Renaissance Philanthropy — in my opinion, the most exciting philanthropic venture in the US — is getting a one-year check-in. Kumar Garg first appeared on the show right before I went on paternity leave, and now we’re back for round two. Before founding Renaissance Philanthropy, Kumar worked in the Obama Office of Science and Technology Policy and spent time at Schmidt Futures.

We discuss…

• How Renaissance catalyzed over $200 million in philanthropic funding in its first year,

• The goals of the organization and how it has responded to Trump’s S&T funding cuts,

• What sets Renaissance apart from traditional philanthropic organizations, and lessons for China-focused research foundations,

• AI applications in education, from tutoring to dyslexia screening,

• Donor psychology, “portfolio regret,” and how to build trust within a philanthropic network.

Listen on Spotify, iTunes, or your favorite podcast app.

The Hedge Fund Model of Giving

Kumar Garg: I like that this is becoming an annual tradition.

Jordan Schneider: Yeah, we've got to set goals this year, and we can hold you to them in 2026.

Can you start off with the 101 of Renaissance philanthropy, and explain how the thesis has played out over the past year?

Kumar Garg: I’m grading myself here, so this is a biased view, but it’s been a very strong year. When we were launching the organization, we were trying to do something different.

Most philanthropic organizations exist in a single model — they work for a single donor. That donor has resources, whether they sit in a foundation, in their DAF, or as personal wealth. The organization works for them, asking how much money they want to give and on what topics, then runs their philanthropic giving.

There’s another class of organizations that are basically the people spending the money — researchers running labs and doing high-quality research. The philanthropic system has mostly operated with givers and takers — folks operating these organizations and folks doing high-quality work.

The idea behind Renaissance Philanthropy was to sit in the middle and style ourselves more like an investment fund — more like what happens in the world of finance. The folks who are the holders of capital, who have the money, mostly don’t spend their time trying to directly deploy that money.

If you work as an LP for a family office, you might have a team of 10, 20, or 30 people, and you’ve got billions of dollars to deploy. What do you do? You go out there and find intermediaries — private equity funds, hedge funds, venture capital funds, or other experts in particular sectors and areas. You give them the money, and they deploy it on your behalf to help you earn a return.

Philanthropy has mostly operated differently. It’s odd, but it’s historically contingent. The investment world moved toward specialization from the ’70s onward, while philanthropy went in the direction of direct giving. You have really large philanthropic organizations, often well-staffed by experts, that do the giving.

The challenge is that there’s a subset of donors who want to build large organizations, and there’s a large set of donors who don’t. The ones who don’t have been sitting on the sidelines. What ends up happening is maybe when they retire, they build an organization, or when they die, they bequest it to a nonprofit or university. That leaves a lot of value on the table.

The idea of Renaissance was, on various science and tech topics, can we do what an investment fund does? We write down a thesis for three years, five years — we want to achieve this goal. We recruit a field leader to run that fund, then treat the donors almost like LPs in a philanthropic fund. We’re not giving them a return back, but they’re putting money to work against that strategy.

A year ago, when I told this story to people — “I’m going to create an organization that does this” — the operative advice was, “Good luck.” You’re going to cover the waterfront across AI, climate, and economic social mobility. You’re going to take on this massive fundraising goal. That seems like a very hard way to operate. You have no natural advantages — you’re not spending one person’s resources. You have to raise the money and deploy it. It seems doubly hard.

What I was interested in was growing the pie — can we use this model to bring new donors in?

A year in, the early grade is strong. We’ve been able to stand up multiple philanthropic funds. We have a fund using AI to accelerate the pace of math research. We have another fund using AI to deliver public benefits better. We recently launched work on climate emergencies — can we solve for runaway climate risks and increase the technology readiness level of various climate technologies?

We have different funds in various areas. Each has this basic structure — they have a thesis they’re driving against, a field leader running it, and we’re recruiting donor money against that strategy.

What I’m hoping for is that this starts to become — not the only way philanthropic giving happens — but a much more credible path. This allows more donors to be active without necessarily having to take on all the operational load themselves.

Jordan Schneider: You’ve launched this in a particularly precarious time for the future of science and research in America. We’ll get to your takes on the policies in a second. But I’m curious from a donor appetite perspective — what has all the tumult in universities and government funding done for those billionaires sitting on the sidelines, giving just 1.2% of their assets annually to philanthropy?

Kumar Garg: That’s a great question. I don’t have one system-level answer — it’s a frequent question I get about how donors are interacting with the environment. They’re interacting the way most people are — there’s an incredible amount of chaos and news every day, leaving many frozen in place.

It’s relative. Government has pulled back on research funding in the short term, causing significant churn. Industry is also holding back as companies figure out what’s happening with tariffs and everything else. Philanthropy, comparatively, is cross-pressured but hasn’t engaged in the same pullback.

There are donors we interact with who are certainly reformulating their strategies. There are others who, as I mentioned, are interested in compelling ideas and looking for those just like anything else. I haven’t seen an overall pullback — just more of a sense of “Is this idea good in itself, even if government didn’t help at all?"

Jordan Schneider: Can you put in order of magnitude the hope of the new model you’re trying to manifest against, I don’t know, NIH budgets being cut by a third?

Kumar Garg: There’s no world in which philanthropy fills the gap. If you step back and ask how the US built its lead — well, the US spends on the order of $200 billion a year on R&D. Once you include basic and applied research across DoD and civilian agencies, that’s an order of magnitude more than philanthropy spends on research.

The place where these new models will get traction is that how you organize scientific organizations has suddenly become much more of a jump ball. It used to be that the academic bundle — being at a top university — had everything stacked on top of itself. You could get really good talent that way — graduate and post-graduate talent, great students. You build your lab there, do cutting-edge work. Usually, the university gives you flexibility to do many things on top of it. If you’re an academic doing well at cutting-edge research, you could do that within the four walls of a university.

Some researchers have left universities and built what are basically academic research labs outside the university. You’ve got the work that Patrick Collison is supporting around the ARC Institute, the Flatiron Institute that Simons supports. You’ve seen the FROs that Convergent Research proposes. For a long time, that’s been a very alternative path — rare to do, often requiring you to figure out what happens to your university affiliation and how it changes your career path.

If you’re a researcher who’s ambitious and wants to do big projects, whether you’re doing them within the four corners of the university or in your own nonprofit research lab and partnering with universities becomes more of an open question — especially in a world where university funding might fluctuate based on political developments.

I don’t know how that will play out over time, but we’re three months into a deeper shift in how institutional financing will happen. That could have big implications generally. On net, if the federal government doesn’t play its important role in funding research, it’s all a net negative. If federal funding returns to a healthy level, researchers will still take this as a wake-up call to think about structuring their research organizations to be more resilient against systemic shocks.

This episode is brought to you by ElevenLabs. I’ve been on the hunt for years for the perfect reader app that puts AI audio at the center of its design. Over the past few months, the ElevenReader app has earned a spot on my iPhone's home screen and now gets about 30 minutes of use every day. I plow through articles using Eleven Reader’s beautiful voices and love having Richard Feynman read me AI news stories — as well as, you know, Matilda every once in a while, too.

I’m also a power user of its bookmark feature, which the ElevenReader team added after I requested it on Twitter. ChinaTalk’s newsletter content even comes preloaded in the feed.

Check out the ElevenReader app if you’re looking for the best mobile reader on the market. Oh, and by the way — if you ever need to transcribe anything, ElevenLabs’ Scribe model has transformed our workflow for getting transcripts out to you on the newsletter. It’s crossed the threshold from “95% good” to “99.5% amazing,” saving our production team hours every week. Check it out the next time you need something transcribed.

Jordan Schneider: I observed the EAs being very excited about how many lives they save based on the bed nets they bought. Then you net that out against USAID no longer existing and all the human suffering that’s going to come out of that. The correct calculation may have been to spend all your money lobbying Congress to get people to focus on this.

I think both of us are pretty aligned — we’ve done other shows on immigration policy, university funding, and what’s happening to NSF and NIH budgets going forward. But why do Renaissance philanthropy when Kumar could be spending 100% of his time in D.C. banging on doors and trying to make it 5% more likely that we get an extra $10 billion a year for this stuff?

Kumar Garg: That’s a great question. Being policy-adjacent is generally very high ROI. No matter how you run the numbers, policy advocacy — especially on science and tech topics — punches above its weight, regardless of what you’re doing. It’s probably why I spent time in government as a policy staffer. It’s partly why, no matter what I’m doing, I’m constantly interacting with policymakers and making the case. It’s also why, when funders ask to what extent advocating on behalf of the research community should be part of their work, I’m strongly supportive.

The reason we structured Renaissance this way is that I wanted to specifically think about growing the pie of philanthropic funding because I thought no one was doing it. There are organizations working on policy advocacy. Very few organizations were trying to bring new donors into the mix.

We would be failing as an organization if we weren’t constantly thinking about how our work could impact shaping the debate on the future of R&D funding. We try to be in conversations with both Congress and the administration, as well as policymakers up and down the ladder, to say, “Here’s why this work matters, here’s why the future matters.”

Part of the new models we’re funding — whether it’s things like FROs or AI accelerating science — is to make the case for why investment should happen. Many of the ideas I’ve funded over the years, you can see echoes of in the new Heinrich legislation around accelerating science through AI, where they’re talking about ensuring these AI investments can actually accelerate the pace of science using new models.

Philanthropy, when done well, opens the aperture for what funding could do. Hopefully we’re playing that role. One area I’d like the conversation to reach is moving beyond this dialectic between “science is important” and “science needs to be dismantled because it made mistakes.” I’d like us to reach a place where we recognize there are important things we can do to help reform how we do science. We should bring more discipline to trying out new ideas, bringing in new funding methods and new voices, and reflecting on past mistakes — while also remembering that the investment agenda around science is critical for its utility. Hopefully we can be part of that dialogue.

In some ways, you’re pushing on something I think about all the time — I am a policymaker at heart. The deep utility of that shouldn’t be forgotten in my story.

Jordan Schneider: All right, I think the answer I would give to you is this is the federalist model of policymaking almost — as you said, the inventiveness that you guys can come up with from a form factor and discipline perspective when it comes to doing science and technology research is the type that’s weird enough that it’s not happening in government. But also, two, three, four years down the line, once you guys have some really awesome case studies, these are the sorts of things that can then get 10x or 100x in our gorgeous NSF circa 2027 that has been remade to fully align with the Kumar vision of how change gets made.

With that stance of optimism, let’s talk a little more in detail about some of the projects you guys have stood up. Take us on a little tour, Kumar. Where do you want to start?

Kumar Garg: I’ll go through a couple of the funds and projects we’ve launched. Just to give people a taste of how the model works, let’s start with our work in AI. Our operating theory in AI is that we’re living through a period of huge capability overhang. The idea is that the core technology is rapidly developing, but the number of people, projects, and overall work that actually applies these tools toward actual hard problems in society is really small.

I’ll give an example. We have an AI and education fund specifically focused on how AI can accelerate the pace of learning outcomes. If you follow social media and others, there are many people who write and talk about AI in education. It would give you the sense that a lot of people are working on AI in education. But if you actually dig into the space, the number of actual technical experts who have knowledge of both how education works and how AI works is still shockingly small.

We run something called the Learning Engineering Tools competition — an annual competition that invites tool developers to present cutting-edge ideas that use AI to actually advance learning outcomes. We’ve been running this competition for a couple years. I started it even before Renaissance and then brought it into Renaissance. That competition is the only large-scale ed tech competition in the world. It still blows my mind. No one is out there in a systematic way asking for sets of ideas from people who want to build AI for education.

We have another part of our AI education portfolio that specifically thinks about moonshots — what’s a really hard problem in education that AI could solve? We picked middle school math. It’s really important for advancing to future degrees, and students really struggle with it. We said, can you actually emulate the results of high-dosage tutoring, which the number of studies that J-PAL and others have done show can really double the rate of learning for students in math? Can you do it under $1,000 per kid — bring it under what would make it such that you could offer it to every kid?

We have that running as a program. We have seven teams in the program. We have two teams that are actually on track to potentially accomplish this goal.

Jordan Schneider: Which is wild, right?

Kumar Garg: When those teams are working on it and we ask them who they’re collecting lessons from, there’s not a big field they can go out to. When they go out and interview the AI labs — the ones that get written about every day — those AI labs talk about education, but they don’t have in-house education teams that can actually help these teams.

The biggest piece I would always say to people is that at the coal face, there’s tons of room to do work because when you actually start to work on it, you realize that the number of people who are actually working on it is shockingly small.

We’re now starting to explore our next moonshot area — should there be something that basically looks like the intersection of AI and early learning? Can we actually build a universal screener to best guess if a child is off track when it comes to early language development by having them speak into a device? There’s a bunch of interesting work happening in this area, but we don’t actually have a way to diagnose early learning challenges like dyslexia just by having a student speak into a device. It could vastly increase our ability to help them get to a speech pathologist, get back on track, and be reading by third grade, which is critical to future reading and learning.

That’s just one track — AI and education. That’s just one compelling thesis.

Jordan Schneider: Obviously AI is going to matter for education. Hard to find people to argue about that. Talk a little bit about finding the donors and finding the teams. What was the work that you guys had to do to make and launch this work?

Kumar Garg: What’s been interesting is it has been hard work for us to build out the team because the number of technical experts who actually know both things — AI and education — is small. We have slowly built out a team of ML experts who have educational backgrounds, basically. We call it a hub model. We basically have created an engineering hub and we recruit technical experts into it that specifically have this technical background.

I have somebody on my team, Ralph Abboud. He has a machine learning PhD, and he did his thesis on graph theory. He’s not an education expert, but we brought him onto the team. He has been working with a lot of these educational teams that we brought in. What’s interesting is that his ideas on what kind of language models they should be building are really good. It took him some time to level up on the education side, but now he is one of their highest value contributors, even though he sits on our team and he’s contributing there.

There is this transition where you can build up talent that sits across these two areas. But in AI and education we had to mostly build it. It was hard to immediately find directly. Now we have a constellation of these AI and education experts, some of which sit on our direct staff, some of which sit inside these teams that we’re betting on. It’s been great. Now we have a field team that can really go after more problems.

On the donor side, we’ve really lucked out. We found that our core donor for a lot of this work has been the Walton Family Foundation. They have a long history in funding education. What’s been interesting is that they’ve been interested in investing more in what they call their innovation portfolio, but didn’t know how to necessarily bridge that technical divide — if we’re going to do more in this area, who are going to be the technical experts who will actually do it? That had actually kept them more experimental. But their partnership with us has meant that they have become way more ambitious on how much investment they want to make on this technical AI and education lens.

That’s our core thesis — can we be the permission structure for donors to go much bigger on innovation? We’ve seen that in other areas. Slowly their support is causing other donors to come in as well. That’s basically whether you’ve been a long-standing donor but not active on science and tech topics, or you’re an early donor altogether.

Jordan Schneider: What’s the RenPhil management fee?

Kumar Garg: That’s a good question. We build our cost recovery into each fund. Usually the way that works is if we’re operating multiple funds, each fund has money going out the door for actual deployment grants, but then we’re building in our cost for the actual staff operating the funds, whatever services and technical support services we’re providing, the work we’re doing to partner with various funders, as well as our overall studio support. It varies fund to fund, but donors have found it — compared to having to try to do this themselves — much more actionable. For us, we want to build a thriving organization. We don’t want to cut corners. We want to build an organization that can both operate those funds and also be looking for the next ones.

Jordan Schneider: Does anyone complain about that?

Kumar Garg: The way it comes up is there’s a type of donor who actually has the answer in their mind. They’re thinking, “I think this needs to happen.” Really, what they’re looking for is an operating partner to just do that — “I want a conference, I want a workshop, I want to fund these three organizations.” Our model is we’re the product. You’re actually hiring us to go build out the strategy, recruit the team, deploy. If you actually have the answer in your head, we often tell them we’re way too fussy for that model. You should just — there are much simpler ways you can operate. That’s where the delta comes in. If you already have the answer in your head and you’re just looking for a partner to execute for you, we’re probably not the right fit.

Jordan Schneider: Yeah, you said this on another show. You were like, “We are for donors to take off the cognitive load.” The idea of being: yes, if I have $10 billion, maybe I’ll allocate $1 billion investing in stuff I know and think I have some subject matter expertise in, but I still have to put the other $9 billion somewhere — probably not cash. Yes, I am comfortable paying a hedge fund or financial advisor a management fee to do that.

Kumar Garg: A big part of it is opt-in. People don’t know what journey they’re on, but what they worry about is: am I going to feel stuck? A lot of folks end up not getting active philanthropically because the decision feels weighted by getting stuck. “Okay, if I hire somebody, and then six months from now, I decide maybe I want to change direction. Now I’m going to have to let someone go.” People hate that.

Or, “I met a researcher. I liked their research. I gave them one grant. But now they’ve reached out and said, ‘There’s so much happening in the world — I’ve lost funding from the government. Can you double the grant?’ I was just giving them a grant because I met them and I thought they were great. But now they’ve sent me a note that they might have to let go of postdoc students. Now I’m in this uncomfortable situation. If I say no, I feel I’m hurting them. If I say yes…”

People have all these experiences where they feel uncomfortable with the relationship they have around their resourcing. Rather than causing them to work through it, they actually hold back. One of the things we say to them is that our model is one where we’re the ones making the decisions. We’re going out there, finding researchers, finding projects, developing strategies. You can be as involved as you want. You want to be meeting the researchers? That’s great. You want to be learning from the strategy so you can do direct giving down the road? That’s great. But if you also take six months off and decide, “That was great, I learned for a few months, now I’m off doing something else,” — nothing will stop. We’re a fully operational organization that will execute on everything we said we were going to do, whether you’re involved or not.

It just takes the pressure off. You can opt in if you want to learn and be involved, but you can also choose not to. That actually frees them up to want to learn without the “Am I about to get stuck?” That sounds very psychological, but people forget how hard it is to get going on things — “I’m going to start to work out more. I’m going to start to do this.” Starting is hard. We want to make starting easy by saying you can provide a lot of value into the system without necessarily having to own all of that execution.

Subscribe now

Jordan Schneider: There are a lot of pieces of people’s jobs that seem like more and more AI can chip away at or enable or launch or whatever. It’s interesting because some of the things that you guys are doing — you have these seven playbooks, ways you can tackle problems — I would love to upload seven of those to ChatGPT and say, “Here’s my problem in the world,” and the AI can help me pick through which one. But talking someone who’s really rich, who’s feeling uncomfortable about giving money, into starting to donate philanthropically in a serious way for the first time seems like one of the more human things where there’s really going to need to be the friendly Kumar Garg, who now has a nice microphone he can do Zoom calls with to — what did Derek Thompson say? — whisper the dulcet tones of comfort and competence in their ear in order to get them on this path. I don’t know, it just seems like a very human thing that you’re engaging in on the donor engagement side. I’m curious for any reflections you have on that.

Kumar Garg: We are very curious about how much of our own internal processes we can automate. Why not? We sit next to AI, we should be thinking, we should be dogfooding. The place where we’ve seen it already provide some value is just what you would consider baseline automations. There’s a lot of grantee reporting that you should be able to do automations on. We’re definitely interested in: hey, we have a hunch around a thesis in this area. Can you do a research report and tell me what’s the relevant stuff to know? Scoping. We’ve even used it for, “Hey, we might do an RFP on this topic. Who are some researchers who should apply?” Sometimes we found some interesting suggestions for researchers that we should affirmatively reach out to.

I will say that we’re still far away from it actually helping on anything that we would consider high stakes. As you’re saying, a huge amount of what we’re doing is making something that feels like a trust fall. Hey, this is an important decision, but one where people who take their job very seriously and put their own personal legitimacy behind the work is an important part of it. When we screw up, it’s on us. We stand behind all of the work. People appreciate that these are serious people who stand behind the work they’re putting before them. They’re not some faceless intermediary. I don’t think maybe that will change, but that’s an important part. Even on the information you should know about various people and stuff, these current AI models are not that great.

The place where I — we have this intuition that there should be parts of being a program leader that you should be able to have an AI assistant for. Right? You take more and more of the task of being a program leader or fund leader and be able to say, “Okay, I want to do a workshop on this topic. Generate me an agenda for how you would run the day.” It generates based on — it takes a bunch of your past workshop flows and generates a sample workshop design. How much of that can we create so that we really could get to a point where a program leader or fund leader is basically able to operate without that much additional support? Obviously we need to create some cross-cutting support that I’m interested in. But the chance that we’re going to get to an AI advisor — we’ll have to wait.

Jordan Schneider: The trust fall works in multiple directions. You need researchers to give up their PhD programs or leave their current positions to spend half their time with you, while simultaneously needing donors to provide funding. Having a recognizable face with a proven track record and skin in the game on the other side of that equation is something that won’t disappear anytime soon.

Kumar Garg: One thing we debate internally is that much of my workflow relies on tacit knowledge. When I’m talking to somebody about their work, twenty minutes into the conversation, I’ll say, “Tell me more about that. Why is the field stuck on this point?” They start describing it, and I realize that if there were a canonical dataset with specific dimensionality, it might solve the problem. When I ask why that doesn’t exist, they explain it’s locked up somewhere.

Part of me constantly strives to figure out how we can make this process more explicit. When we recruit somebody new to the team, they ask if they can sit in on my calls and watch me work through problems with researchers. There’s something that feels wrong about just saying, “You develop this intuitive feeling for opportunities — just pull on that thread.” The more we can transition from tacit to explicit knowledge, the better. Right now, we operate on an apprenticeship model where people learn by doing and being embedded in these structures, but I don’t think that has to be the endpoint.

Jordan Schneider: Much of what you do involves human matching — putting people in touch with each other. While you could potentially feed all your past calls into an AI system, there’s an emotional and personality matching component that you’re handling. That remains very much a human process that current models aren’t quite ready for yet.

Kumar Garg: The matching capability changes over time, but I think what people really value when I connect them is that I took time out of my day to think the two of them should know each other. That’s the actual signaling value — that my time is precious.

Jordan Schneider: Slight tangent, but if people want to establish trust and rapport, the first thing they should do is spend $150 on a microphone for their Zoom calls. That’s my recommendation for everyone. When I do my calls, I sound the same as I do on my podcast, and people respond positively. You feel like an embodied person rather than a compressed, distant voice through AirPods. It’s advice for anyone who wants to make connections and raise money from billionaires on Zoom.

Kumar Garg: I’ll echo that point, though I haven’t practiced it myself. There’s an old political adage about microphone technology. If you look back at politicians historically, there was a time when microphones couldn’t pick up subtle intonations well — speakers were just projecting loud sound. Once microphones could capture subtle intonations, politicians who excelled at that style of speaking began to dominate.

People point to President Clinton as an example — he was exceptionally skilled at subtle microphone use. I remember reading a paper arguing that this was possible because the technology had improved to support that communication style. Politicians offer good lessons here because ultimately, communication is central to building trust with the electorate. [Here are papers that explore this]

Jordan Schneider: Absolutely. If you listen to old clips of Warren Harding or Teddy Roosevelt speaking, they’re basically screaming into microphones — which was necessary at the time. Teddy Roosevelt was exceptionally good at that style of projection. You needed to be very loud to stand on a soapbox and reach people twenty rows back. Now we have the dulcet tones that modern microphones enable.

Here’s another fun fact, Kumar — the microphone I’m using has been manufactured for sixty years. It’s remarkable that microphone technology for voice pickup has essentially reached its peak — we’ve basically maxed out the capability.

Kumar Garg: I should try to find that paper I mentioned. I wonder if it’s about mobile situations — being in some random union hall where you need to set up a handheld mic in front of a politician. Perhaps that’s why microphone technology improvements became so important. That’s an interesting angle.

Jordan Schneider: I’m curious about that. Alright, shifting topics — Yascha Mounk recently wrote on Substack about attending gatherings, conferences, and dinners where leaders of America’s biggest foundations have been strategizing how to defend democracy. Few were as openly devoted to extreme forms of identitarian ideology as they might have been a few years ago, but the reigning worldview at the top of the philanthropic world assumes little has changed since summer 2020.

The general consensus holds that voters turned to Trump because American democracy failed to deliver for the “historically marginalized,” and the solution supposedly revolves around “mobilizing underrepresented communities.” The most urgent imperative is to “fight for equity” and “listen to the global majority.” I find this perspective fascinating. Kumar, as someone who’s a new entrant to this world, how do you interpret this?

Kumar Garg: Several different dynamics are happening simultaneously. Some philanthropic responses resemble dinner table conversations — people sharing hot takes about why the election unfolded as it did and offering their views on America or the American people. Much of this sounds as random as hosting a dinner party where guests share their political opinions.

There’s also a genuine state of confusion about what’s happening. The first hundred days of the Trump administration have been exceptionally active across a range of unexpected areas. Many people expected it to feel similar to the previous Trump presidency, so they examined their portfolio of issues and anticipated certain outcomes — but that’s not what materialized.

Regarding how much people are actually rethinking their approaches, that’s a valid question. The most immediate reconsideration I’m seeing centers on identifying what we’re missing. This is particularly evident in the science community, which is confronting devastating across-the-board cuts. Researchers are losing funding, university funding is being paused, and graduate students working on topics relevant to competitiveness are having visas revoked.

The community is asking, “We don’t remember this being a major campaign debate topic, so how exactly did we become a political football?” There’s extensive questioning about what we’re missing — whether there was a conversation we weren’t invited to where we were being discussed, and what we’re failing to understand.

While donors with certain political orientations likely won’t change their fundamental positions, the confusion centers less on American domestic politics and more on why certain issues became contentious. Foreign aid is a good example. The extent to which US foreign aid posture and system effectiveness became campaign issues wasn’t apparent during the election cycle. People are asking whether we missed a major debate that suggested the United States should dismantle its leadership on these topics overnight. What policy debate did we miss? That’s where much of the confusion originates — donors being puzzled about the sources of these developments.

Subscribe now

Jordan Schneider: There’s an interesting dichotomy between foundations with living leadership and those with deceased benefactors. Gates recently indicated he would spend down his wealth faster than previously planned, presumably responding to recent events. When foundations have active leadership that engages with current events, they can be more responsive.

However, when you have flagship philanthropists who died seventy-five years ago and organizations that have built programs around worldviews that are no longer relevant or don’t meet current demands, pivoting becomes much more difficult. You encounter institutional blockers, boards, and established structures, whereas a living person with decision-making authority can simply redirect resources.

Kumar Garg: That’s definitely part of it. The piece I’d add, which connects to our Renaissance model, is that people underestimate how much philanthropic organizations become tied to their existing programs. This isn’t necessarily negative, but consider the process: you spend two and a half years scoping a program, conducting field research on topics, then executing a national search for program leadership. You recruit and convince someone to relocate for the position, provide coaching for test grants, and they’re six to twelve months into the grant cycle.

If you then decide the world has changed and want to cut the program — after issuing press releases announcing this as a major new strategic direction — it appears chaotic. People develop what I call “portfolio regret.” If they could start fresh today, they’d create different programs than what currently exists.

One argument we make to donors is structuring themselves more like limited partners, deploying money into funds where all capital remains fresh and available. You avoid the incumbency problem where team members question every pivot attempt because they have specific responsibilities you hired them to fulfill.

Flexibility requires both mindset and structure. Donors sometimes create substantial built-in costs and barriers to pivoting when they could maintain lightweight, flexible operations if they chose to do so.

Jordan Schneider: That’s fascinating because you’d think giving money away would be straightforward — you should be able to distribute funds however you want. But the emotional sunk cost around philanthropy wasn’t something I had necessarily considered.

Kumar Garg: This creates a situation where people spend enormous amounts of time operating like a duck swimming on water — their feet are moving rapidly beneath the surface while they try to keep the strategy looking consistent above water, all while changing the actual content underneath to pivot to current circumstances. This leads to significant conceptual confusion because you claim to have always had this program, but underneath, the program is completely different because the situation changed.

Part of why I favor the philanthropic fund model is transparency — what’s on the cereal box is what you get. It’s a three-year fund with specific objectives that will begin and end. Maybe it’s not perfectly timely, and that’s fine, but your new program can be timely. The alternative — constantly maintaining broad programs that you’re perpetually reworking underneath — makes evaluation nearly impossible. If I ask whether a program has been successful, people respond that the program has been changing constantly. This makes it extremely difficult to evaluate it as a focused initiative that ran for a specific period with defined goals. Did we achieve those goals? People simply don’t engage in these basic evaluations.

For example, I was speaking with a donor and pointed out that in the investment world, people prominently display their successes — putting “first check into [major company]” in their Twitter bios to demonstrate their betting acumen. I asked, “Who are the ten best program officers in America? Who on the philanthropic side has been the most effective check writer?” They responded, “How would we even know that?” Even with qualitative measures, wouldn’t you want to identify the best check writers?

A fund model, even for philanthropic goals, enables more honest assessment. You can say that fund paid out successfully, that one was moderately effective, and another one failed completely. The person who led that fund can then take that track record to their next position as a legitimate career advancement. We deny people this opportunity when we maintain the fiction that we’ve always had these programs run by different people with slightly different strategies. This obscures rather than clarifies outcomes. We could simply be honest: we executed that fund, now we’re doing something new with a clear beginning and end.

Improving China Research + Why Bother?

Jordan Schneider: It’s remarkable. I’ve spent considerable time on foundation websites researching whether organizations like the Ford Foundation might fund ChinaTalk. They have mission statements about “democratizing equity,” which is admirable — I agree we should advance democracy and create more opportunities for people. However, the problem arises when you’re only a passive recipient of pitches. You’re essentially letting grantees define what success means, and the counterfactual becomes very difficult to assess. The organizations you’re funding would probably exist whether you give them $100,000, $500,000, or nothing at all.

The alternative approach would be starting with specific objectives — “We are trying to achieve X by Y timeline,” then working backward from that goal orientation to identify the people and organizations who can take your money and provide the highest probability of achieving that outcome. This approach is far more strategic, and my frustration isn’t about the politics of how they set their goals — it’s that they need to engage more seriously in the process.

Kumar Garg: Let’s work through this together. Imagine you and I were designing a fund model versus a program model for increasing collective intelligence on US-China relations.

There’s a vague way we could structure this — “This is the US-China program. It will have three tracks — funding scholars studying China, engaging policymakers about those insights, and warehousing data and research publications on these topics.” This resembles how most programs operate: they establish a broad framework with several tracks, then people apply under those categories. But if I asked what constitutes winning or how we’d know this program succeeded, you’d probably say, “Well, people applied and we distributed grants."

You could execute the same concept with much sharper focus by asking, “What would success look like in three years?"

Jordan Schneider: Exactly. I want ten books written that are so thoughtful and essential to the future of US-China relations and American policy that Ezra Klein would be compelled to feature these authors as guests because the thinking these grants produced is indispensable. Then we work backward from that goal to determine the budget.

We’d estimate the costs — ten books, assuming a one-in-five success rate for people with strong proposals to execute effectively, calculate the pipeline requirements, and arrive at a number that gives us a 75% confidence interval for producing those ten books by 2029.

Kumar Garg: Exactly. That approach feels like a sound tax strategy built on what resembles a very tight OKR, which you may completely fail to achieve. You might be wrong, but it’s precise. Then you build your strategy around that goal.

You find someone to execute it — let’s call it the Jordan Fund. If you succeed, people ask, “Jordan, how did you pull that off? You wrote down this ambitious goal, built a strategy around it, and executed successfully. You’re clearly skilled at this.” When you pursue your next initiative, you can say, “I ran this fund called the Jordan Fund. We set this audacious goal to produce ten bestsellers on US-China relations, and we achieved it."

This feels much more tangible as an actionable strategy — something that field leaders can point to as real-world impact, even if it fails. Let’s say you only achieve partial success — you still have concrete lessons. Compare that to “the US-China program makes some awards and does some things.” How do I assess whether that’s working?

Jordan Schneider: It’s remarkable because the market is so powerful — you can’t get away with this approach when running and scaling a business, especially when taking other people’s capital and trying to generate positive returns for them.

What’s curious, Kumar, is that very capitalist people become surprisingly touchy-feely when it comes to philanthropy. There’s an emotional layer where they think, “This is giving, so we shouldn’t apply business mindsets and OKRs.” It feels somehow dirty to them.

Kumar Garg: Here’s what’s important — we need to distinguish between current donors and potential donors. People focus extensively on today’s active donors, but if you examine the statistics on potential giving, current donors might represent only 1-2% of the actual addressable universe.

The question becomes, would we attract an entirely new class of donors if we brought this level of rigor, precision, and targeted approach to philanthropy? This would feel much more familiar to their professional experience.

Why don’t existing donors demand this rigor? I believe there’s significant pent-up interest in this approach. People oscillate between thinking, “Since this isn’t about making money, we’ll substitute with having a really complicated theory of action — that’s where we’ll apply our intellectual capacity.” I often say that just because something has numerous boxes and slides doesn’t substitute for having a clear attack vector.

Jordan Schneider: Exactly — rigorous thinking. The median nonprofit worker is about five times more likely to be socialist than the average person, so perhaps people more attracted to touchy-feely logic are simply concentrated in current organizations.

Kumar Garg: Sometimes the nonprofit and philanthropy sectors spend too much time engaging in collective mission statements, as if shared purpose alone is sufficient. But we actually have distinct roles to play, including making high-quality decisions about where to deploy finite resources. Because money is limited, you must make decisions strategically and place informed bets.

This may feel reductive, but it’s actually the responsibility inherent in this work. You must be a responsible steward because high-quality decisions produce more good. People sometimes struggle with that reality.

Jordan Schneider: Without high-quality decisions, you end up with USAID getting canceled. That’s our current reality. Organizations that weren’t evidence-based and couldn’t effectively justify their impact had some good projects and some poor ones. They faced criticism from small but vocal movements — organizations like Unlock Aid, whose founder we featured a few years ago and will have on again — arguing for more rigor because there was substantial waste and inefficiency.

If you let these issues fester too long, consequences follow. I don’t want to say universities, the NIH, or the NSF “had it coming,” but one of the best defenses you can have is a tight, well-justified organization that can stand up for itself.

Kumar Garg: I don’t want to engage in victim blaming, and I don’t want to excuse what I consider sometimes bad-faith behavior. However, your point about systemic advantages is valid — caring deeply about systemic impact and bringing that rigor and constant evaluation is useful for the work itself, but also valuable when those political fights emerge. You can say, “Look, we’re building something substantial."

In some of these cases, who knows what impact rigorous evaluation might have had. We’re living through unusual times, but I believe we’re gaining traction because there’s significant pent-up demand for this approach.

Jordan Schneider: Good. That makes me feel somewhat better, I suppose.

Kumar Garg: What should donors know about China? That’s my question for you.

Jordan Schneider: The original impetus for ChinaTalk was thinking about long-term national strategic competition and competitiveness from an industrial systems and technology perspective — identifying things people could do to nudge outcomes in liberal democracies’ favor. During the Biden era, I observed errors that legislation and executive action could fix with modest improvements — 5% here, 10% there. A sophisticated understanding of what’s happening in China could meaningfully help squeeze that extra 10% out of various decisions.

However, the policy changes we’ve witnessed over recent months regarding long-term strategic competition — how the US relates to allies, approaches to global nuclearization, science and technology funding, and immigration — are much more fundamental. Getting to a better place doesn’t require understanding what made BYD successful, how SMIC is developing its chips, or even China’s new AI policy. These are much more basic issues.

The thesis I operated under during the Biden era was that deeper, more considered understanding of China would lead to smarter policies. That’s now become a sideshow compared to more fundamental questions. If we accept the base case that science is important and immigrants are crucial for better science, then we should pursue those priorities directly. I would choose that approach ten times out of ten.

Returning to the order-of-magnitude questions I asked at the beginning, I would choose a NATO that functions as a genuine alliance ten times out of ten over determining the right tariff level for Chinese electric vehicles or batteries. That’s why I lean toward “be nice to allies” bumper stickers and NSF funding advocacy rather than tightly nuanced “we need to better understand China” approaches when considering ChinaTalk’s decadal competition mission.

Kumar Garg: One thing I’ve been considering, though I don’t have the answer yet, is what new institutions we need. Much of what I care about regarding how science operates in this country has been overturned. The idea that we’ll navigate this period with identical institutions seems unlikely — whether it’s who makes the case for science, who serves as science messengers, or how we conduct science itself.

This raises questions not just about policies, but about institutions themselves. Obviously Renaissance is part of that response, but I have broader concerns. We’ll probably need new institutions because the players on the field will have to change. Systemic change of this magnitude requires that everything else engage in significant adaptive change for us to succeed. That seems unlikely with current structures.

That’s a major meta-question I’ve been asking the team, “We won’t be able to handle everything directly, but what institutions would restore us to better footing? Do we have them? Do we need to create them?”

Jordan Schneider: I’m starting to focus my energy differently because I’m uncertain whether additional ChinaTalk podcasts about the importance of allies will accomplish much. However, one constant you can expect over the next four years is AI development and rapid technological change.

Regardless of controversial Trump policies, the Defense Department will persist, and America will still need to protect itself. America engages in conflicts approximately every three years, so that pattern will likely continue. Perhaps this is just me entering a kind of intellectual monk mode after dynastic change, but I’ve been reading extensive military history and examining periods of rapid technological change — specifically, what it means to deploy these tools more effectively than adversaries.

This doesn’t directly answer your question, but I’m pursuing intellectual journeys rather than policy ones.

Kumar Garg: Here’s what I’d say, which connects to the role you and ChinaTalk are playing. One thing I mentioned to Jordan before we started is that frequently when people reach out to me, they reference hearing a great ChinaTalk episode. You may not have set out to do this, but you’re playing a valuable role in shaping how other people — especially technical professionals — think about problems worth solving and their mental frameworks for our current age.

People are seeking understanding and meaning in this moment. The question isn’t about marginal additional podcasts, but whether you’re providing people with new vectors for their lives and careers. When we first met, you described yourself as a nerd passionate about these topics but uncertain where to channel that energy. You’ve created something quite distinctive.

We might be living in an era of unusually shaped careers, and we need to give people more space for that kind of professional evolution.

Subscribe now

Jordan Schneider: That’s fair. Some of this material feels obvious to me but may not be obvious to people who don’t live and breathe these topics daily. It’s strange because I don’t feel like I’m part of some resistance movement — I’m just a guy with opinions on various issues. Some days I wake up feeling helpless, others I feel genuinely empowered. This isn’t a direct response to your question, but that’s my reality.

Kumar Garg: You’re reasoning in public. You’re expanding people’s understanding of how to think through these complex issues and broadening their sense of who engages in this work. What I consistently observe is that people operate in highly siloed environments. They’ll mention their sources, and I immediately know exactly who they’re reading or listening to.

If you can expand their perspective while providing actionable next steps, that’s valuable. Part of my goal is always emphasizing that there are numerous hard, interesting problems to solve across every arena. No arena is the wrong one. People who dismiss politics are missing a crucial point — we’re all living within political systems whether we acknowledge it or not. Don’t dismiss any arena; simply understand that there are multiple dimensions to engage with. There are challenging problems to solve, and nobody benefits if you just remain a passive observer in the cheap seats.

Jordan Schneider: Perhaps the way I justify all my World War I reading is that no one else is doing this specific work. I’m bringing historical insights to current issues as someone who also reads contemporary news and has the freedom to spend ten hours weekly on intellectual journeys exploring topics I consider relevant to today. That seems like a natural conclusion.

Kumar Garg: We’ll do an update in a year. Hopefully, the republic endures.

Jordan Schneider: Kumar, you need to set ambitious goals for yourself — accomplish so much that I have to bring you back in six months.

Kumar Garg: Absolutely. One of our primary goals is becoming more international. We have a partnership with the British government to develop their R&D ecosystem, and we want to expand that model to additional countries. We’re building an organization that continues to internationalize because science and technology are inherently global.

I’m hoping our fund model will attract new donors who have never engaged in philanthropy before. Beyond discussing our approach, I want the actual work to manifest tangible results in the world.

Jordan Schneider: Contact info@renphil.org if you’re wealthy, have innovative ideas, or simply have a technology-related challenge you need help addressing.

Kumar Garg: I’m also available on LinkedIn — please reach out. We consider ourselves fundamentally a talent network, so I’m always eager to connect with people who have compelling ideas.

Jordan Schneider: We should do a little parent corner. We’ll keep this as part of the annual check-in. We talked about slime last year.

Kumar Garg: We did. Here’s something we discussed before we started recording — I asked you about sleep training, and you mentioned being hesitant to push sleep training advice on others. I’m strongly convinced that sleep training is a gift you give your children. We had twins and committed to sleep training. They’re eleven years old now and remain excellent sleepers today. We attribute this directly to that early sleep training.

For any parent listening who’s on the fence and wants random advice from someone they’re hearing — I can’t offer this to everyone, but I usually tell people I know that I’m always happy to be anyone’s texting buddy, providing extra support to get through those terrible first few days when it feels like you’ve made a horrible mistake. On the other end, you have children who can sleep well, which benefits everyone.

Jordan Schneider: I’m with you. I outsourced this decision to my mother — maybe one of the best decisions of my life.

What’s a cute development we’ve observed recently? I bought a ukulele two years ago, thinking it would be nice to play with my kid. What’s been charming is that my daughter is nine months old now, and there was a period where her manual dexterity only allowed her to grab the strings and pull them. But one day she figured out plucking, and now she’s actually plucking the strings. It’s such a cool activation moment for her — realizing “I make the sound now” instead of just dragging this thing around the room.

Kumar Garg: The period from nine months to eighteen months is truly remarkable. You’re approaching walking and first words, then vocabulary takes off exponentially. It’s incredible developmental progress, so I’m very excited for you.

Jordan Schneider: Perfect. Let’s wrap it up there. Kumar, thank you so much for being part of ChinaTalk.

Kumar Garg: Thank you for everything you’re doing. I’m excited to have participated and look forward to future conversations.

Today we’re running an interview of Huawei founder Ren Zhengfei that made it to the front page of yesterday’s People’s Daily, the premier Party outlet.

What struck out to me in the Q&A was just how hard he hit on the importance of basic research to long term competitiveness. “Without basic research,” Ren says, “there are no breakthroughs, and we’ll never catch up with the U.S.” The People’s Daily giving Ren Zhengfei the airtime to promote this view underscores just how central China’s leadership today sees this work.

We’ve covered the US side of how the Trump administration may be killing our golden goose in this regard with its treatment of the NSF, NIH, university ecosystem and immigration system in recent episodes on America’s R&D Reckoning, our interview with Josh Wolfe of Lux, and a recent op-ed on the war on S&T talent.

This translation also serves as a reminder to subscribe to ’s excellent and meditative daily People’s Daily recaps in his substack Tracking People’s Daily.

"The more open a country is, the more it drives us to progress" — A Dialogue with Ren Zhengfei

By reporters Hu Jian and Chen Jiaxing People's Daily (June 10, 2025, Page 1)

Recently, at Huawei headquarters in Shenzhen, focusing on some hot topics of public concern, a group of People's Daily reporters had an in-person interview with Huawei CEO Ren Zhengfei.

Q: Facing external blockades and pressure, encountering many difficulties, what are your thoughts?

A: I haven't thought about it—thinking about it is useless. Don't think about difficulties—just get on and do it [干就完了] and move forward step by step.

Q: Ascend chips have been "warned on" about usage risks [BIS . What impact does this have on Huawei?

[BIS in mid-May warned firms anywhere in the world that using Huawei Ascends, which were almost certainly fabbed by TSMC under false pretenses, would be considered a violation of export controls. The Information recently reported that “The fallout [from this BIS action] was swift: one Chinese data center firm promptly halted even the small order of Huawei chips it had planned to make, while its legal team scrambled to understand how it could comply with the regulations, according to one person with direct knowledge of the change. Meanwhile, executives at ByteDance and Tencent, both of which have significant business footprints outside China, worry adopting Huawei chips would invite scrutiny from the U.S. government, according to people familiar with their thinking.”]

A: There are many chip companies in China, and many of them are doing quite well—Huawei is just one of them. The US has exaggerated Huawei's achievements; Huawei isn't that formidable yet. We need to work hard to live up to their assessment. Our single chips are still one generation behind the US.

[The incentives are interesting here. On the one hand you have Jensen Huang going around saying Huawei is an enormously powerful competitor who dropped Nvidia’s China market share from 95% to 50%—even though, as the Information very generously put it, “Market share data to confirm Huang’s statement is hard to find.” Jensen, of course, wants the US government to lift export controls not on semiconductor manufacturing equipment but his ability to sell AI chips and racks into China.

The Chinese government and Huawei have every incentive to sandbag Huawei and SMIC’s capacity to compete. They’ve deeply internalized the backlash from Made in China 2025 and want the Trump administration to think that easing export controls on inputs to semiconductors and not investing in the state capacity to enforce the current ones is less of a give than it appears.]

We use mathematics to compensate for physics, non-Moore's to compensate for Moore's [用数学补物理、非摩尔补摩尔], and cluster computing to compensate for single chips. In terms of results, we can still achieve practical applications.

[For more color here, see SemiAnalysis’ coverage of the CloudMatrix 384, Huawei’s answer to Nvidia’s rack solutions]

Q: If there are difficulties, what are the main ones?

A: Difficulties are just difficulties—when hasn’t humanity faced them? Wasn't slash-and-burn agriculture difficult? Wasn't the Stone Age difficult? When humans used stone tools, how could they have imagined high-speed rail? China has opportunities in mid-to-low-end chips, with dozens or hundreds of chip companies all working very hard. Especially in compound semiconductors, the opportunities are even greater. For silicon-based chips, we use mathematics to compensate for physics, non-Moore's to compensate for Moore's, utilizing cluster computing principles to meet our current needs. Software cannot be strangled [卡不住脖子]—it's mathematical graphical symbols, code, and advanced operators and algorithms stacked together, with no barriers blocking it. The real bottlenecks lie in education and talent pipelines. China will eventually have hundreds or thousands of operating systems supporting progress in Chinese industry, agriculture, healthcare, and more.

Q: Huawei is getting a lot of praise and public recognition these days.

A: Praise brings pressure. A bit of scolding keeps us clear‑headed. We make products; once people use them they will criticize—that’s normal. We allow criticism. As long as people tell the truth—even if it’s negative—we welcome it. Praise or blame shouldn’t matter; what matters is whether we can do our job well. If we do, there’s no problem.

Q: From your attitude toward difficulties and criticism, we sense you have a strong inner core—not caring whether it's praise or criticism, but firmly doing your own work well. This should be an important reason why Huawei has come this far.

A: There's still too much praise for us. People should pay more attention to understanding those engaged in basic research. They are highbrow and solitary [曲高和寡—literally ‘difficult songs find few singers’], ordinary people don't understand them, and it takes decades or centuries to see their contributions. Groundless accusations against them are detrimental to the country's long-term development. We need to understand and support those doing theoretical work. We need to understand their broad-mindedness—their great anonymity is our country's hope. Don't elevate one while suppressing another; those engaged in theoretical research are the hope of our nation's future.

"Basic research scientists are lonely—we must have strategic patience and understand them"

Q: How do you view basic research?

A: When our country possesses certain economic strength, we should emphasize theory, especially basic research. Basic research doesn't just take 5-10 years—it generally takes 10, 20 years or longer. Without basic research, you plant no roots. And without roots, even trees with lush leaves fall at the first wind. Buying foreign products is expensive because their prices include their investment in basic research. So whether China engages in basic research or not, we still have to pay—the question is whether we choose to pay our own people to do this basic research.

Q: Regarding basic research, people might not understand the point of it.

A: Scientific breakthroughs—there are inherently few people in the world who understand them, so those who don't understand shouldn't evaluate them. Einstein's discovery that light bends was only proven a hundred years later. In Guizhou, there's an agricultural scientist named Luo Dengyi. In the 1940s, when analyzing the nutritional components of fruits and vegetables, he discovered a wild fruit called cili [刺梨] with very high vitamin content. China was still in the Anti-Japanese War period, education levels were low, and few understood. Later he wrote a paper saying cili was the "King of Vitamin C." After nearly a century, Guizhou has made it into a natural vitamin-rich cili beverage, a luxury item among vitamin drinks at nearly 100 yuan per bottle, highly sought after. The cili industry has become a channel for farmers to escape poverty and become prosperous. Only then did people truly recognize Luo Dengyi, who worked at a broken table during the flames of the anti-Japanese war.

[A deep cut by Ren Zhengfei here but this does look incredibly tasty!]

Q: Many results that look irrelevant at first yet turn out to be hugely useful.

A: Theoretical scientists are lonely; we must have strategic patience and empathize with them. Tu Youyou’s discovery of artemisinin is one case; So was Huang Danian, who embodied "exploring innovation and serving the country with utmost sincerity." The symbols, formulas, and thinking in their minds can only be communicated with by a few people in the world. We must respect theoretical scientists because we don't understand their culture. Society must be tolerant, and the nation must be supportive.

Q: Basic research takes a long time, but companies must deliver returns.

A: We spend roughly 180RMB billion a year on R&D; about 60 billion goes to basic research with no KPIs, while around 120 billion is product‑oriented and is assessed. Without basic research, there are no breakthroughs, and we’ll never catch up with the U.S. [没有理论就没有突破，我们就赶不上美国].

Q: That’s true long‑termism. We heard Huawei even has a “Huang Danian Tea‑Thinking Room.”

A: Professor Huang was a great scientist. China noticed him during the Gulf War: the U.S. used a pod under helicopters to detect Saddam’s buried weapons and destroy them instantly—that pod had been developed by Huang at a UK university. He later resigned from his UK university position and taught at Jilin University. He spent his own money to get a 40 m² room, offering free coffee so students could “absorb cosmic energy over a cup of coffee.”

[Honestly this looks pretty lame, Huawei didn’t do the guy justice with this underwhelming study hall]

We have his family’s authorization to use his name for a non‑profit online platform—the Huang Danian Tea‑Thinking Room—where anyone can freely access global scientific information.

We also keep the “mouth of the funnel” wide open for basic research, partnering with universities. These are strategic investments with no performance metrics. Internally, we accept that we don’t know when theoretical work will bear fruit and impose no demands on the scientists.

“One purpose of socialism is to advance society”

Q: [A cringe leading question here…] Economist Richard Wolff and others argue that the U.S. lacks a high‑speed rail system because capitalism demands profit, while China’s socialist approach values social benefit: high‑speed rail, heavy‑haul lines, advanced power grids, expressways, rural cement roads, irrigation works, power plants… none make money directly, yet they underpin modernization. Competitive goods, meanwhile, are market‑regulated. What’s your view?

A: Why is it that only socialism tackles unprofitable tasks? One goal of socialism is precisely to develop society. China’s socialist market‑economy model is a magnificent undertaking. In infrastructure we had to follow this path—otherwise high‑speed rail, expressways, dams and so on would never have been built.

Q: What’s your outlook on AI?

A: AI might be humanity’s last technological revolution—barring perhaps fusion power. It will take decades or centuries to unfold. No need to worry; China has many advantages. [This line made leadlines in non-Party media]

Q: What advantages?

A: Hundreds of millions of young people—they’re the nation’s future. The General Secretary has said a nation’s strength rests on cultural flourishing. Technically, AI hinges on ample electricity and advanced networks. China’s power generation and grids are excellent; its communications network is the world’s most developed. The “Eastern Data, Western Computing” vision can become reality. [For more, see ChinaTalk’s past coverage of EDWC.]

Q: Anything else?

A: Chip worries are unnecessary: with stacking and clustering, our computing outcomes match the state of the art. As for software, open‑source ecosystems numbering in the thousands will meet society’s needs. [Wait, earlier in this interview I thought we were stressed about how behind Chinese chips were…?]

Q: How do you see China’s future?

A: After leaving our company, Tom Friedman bought a second‑class high‑speed‑rail ticket on his own to experience China. He later wrote, “I’ve seen the future, and it’s not in America.” [See ChinaTalk’s interview with Ezra Klein where I discuss this Tom Friedman piece with him. I, for one, would not take cues on analyzing China from Tom, who said in his Ezra interview that “Whether I’m writing about China from Washington, or whether I’m writing about China from China, I’m always just writing about America. My goal is to use China as my permanent Sputnik.”]

Q: Friedman said in that piece: “What makes China’s manufacturing juggernaut so powerful today is not that it just makes things cheaper; it makes them cheaper, faster, better, smarter and increasingly infused with A.I.”

A: Fundamentally, algorithms aren't in the hands of IT people, but in the hands of electrical experts, infrastructure experts, coal experts, pharmaceutical experts, and experts from various industries. At the practical level, Chinese manufacturing's application of artificial intelligence is very rapid, and many Chinese models will emerge.

Q: What support do private enterprises need from the state for their development?

A: Rule of law and marketization, with government administration according to law and regulations. Enterprises should focus mainly on value creation, technological breakthroughs, law compliance, and legal taxation. This harmonious development model will gradually release economic vitality.

Q: How do you view openness and development?

A: The country is becoming increasingly open, and openness will drive us to greater progress. Under Party leadership, the nation is administratively unified with smooth policy implementation. The gradual formation of a unified large market is possible and will certainly break through all blockades to achieve great rejuvenation.

[Ren Zhengfei of course got the seat of honor during the the Feb 17 Symposium featuring Xi which doubled as the Party’s olive branch to its leading S&T firms. We covered this meeting here on Xi’s Hard Tech Avengers.]

For more, see other core Ren Zhengfei speeches we’ve translated.

Read more

Has America already lost its dynamism in basic research? Josh Wolfe, co-founder of Lux Capital, joins the podcast today to discuss.

We get into…

• Why the Trump-Elon fallout matters less than you probably think,

• How much payoff corporate leaders are reaping from their campaign to appease Trump,

• The erosion of the U.S. research ecosystem, and how we should think about philanthropic giving amid that chaos,

• Parenting, strategies for emotional resilience, why short videos aren’t terrible, and the history of the machine gun

It’s a good interview. Listen on Spotify, iTunes, or your favorite podcast app.

Dreams of the Tech Right Meet Reality

Jordan Schneider: We have to start with Trump and Elon. I’ll let you pick the frame.

Josh Wolfe: I’m going to take a contrarian position — it’s irrelevant. It’s irrelevant because it dominates front-page news and captures everyone’s attention. I joked yesterday that if you were planning fraud or were a terrible company with bad news, yesterday was the perfect day to release it while everyone focused on this inevitable outcome.

People were making predictions six months ago — over/under bets on whether it would last three months or maybe a year.

Jordan Schneider: There were actual betting markets created for this.

Josh Wolfe: Exactly. You were betting under a year, maybe over three months, but none of this was surprising. This was an unstoppable force meeting an immovable object — kinetic chaos was inevitable. Elon served as a useful foil for Trump and did an excellent job helping him win the election. They were strange and unusual bedfellows.

Trump has historically been a strong China hawk, albeit with nuanced moments regarding Xi Jinping. Meanwhile, Elon depends heavily on China in ways people don’t fully appreciate — particularly Tesla’s production levels and the profit margins China contributes.

Jordan Schneider: Well, the China Talks audience understands this, but perhaps not everyone does.

Josh Wolfe: Right, not the broader ecosystem. This creates a significant vulnerability. You’ll see Elon Musk historically position himself as an outspoken advocate of free speech — though whether Twitter was genuinely about free speech is debatable. I’m more cynical about his true aims.

What you’ll never see him discuss publicly is anything about China. He’ll never criticize Xi for human rights abuses in Xinjiang province or regarding the Uyghurs, and he’ll never mention Tiananmen Square. Nothing. It’s “Free speech for thee, but not for me” when it comes to China.

Inevitable clashes were bound to occur. Was this really about the spending bill, or something deeper? When the leaks started — clearly planted by the White House over the past five days about drug use, and then the Epstein connection — it became salacious and interesting for everyone, but none of it was surprising.

My position is that it doesn’t matter. It’s similar to tariffs — Trump campaigned on tariffs, Democrats attacked him, saying he’d implement tariffs that would be economically troublesome. I don’t understand why these developments are surprising when they shouldn’t be.

Jordan Schneider: Speaking about your profession, many people in the tech ecosystem — particularly with the rise of the tech right throughout 2023 and 2024 — pinned their hopes on this man, similar to how Elon did. They hoped that by being inside the tent, they could influence policy trajectory. From my perspective, benchmarking to June 6, 2025, this has been pretty dramatically disappointing.

What are your reflections on that psychological arc? Are these guys all just patsies?

Josh Wolfe: During November and December of 2024, I remember peak rhetoric around the zeitgeist of America’s golden age. All you could hear was “we’re back, baby” and these ridiculous 1980s Top Gun-infused, Hulk Hogan maverick-style videos celebrating America.

Jordan Schneider: Sure.

Josh Wolfe: I told my wife, who runs an activist public hedge fund and manages our personal money, “We need to buy long, deep out-of-the-money puts” because all I heard was “to the moon” and “American greatness.” It felt like an echo chamber of optimism.

Two to three months ago when the tariffs hit, those positions performed very well in our portfolio. We’ve been relatively unaffected on the venture side.

One area where real change is happening — and it’s always dangerous to say “this time is different” — but it feels very different: defense. We’re seeing a combination of a nearly trillion-dollar budget and a huge shift toward autonomous systems, AI-driven software systems, and space satellites. Remember, Space Force was an absolute joke eight years ago.

Jordan Schneider: Okay, we’re pivoting a little…let’s stay on track.

Josh Wolfe: But this is where the venture world has made noise and impact. There are sympathies, appreciations, and influence, particularly in tech and defense.

Jordan Schneider: Sure.

Josh Wolfe: Regarding everything else: you had everyone pledging fealty, taking a knee, and donating a million dollars to the inauguration campaign. Zuckerberg, Bezos, Satya, Sundar — they were all paying homage to Trump, hoping to escape DOJ, FTC, or antitrust scrutiny. How that plays out remains to be determined.

The Trump-Elon situation was both weird and destined to end inevitably. There will be some strange reconciliation, and they’ll be bros again. But I don’t think this is just about the spending bill — it’s much deeper.

Jordan Schneider: One of the things the tech right didn’t price in is what’s happening around immigration and the basic research ecosystem. Let’s start with basic research. There’s this argument that we don’t need the NSF or NIH because corporate R&D has increased significantly over the past 30 years and can drive innovation forward.

As an investor who focuses on scientifically ambitious investing, what do you think the economy gets from basic fundamental research that happens in labs and universities versus corporate R&D?

Josh Wolfe: Long-term: everything. Short-term: it’s hard to see the value. For policymakers looking at budget cuts — particularly those focused on short-term gains — these programs seem like easy targets. This isn’t just a Trump phenomenon. In 20 of the last 22 years, we’ve had federal cuts to federal science funding.

Jordan Schneider: We had a brief moment with the Endless Frontier Act.

Josh Wolfe: Right, it started with Endless Frontier, which referenced Vannevar Bush’s work from around 1945. Then you had “The Gathering Storm” by Norm Augustine, the former Lockheed executive, warning about America’s talent base and cultural shifts in what people were drawn to and celebrating.

What they didn’t mention in that report — now over 20 years old — was this anathema, this zeitgeist against the military-industrial complex, precisely when China was not only embracing but mandating military-civil fusion.

Over the past two to three months, we’ve witnessed a perfect storm. Now you have the politicization of academia. Harvard sits number one in Trump’s crosshairs — some speculate because Barron applied and didn’t get in, creating some Shakespearean vendetta. Whatever the reason, while anti-Semitism was the stated concern, it was far worse at Columbia, right here in New York City, than at Harvard.

Subscribe now

According to Nature magazine’s rankings — which publish various metrics including H-index measurements — when you look at the most important scientific publications by institution, the top 10 includes Harvard at number one, but positions two through ten are all Chinese universities.

Fifty percent of AI graduates worldwide come from China. Thirty-eight percent of the AI and computer science workforce here comes from China — not native-born Americans. They’re outpacing native American citizens by 35-37%.

We’re losing the talent game. What should we do to win? What we did in World War II and beyond: attract the best and brightest. Create brain drain from other countries because people are repelled from their home nations.

Eastern European Jews in the 1940s established the Institute for Advanced Study and brought Einstein here. Having the atomic bomb developed here rather than Germany was a net positive for the world. Soviet émigrés in the 1980s escaped communism for capitalism during the Cold War.

We should be stapling visas to every Chinese, Indian, Pakistani, Israeli — anyone who wants to come here. We should help their parents immigrate too, because often family remaining in authoritarian home countries becomes leverage that governing regimes use to control whether these individuals can leave or return.

The last Republican I remember discussing this was nearly 20 years ago: Newt Gingrich, who called for tripling the NSF budget. He understood that everything from Genentech (which emerged from the UC system) to Google (from Stanford) was premised on long-horizon scientific research.

This isn’t just computer science — it’s chemistry, physics, materials science, and all the breakthroughs that emerge from these fields. If we knew what those breakthroughs would be, we’d fund them today. But we don’t. We rely on this rich ecosystem in our own self-interest to support brilliant people who generate breakthroughs.

These breakthroughs often result from combining insights from one department with research from a different university. More than ever, this cross-pollination is critically important. We’re ceding academic and intellectual leadership to other countries. China will likely be the greatest beneficiary.

International students used to comprise roughly 25% of our student body. Now it’s down to 12-15% and falling. China is actively recruiting with programs like their Thousand Talents initiative.

This represents the greatest self-inflicted wound Americans have created in several generations.

Jordan Schneider: You had this line: “If the Politburo were drafting America’s self-sabotage plan, they would politicize science and freedom of inquiry, starve young and ambitious investigators of funding, and discourage the best immigrant minds."

To recap for everyone: the NSF is spending at half the rate it was in 2024, despite authorized funding. The new budget dramatically cuts science and research spending. Universities have been cut off wholesale from billions in research dollars. Johns Hopkins is laying off 1,000 people because they’re worried grants won’t materialize.

Trump said yesterday that we actually want Chinese students, which creates bizarre mixed messaging when his administration simultaneously tells Harvard they can’t admit Chinese nationals.

Josh Wolfe: Some of these grant cuts are both tragic and absurd. There was a researcher studying neurotransmitters for neuroscience whose grants were cut. They couldn’t understand why until they realized the word “trans” triggered a five-letter string search that flagged anything with “trans,” “DEI,” or similar terms.

Now researchers have to change grant language — writing “neurons firing intracranially” instead of “neurotransmitters."

Jordan Schneider: That’s not funny, though.

Josh Wolfe: You’re right. It’s tragic.

Jordan Schneider: Why don’t you talk about the $100 million pool you launched?

Josh Wolfe: It’s specifically not grants because we’re not trying to substitute the charitable giving of our government — funded by taxpayers — which actually makes America great when we have scientists working on breakthrough research.

Historically, about 10% of our investments are de novo new companies, and most emerge from academic labs. We find a principal investigator — a fancy term for a scientist at an academic institution — whether at Harvard, Princeton, Yale, Georgia Tech, Cornell, or elsewhere.

The early 1980s Bayh-Dole Act, sponsored by Senator Evan Bayh and Senator Bob Dole, allowed universities to own intellectual property created with federal taxpayer research funding. When a scientist at these institutions files a patent, the university becomes the assignee while the scientist remains the inventor.

We, as a venture capital firm, can license that technology. There’s a well-established deal structure involving licensing royalties and equity — the scientist typically gets about 25% ownership. This mirrors Google’s “one day a week” policy, allowing employees to spend time on personal projects, but applied to exclusive company work.

This approach for launching companies from academic labs is well-established. Professors often stay at their institutions while postdocs handle translational research and join the company. We’ve created about 25 de novo companies this way, spanning everything from 4D lidar for autonomous vehicles to digital olfaction — essentially “Shazam for smell” — to cancer therapeutics and Nobel Prize-winning work from the Howard Hughes Medical Institute that enables real-time cellular imaging.

Historically, this represents about 90% de-risked investment, meaning we take roughly 10% scientific risk. Now, with grant cuts and layoffs at institutions like Johns Hopkins, exceptional research is being abandoned — like proverbial Rembrandts in the attic. Scientists are wondering what to do next. If they wait six months for resolution, they need to find work elsewhere, whether in academia, nonprofits, or the private sector.

Our response is the Lux Science helpline — a bat signal for struggling researchers.

We’re dedicating $100 million from our latest $1.2 billion fund to double down on early-stage science risk. Instead of taking 10% science risk on 90% proven concepts, we’re willing to take 50% science risk on half-baked ideas.

We’ll help license patentable work into our existing companies or arrange sponsored research to continue funding relevant projects. For example, someone contacted us about novel materials for radiation hardening in space. We have 17 companies across the space ecosystem that could sponsor continued research.

Other scientists ready to enter the private sector can work with us to start new companies around their expertise. We’ll license their work, assemble their team, build laboratories, and launch them.

This is just a drop in the bucket — we can’t do this alone. We need dozens, if not hundreds, of other VCs to recognize the value of funding early-stage scientific work. It’s profitable, beneficial for national security, and good for business. Our $100 million allocation, while significant, isn’t sufficient by itself.

Jordan Schneider: I’ve received numerous emails from intelligent people essentially doing backfill work for USAID funding cuts. While it’s admirable that people are donating $10 million or $50 million, tens of millions of people will suffer, and we’re delaying future breakthroughs because we’re not conducting this research. The laboratory animals aren’t being fed, and you need the time, energy, and tacit knowledge built over years of lab work to accomplish meaningful research.

Josh Wolfe: Absolutely. This isn’t purely black and white — there are legitimate concerns about USAID politicization and questionable funding destinations. There are appropriate and inappropriate uses for that money.

My main criticism of the NIH isn’t to strip their funding, but to bias toward young investigators. Too much grant money goes to, frankly, older researchers. Older scientists can be set in their ways and resistant to new approaches.

You need that potent combination of naivety and ambition that drives both entrepreneurial and scientific discovery. Young researchers have the arrogance to say, “I know better than you — why wouldn’t we try this?” while older researchers respond, “Why would you do that? I’ve tried it ten times and it never works."

I would restructure the NIH budget and federal funding to favor younger investigators.

Jordan Schneider: I wonder to what extent the Silicon Valley tech ecosystem that embraced Trump and the DOGE energy — which led to all these cuts — stems from the fact that for the past 20 years, you could make the most money by being 19 and writing software. That’s not something you necessarily need a PhD for. The social returns to that work are very different from what you see when making novel materials and drugs.

Josh Wolfe: Absolutely. Even the term “engineer” shifted from physical engineer to software engineer. For at least 20 years post-internet boom — through the SaaS enterprise boom and cloud boom — that was the dominant paradigm.

Lux was on the periphery of that trend because we don’t really fund software. About a third of our investments focus on healthcare, biotech, robotic surgery, and medical devices. Another third covers aerospace, defense, manufacturing, and industrial applications. The final third is what we call “core tech,” defined more by what we don’t do — very little internet, social media, mobile, or ad tech.

Great fortunes were made and great companies built during the software boom. Interestingly, there’s a China parallel here. Fifteen years ago, Xi Jinping designated software as a domain China would fund heavily. This created incredible companies and ecosystems, and many US investors funded them, believing China was a democratizing, growing market. Knowing what they know now, it would be difficult to justify investing in companies like ByteDance.

Jordan Schneider: Unless you’re Bill Gurley [who recently funded Manus AI].

Josh Wolfe: I love Bill, and he’s not necessarily part of the partnership decision. We’ve discussed this, and he would argue there might be scenarios where teams like Manta are trying to exit China. I think this deserves more study rather than just criticism. However, I personally wouldn’t invest in Chinese companies that are part of CCP military-civil fusion.

The shift also happened within China. At the last Politburo gathering about a year and a half ago, the people close to Xi weren’t software engineers or computer scientists — they were from space, biotech, rockets, and defense. All hard sciences.

This coincided with when Jack Ma went to “spend time with his family” and when entrepreneurs were capped at $9.9 billion. Anyone exceeding that threshold was essentially decapitated — they could start schools or enter education, and either they or their family could leave the country, but not both simultaneously.

The new directive became hard sciences. I used to joke that if you wanted to make money, the greatest capital allocator wasn’t following Warren Buffett or Seth Klarman or other great value investors — it was listening to what Xi Jinping was funding, because that’s where the world was heading.

I agree that for the past 10-15 years, the focus has been software. Many VCs, including Marc Andreessen who said “software is eating the world,” were drawn to Trump partly because they felt rejected by Biden. They were spiting Biden and possibly spiting themselves long-term.

I still believe the greatest entrepreneurs don’t really care about political developments or the 10-year interest rate. They’re building something because they have a chip on their shoulder — and as I always say, chips on shoulders put chips in pockets. They’re driven by private ambitions with timeframes that supersede one, two, or even three presidential cycles.

I remain bullish on great entrepreneurs. There’s simply a shift from software to hardware.

Jordan Schneider: Something worth exploring is the importance of tacit knowledge in hardware versus software and the learning that happens almost entirely in universities during master’s and PhD programs. When you’re doing actual scientific work, you need some sense of navigating the dark forest.

Josh Wolfe: That’s absolutely true. This applies even to semiconductors and manufacturing. TSMC’s Arizona facility is starved of talent — not just union workers doing physical assembly, but specialized expertise. There’s a scarcity of talent because much of that tacit knowledge remains in Taiwan and is difficult to transfer here.

You see this dynamic in companies working on the most sophisticated technologies. My partner Sam Arbesman, who you know or have spent time with, is a brilliant scientist-in-residence here. He has a new book coming out called “The Magic of Code,” and his previous book “Overcomplicated” offered a modern version of “I, Pencil” — the thought experiment about complexity.

If you were to make a semiconductor today, or an Apple iPhone, consider the number of components, the tacit knowledge required, the number of countries and companies involved — it’s extraordinarily complicated. No single person can make a pencil, let alone a semiconductor, chip, GPU, or field-programmable gate array.

Culturally, we get what we celebrate. For 25 years, we at Lux have complained that American culture celebrates celebrities — the Kardashians, the Hiltons, and similar figures. You see this manifested in TikTok and what gets fed to American users versus what’s not even allowed in China and what China celebrates in terms of STEM education.

We’re losing this battle terribly. Looking at the labor pool: we have 300,000 undergraduates in science here. China graduates approximately 1.2 million. Half of our 300,000 are foreign students.

This represents a cultural crisis regarding what we want our children doing and celebrating. We don’t need more people in marketing, advertising, or selling products. We need people inventing things that everybody else in the world wants to buy.

Jordan Schneider: And podcasting.

Josh Wolfe: But you produce intelligence and insight, which is valuable. It’s unique insight because, as you noted, people who follow China Talk understand things that others don’t. That’s an advantage — you produce something intellectually valuable.

Jordan Schneider: Back to the narratives. There was this fascinating Twitter exchange a few days ago. You mentioned the atomic bomb and the Apollo program, and J.D. Vance framed it as “we didn’t need foreigners to do this — we had 600,000 Americans weaving the wires to connect everything together."

Both perspectives are true: his narrative about American workers is accurate, and your narrative about leading lights of both the rocket and nuclear programs coming from Europe is also true. But this narrow-minded “you have to be born here to be part of the circle” mentality — let’s discuss what that means.

Josh Wolfe: This created a significant fissure. People entering the administration who were immigrants from India were being lambasted by these self-described “heritage Americans.” Elon defended them because he’s an immigrant himself — not the classic immigrant from poverty in India or Russia, but an immigrant nonetheless.

He argued there’s a distinction between immigrants crossing the border and taking blue-collar jobs — who may not conform to American society and could include criminals or drug dealers — and brilliant people making extraordinary contributions. For example, the recent genetic medicine breakthrough that cured a baby using one of the first in vivo genetic editing approaches involved two scientists, one whose parents arrived from India 40 years ago. Thank God those individuals came here and had children.

We want this brain drain — from World War II through the Cold War to our current version of Cold War competition. I don’t understand why people consider “immigrant” a dirty word. Immigrants literally are the fabric of this country.

One of my mentors who put us in business, Bill Conway, who founded the Carlyle Group, focuses his philanthropy on addressing one of America’s biggest problems: nursing shortages. He says if you’re anti-immigrant, good luck getting sick, because our entire healthcare system depends on immigrants playing critical roles — from hospital orderlies to doctors to neurosurgeons.

We want the best and brightest coming here. It’s a complete self-inflicted wound to say “America First” means shutting out talent. America First doesn’t mean America Alone, as Scott Bessent has said. Scott is one of the great adults in the room and a friend.

People ask why Scott doesn’t speak out more publicly on these issues. Scott is a student of markets first — he understands currencies and countries — but he’s also a student of human nature. He understands in a Shakespearean way who his boss is, who he works for, and how to manage that relationship. We want him doing exactly that because we don’t want him out of the job.

Jordan Schneider: We discuss immigrants and science frequently on China Talk from a US-China competition perspective, and that’s all valid. But the fact that people want to come here is remarkable. Look, at the margins there are legitimate concerns about completely open borders, but limiting human flourishing by building walls and splitting families is just —

Josh Wolfe: I’m going to be blunt because I come from Brooklyn: this approach is counterproductive and self-defeating. The idea that we’re limiting the immigration of brilliant people who could make this country better is fundamentally wrong.

Our greatest export isn’t our music, Hollywood, fashion, high-tech companies, banking system, or rule of law — it’s the American Dream. The American Dream is arguably our greatest export; it’s what everybody wants. The measure of a great country is whether more people want to enter than want to leave.

Walter Wriston said, “Capital goes where it’s welcome and stays where it’s well-treated.” This applies to both money flows and people flows. I hope the rhetoric will change and that there will be enough great American leaders to counter jingoistic “America First” voices.

There’s a delicate balance: we must remain that shining beacon on the hill, attracting people while distinguishing between good people who genuinely want to build better lives here — who want to embrace the American Dream and become proud Americans — and bad actors who want to undermine the country. Many of the latter are coming through our university system and may be funded by foreign actors, but that’s a separate issue.

There’s virtue in virtuous people coming here and the American Dream being upheld and celebrated. Right now, we’re casting shadows on that dream.

Jordan Schneider: You opened by saying the Elon-Trump fight is noise. But the signal is that people around the world would rather live here than in China.

Josh Wolfe: They still do. My wife mentioned recently that many countries hate us right now — Canada, tourism is down — and there’s probably some schoolyard “arms crossed, I’m not playing with you” attitude.

Jordan Schneider: Sure.

Josh Wolfe: The reality is, I read about how Canada supposedly hates us, then the next headline I see — because I read many papers each morning, including the Globe and Mail — is “Canada wants in on Iron Dome."

The rest of the world, even if they hate us, still wants our superior military technology. The zeitgeist of popular antipathy will fade, but people will still want our materiel — our military materiel.

Jordan Schneider: We’re recording this a few days after the Tiananmen anniversary. The fact that hasn’t happened here — the fact that there is free speech, elections, due process, and habeas corpus — represents incredibly powerful long-term advantages. These aren’t just abstract principles — they’re things that make life worth living and make you excited to get out of bed in the morning.

Josh Wolfe: I remain relatively optimistic personally. I’m always optimistic about technology, science, and ultimately the human condition. I’m always skeptical about people because I’ve read a lot of Shakespeare. Technologies are amazing, science is amazing, and both will continue to progress. People generally disappoint because they’re vainglorious, full of ego, petty, jealous — they connive and deceive.

People ask how I can be optimistic yet cynical. That’s the pairing — I’m optimistic about science and technology, skeptical about people. I love Tom Wolfe’s answer when asked why he writes about space, astronauts, and moneyed Upper East Siders but not politicians. He said politicians are like passengers on a train, and the country is the train. The tracks might go up and down, but it’s generally heading in the right direction over time. Inside the train, there are clowns on one side in red and clowns on the other side in blue, throwing pies at each other nonstop. Every four years, the engineer changes, but the train stays generally on track.

I take that view: the real thrust comes from people, values, that great beacon on the hill, the American Dream that attracts people here. We shouldn’t deter them from building here on that track. Most entrepreneurs couldn’t tell you much about the political system or what’s happening at the Fed or Treasury — they’re focused on building.

Jordan Schneider: But the fact that they have to worry now, that they have to fear their visas getting revoked — look, in other moments that might be manageable. But within our range of expectations, there are definitely scenarios where an executive acting alone can really bend those train tracks.

Choosing Optimism + Parenting

Jordan Schneider: Speaking of Shakespeare, we did a show with Eliot Cohen about his wonderful book The Hollow Crown and another episode as a Biden emergency discussion. The Trump-Elon situation probably deserves its own Shakespeare emergency episode.

told me to ask you this question — does America need a Gen Z Marsha Linehan?

Josh Wolfe: Marsha Linehan is one of the founders of CBT — Cognitive Behavioral Therapy.

I love Derek. I got into psychology maybe five or six years ago. My kids are 15, 12, and 9. My oldest daughter is much more like me — emotionally volatile. I’ll have high highs (not bipolar…emotive!) then calm down. My wife and middle daughter have slow burns with grudges lasting three or four days.

We learned CBT and DBT techniques as a family. Derek’s question is whether Gen Z needs this — I think they do. CBT is essentially Stoic philosophy in a clinical psychology setting. I wish I’d learned this 20 years ago as a kid or teenager — I would have had healthier relationships. I wish I’d learned it 15 years ago before my first child was born — I would have been a better parent. We have a better marriage, family relationships, and even professional interactions because of these principles.

First, avoid extremes like “you always do this” or “you never do that.” If I said “Jordan, you’re always late for podcasts,” your first reaction would be defensive because it’s not true — nobody “always” or “never” does anything. People shut down and become defensive. You want to find the dialectic, avoiding black-and-white thinking. “Sometimes I feel frustrated when you’re late” is much more effective.

Second, when someone has an emotional outburst, they’re dysregulated for various reasons — hunger, a bad day, a lack of tools. If my daughter has a fit, she’s not thinking “I really want to have a tantrum and lose my phone for a week.” You appreciate that they’re doing their best with what they have — they just lack the tools.

Third, validate emotions. When someone’s really upset about something, you could say, “That’s ridiculous, it’s just a math test, you’ll get an A.” But explaining away someone’s feelings doesn’t help. People need validation. They become more emotionally frustrated when they don’t get that release valve. Saying “I can see you’re really upset” sincerely can help reduce their emotional burden.

Returning to Derek’s question about Gen Z, many people seem massively oversensitive. I grew up in Coney Island, Brooklyn, where people talked trash to each other, were rough around the edges, and the world wasn’t safe. You could say all kinds of things. I’ve raised my kids similarly — I don’t want them so soft that when someone says something offensive, they appeal to authority, running to school or teachers saying “he said this” or “she said that.” Take care of it yourself. Have a conversation. I’m not advocating violence, but handle yourself.

My kids have experienced this and tell us about various scenarios. The younger generation — whether you call them far left or woke — has lost some ability to engage with people. When encountering ideas or comments they disagree with, there’s hysteria.

I was a center-left Democrat my entire life. I didn’t vote for Trump, but I also didn’t vote for Kamala. I voted for Bloomberg-Romney, which was not a ticket, but that’s where my values aligned, in a non-swing state where my vote didn’t matter. I had the luxury of voting with my conscience.

I believe many center-left Democrats didn’t move right because they were attracted to the right. They were repelled by the histrionics, noise, whining, and complaining from the left. Enough was enough.

Share

Jordan Schneider: Let’s stick with parenting. You mentioned earlier that Steve Jobs accomplished incredible things, but everyone around him in his personal life had a totally different experience.

Josh Wolfe: They hated him. This is a fascinating phenomenon.

Jordan Schneider: This really resonated for me when I was reading a biography of Richard Holbrooke — the most storied American diplomat of the second half of the 20th century, though perhaps not well-known in the Asia-watching community. There was a line where his kid said his father wouldn’t recognize his grandchildren in a toddler lineup.

Growing up, I dreamed of being Secretary of State or a diplomat, but trading family relationships for that achievement gave me pause. Many people choose one path over the other. What are your reflections on this dynamic of accomplishing big things while being awful to the people closest to you?

Josh Wolfe: I used to debate this with one of my best friends, a famous journalist, who talked about how great Jobs or Elon were. My response was that the people around them — their families — hate them. They’re celebrated by strangers, feeling love from millions of people they’ve never met who celebrate a caricature of who they actually are. Meanwhile, their true character is loathed by those closest to them.

I’ve admired something I have to admit I’ll never achieve — I am too rough around the edges, have burned too many bridges, and have been abrasive to too many people. When Warren Buffett gave the eulogy for Coca-Cola CEO Don Keough, he summarized it in three or four words — “Everybody loved him.” I thought that was beautiful.

People will not say that about me, but I can control the decisions I make regarding my children. Will my kids feel that way about me?

I experienced something very salient and memorable involving my grandfather, who raised me. I grew up in Coney Island, Brooklyn — my mother, grandparents, and I shared a two-bedroom, one-bathroom apartment. We were very poor. My grandfather wasn’t my biological grandfather; he was my grandmother’s second husband, but he treated me like his son. He delivered the Daily News at night and was the most important man in my life.

He passed away the month before September 11th. At his funeral, his biological son was present. During the Jewish ceremony, when family members put dirt on the grave, his biological son approached with such animus. It was like “good riddance” — he took the shovel, threw dirt on the grave, and walked away. That sound still echoes with me.

I’ve always loved this quote from Carl Jacobi, the 18th-century mathematician: “Invert, always invert” — flip it on its head. My father wasn’t present in my life, so I became the father and husband he wasn’t. But watching that funeral moment, I resolved that would never happen to me. I never want my kids to feel that animus and animosity, wanting to dump dirt on my grave and walk away saying “good riddance."

The most important thing to me is what Adam Smith wrote about — not just the invisible hand, but the idea of being lovely and being loved. That second part, being loved, is scarce and valuable.

Scott Galloway recently gave a rant — I believe it was on Piers Morgan — where he talked about our obsession with Elon, innovation, and money while ignoring people who are hurting and suffering, including those losing USAID funding. He asked what it means to be a man. In his view, it wasn’t toxic masculinity — being a man means being able to take care of people.

We’ve lost that. There aren’t great male role models in public life today that young people celebrate for being good men through self-sacrifice. Everything centers on self-aggrandizement.

Regarding how to encourage or discourage people on certain paths: I want my kids to be truly happy. I ultimately don’t care where they go to college — the world has changed significantly. I don’t care what they do professionally. I want them to find purpose and meaning.

This relates to something interesting. Dan Senor works at Elliott Associates but is very active in Jewish and Israeli life. We attended Shabbat dinner at his house. I’m not religious — I’m an atheist but a tribal Jew. He blesses his children, and what struck me was that the blessing wasn’t about success, money, or career achievement. The blessing was “I want you to have a life full of meaning and purpose."

I thought that was beautiful. That’s what I want for my kids.

Philanthropy, Parenting, Short Videos

Jordan Schneider: We have five minutes for five quick questions. Take your pick — future of media advice for China Talk, New York mayoral race hot takes, what you’d do with a $10 billion philanthropic foundation, or something else.

Josh Wolfe: I don’t know.

Jordan Schneider: Where do you want to go?

Josh Wolfe: Mayoral race. I really hope Cuomo wins.

On advice for ChinaTalk, people get the audiences they deserve, and you have a smart, sophisticated, engaged audience looking for signal among the noise. Keep doing what you’re doing — your combination of Substack, podcasts, video content, periodic pieces, and great guests works well. I feel privileged to be here with you and enjoy reading your work.

Just maintain it because it’s a high-signal, no-BS voice. These things aren’t linear — there are step-change functions where suddenly something goes viral and you gain another thousand or ten thousand subscribers.

You should periodically write op-eds in major publications like the Wall Street Journal or Financial Times, sharing insights from your work. There are proxies like Stratfor and others in popular geopolitics, but you own a valuable niche. Over time, whether you do it or someone else does, there will likely be aggregation and acquisition — a China expert, Africa expert, defense expert — building a new media ecosystem where you could benefit from that outcome.

Jordan Schneider: Second and third questions — you’re pretty wealthy and likely to become wealthier. How do you think about big philanthropic investments beyond what you’re doing today?

Josh Wolfe: There are two aspects I really admire. Take what Bill Conway did — he’s not pursuing vainglorious naming opportunities. He literally identified deficits. His late wife was focused on nursing, so they fund nursing schools and programs because we have absolute scarcity there.

Considering the arc of AI, I believe in abundance and scarcity dynamics. What’s abundant will be machines helping with intellectual tasks — white-collar jobs will be hit while blue-collar jobs will surprisingly be safer than people think. The care aspect of healthcare will be critical.

If you ask what I want my kids doing: I grew up playing extensive video games and watching tons of TV, and I allow them that — not Monday through Thursday, but weekends they can play games, watch TV, engage in pop culture, watch sophisticated content like Fareed Zakaria and Jeopardy.

But I want them fully versed in AI. My 9-year-old is better than my wife at ChatGPT queries and Midjourney prompts for images. It’s creative expression.

Jordan Schneider: That’s a great age for it, right?

Josh Wolfe: Absolutely. My oldest daughter had to do an evolution project in seventh grade by hand — organisms with sharp teeth survive candy rain while weak-toothed ones die. She drew it manually. My current seventh-grader in that same class is using one of our companies, RunwayML, for AI video generation, creating full videos of different organisms in her scenario.

I want them totally versed in AI because ultimately — and you discussed this recently, maybe with Wang — the power isn’t in who has the chips, but who’s using them. Similarly, power isn’t in who has the applications, but who’s using them.

There’s a significant push for the world to use US-driven open-source or closed models rather than China-driven models that approach an asymptote of truth but never discuss Tiananmen Square, Xinjiang, or Uyghurs. We want the Global South influenced by American ideals of truth, Popperian hypothesis, conjecture, and criticism rather than Chinese systems.

But I want my kids using all these tools and understanding them. What will be scarce against all that abundance is human connection. They need to understand people, make eye contact — you’d be amazed how many kids, because of screens, have awkward, almost autistic interactions.

Being able to connect with people, understand them, read Shakespeare — that’s timeless. People change, costumes change, stages change, but human nature hasn’t changed since the Pleistocene African savannah.

That’s what I’d fund philanthropically. Derek mentioned CBT programs for young people earlier. I started a charter school 17 years ago in my native Coney Island, Brooklyn. We began with 90 fifth-graders in the projects. Now we have 1,000 scholars, 200-plus faculty, 100% college acceptance rate for first-generation college students. Eighty percent of families qualify for free and reduced lunch — a euphemism meaning a family of four makes less than $30,000, which is insane.

These families lost the ovarian lottery — the classic John Rawlsian veil of ignorance. These kids are no less intelligent than those born in Greenwich, Connecticut. But there’s no Army recruiting station on Greenwich’s Main Street — there is one on Cropsey and Stillwell in Coney Island. That’s not fair.

Those are worthy targets for philanthropic dollars.

Jordan Schneider: Okay, but let’s start from a $10 billion bucket. What are we talking about here?

Josh Wolfe: Where would I give? I would fund universal CBT for everybody in the country. I don’t know that it needs that much money — it just needs celebration in the country that helps people become more emotionally regulated and be their best version of themselves. It will reduce problems in our criminal justice system. It’ll reduce problems in corporate America. It’ll reduce a lot of problems across the board.

Jordan Schneider: Not to rag on you too much — you give very sophisticated answers to how to invest in the future of science and technology. That was a fine answer. But a lot of people, at a certain point in their life, switch from the sort of answers we discussed in the first 80 minutes to the question I just asked you. I’m curious: do you see the future differently when it comes to philanthropy?

Josh Wolfe: Where we give philanthropically right now reflects things that we prioritize. For me, complexity science through the Santa Fe Institute — brilliant people. I love it. I believe that’s a source of tremendous value. I’ve been part of that for 10 years as a trustee and believe deeply in it.

The charter school movement — I believe deeply in that because it’s a form of civil rights for people. My mom was a public school teacher, so this hits close to home.

Jordan Schneider: Let me try one more time. This is more of a meta question. The sort of investigation that you need to understand how to use philanthropic dollars efficiently and effectively — I’m curious how similar or different you think that is from investing?

Josh Wolfe: The only similarity is finding an amazing social entrepreneur. It’s like when we started the charter school — we basically backed this guy Jacob Newkin, who was starting the school. It’s the same thing. I used to talk publicly about Jacob: he’s the greatest social entrepreneur that I’ve backed by spending my time and money with him.

But we weren’t doing analysis on the market and the unmet need and that kind of thing. It tends to be something personal. For my wife, it’s the Center for Reproductive Rights. She’s on the board there, making sure that women have access to contraception and abortion and autonomy over their bodies. That’s a really important thing to her.

She’s not doing an analysis of where’s the best place to give or whether we should give more. It’s just: Roe v. Wade got overturned. There are women who are going to die in certain states because they can’t get abortions. What can we do about that? She gives a lot of money.

Jordan Schneider: Maybe we’ll close on this topic. If we’re entering a world where science receives less funding — Danny Crichton, who works for Lux Capital, wrote a really interesting piece about this — when the total amount of science that the US funds decreases, there’ll be a little spillover to China and the EU, but we’ll just have less science overall. Beyond CBT, what encouraging developments do you see for science and technology’s future? What basic research do you think people should be funding — the stuff that’s too risky for any venture capitalist to invest in?

Josh Wolfe: This might be controversial, but people should be spending far less money on climate philanthropy. The answer lies in what I call elemental energy and nuclear power. All that money should be redirected toward early-stage science and psychological and behavioral health research, because that will make society better.

The Gates Foundation, Bloomberg Philanthropies, Ford Foundation — these organizations were captured by the climate movement over the past 15 years. Honestly, I don’t know what it’s accomplished. It’s been a colossal waste of money.

Jordan Schneider: Speaking to the culture issue you raised earlier — it’s striking to me. I worked at an organization that was half nonprofit, half research fund, focusing on China and climate. On the climate side, they could get money to fund literally anything.

Josh Wolfe: It’s popular and makes you feel good. You go to a cocktail party and say, “Oh, I’m funding climate research.” Great, you’re doing wonderful work — Al Gore would be really proud. But the money is misplaced.

Jordan Schneider: It’s almost downstream of the culture. Who are the funders and trustees, and what’s popular with them? They’re not scientists conducting expected value calculations on human flourishing or whatever. Not to disparage anyone, but it’s something that resonates with them personally.

Josh Wolfe: Look, Sam Bankman-Fried was the emblematic figure of this, but the effective altruism movement was rational in trying to determine where we can do the most good. They approached it economically, looking at low-probability, high-magnitude events and identifying opportunities where small amounts of money could have significant leverage.

Going back to Conway’s nursing initiative — that’s not popular. People don’t get excited about addressing the massive nursing shortage. But he identified this as crucial, and they’re putting several billion dollars behind it. That’s noble work.

Bloomberg’s urban initiatives and charter school funding are excellent. People funding the arts because of personal passion — that’s great too. But we have massive problems with criminal justice reform and behavioral health domestically.

I’m not talking about everyone needing mental health days, but implementing cognitive behavioral therapy in schools at a young age. Before children’s prefrontal cortex develops at 25, we could help them develop better self-regulation. The world would be a much better place.

It would be tremendous to see philanthropists return to funding institutions like the old Cavendish Laboratory — putting billions of dollars into institutes that enable knowledge discovery.

Jordan Schneider: Rockefeller University — that’s incredible work. People need to get with the program.

Josh Wolfe: Carnegie Mellon, exactly. These institutions started with robber barons who decided to redirect money into academic institutions. This will happen again.

If you examine philanthropic funding historically, it began with private individuals, then government labs like Los Alamos, followed by Bell Labs (born from monopoly), then IBM Research. IBM centralized initially, then distributed with locations in Zurich and Almaden. Then came Google, Intel, and Microsoft Research.

Some of these corporate labs are under pressure now because they haven’t yielded significant results. But we’ll see the rise of private labs. You can see this already with the ARC Institute.

Jordan Schneider: Absolutely.

Josh Wolfe: The Collison brothers are major supporters there. The Chan Zuckerberg Initiative represents major scientific initiatives comparable to Howard Hughes — himself a former defense contractor who invested enormous amounts into what became the Janelia campus, now one of the great sources of Nobel Prize winners.

Jordan Schneider: That’s a nice place to close. 50% of Howard Hughes researchers came to the US on visas! Joshua, thanks so much for being part of this.

Josh Wolfe: Great to be with you, man.

Jordan Schneider: Awesome. We’ll do this again in 10 years, and then you’ll be saying, “Here’s this philanthropic vision — look at all these molecules I found."

Josh Wolfe: I believe mostly in free enterprise, science, and technology. Early-stage ventures will handle a lot of that, but not the basic science research. There’s no market for that.

Bonus Riffs on Books

Cool. What are you reading? Anything good? Binge-watching anything inspiring or fun?

Jordan Schneider: Two books. I’m a new parent — I have a 10-month-old at home.

Josh Wolfe: Boy or girl?

Jordan Schneider: Girl. First child.

Josh Wolfe: Wow.

Jordan Schneider: I read all these books on paternity leave about “How I Raised My Child in X Country.” They were not good. But one was excellent: Italian Education. It’s by a cranky British guy who married an Italian woman and raised his kids in late 1980s-90s Verona. All these other books are basically backhanded critiques of American parents from whatever direction. But he’s just observing this really interesting, weird society where you have very tight connections between parents and kids — for better and worse, from my perspective, across many dimensions.

It was engaging, funny, smart, with nice vignettes. Each chapter stands on its own, which is good for 3 AM reading.

Book number two — A Social History of the Machine Gun by John Ellis.

The book tells the story of the machine gun through different lenses. He’s a military historian who wrote about World War I and World War II tactics. But in this book, he explored the technological evolution from the Maxim gun all the way through World War II — who invented it, why, and where it came from.

There’s a fascinating acquisition story too, because people didn’t think it was real and didn’t want to buy it. There were prototypes but no factory yet, so you had these hucksters trying to —

Josh Wolfe: What years was this?

Jordan Schneider: At the start of World War I, the British had one gun for every 2,000 people. You had to get to 1916-1917 for them to actually be making and buying enough machine guns.

Josh Wolfe: That’s amazing.

Jordan Schneider: It took an enormous amount of time. The technology was already there in the 1880s and 1890s. There were examples from wars in different countries — the 1905 Russo-Japanese War, the Crimean War. You could see it if you were looking at the right things.

Ellis gives examples of smart colonels saying, “Guys, we need to buy these guns — they’re a big deal.” But people said no, and they all said no way too late.

The other story he tells is about the psychology of not just the people buying the guns, but the officers themselves who had to abandon their mindset about what made a successful officer. Being a sharpshooter wasn’t considered honorable. What won wars throughout the 19th century was discipline, standing in line without fear, marching together to bring maximum power. That’s what the technological paradigm demanded — willingness to maintain rank.

It took over 100 years for people to change their mindsets and understand that you actually need to be distributed, use natural cover on the ground, and get away from the Napoleonic mindset of gallant charges. Those charges were the correct evolutionary answer in a different time period, but not by the US Civil War, definitely not by the end of the 19th century, and absolutely not by World War I.

Josh Wolfe: It’s interesting — the juxtaposition of the two books. One is arguably about technology of life (all parenting is a form of life technology), and the other is about technology of death. It’s a nice contrast in what you’re reading.

Quick parenting observations — First, if you walk into a bookstore like Barnes & Noble — which really don’t exist anymore —

Jordan Schneider: There’s one three blocks away!

Josh Wolfe: The mere existence of a handful doesn’t change that they’ve largely disappeared. There used to be 20 in New York City; now there are one, two, or three.

My point is, when you go into any section — investing, relationships, or parenting — and see 200 books, it means nobody has any idea what they’re doing. If they did, there would be one book with all the answers.

What I learned, especially with our first child, is that you develop a whole bunch of lessons, then you have a second child and they’re all wrong. Once you have more than one child, the nature versus nurture debate is settled. They are genetically different from day one — their predispositions, attitudes, sleep patterns, crying patterns, wants and needs. Their personalities persist from birth. The one who’s more reactive, the one who’s more smiley — it’s absolutely fascinating.

Second point on parenting, which relates to China — when I was growing up, my mother said, “You need to learn golf and Japanese because that’s the lingua franca of business in the ’80s.” Then it became, “You need to learn Mandarin and coding because that’s the lingua franca."

Neither of those things really matter today. The pace of AI development means coding can now be done by AI agents. Computer scientists who thought they were in a valued position are suddenly thinking, “Oh my God, I’m being replaced by agents."

Translation is pretty incredible now too. I don’t know what the next parental trope will be, but it’s usually wrong.

Subscribe now

Jordan Schneider: We can take it back to the machine guns, right? What was successful for you as you were coming up in the world and in the institutions that shaped you isn’t necessarily going to be the thing for the next era. The humility to understand that — both from a defense acquisitions perspective and a parenting perspective — is really hard.

Josh Wolfe: I’ll give you an investing version of that. First, the most dangerous words everyone always says are “this time is different” — because it never is. If you know Shakespeare, then of course it’s never different. Human nature is constant.

But when parents utter certain words, it’s predictive — similar to the defense acquisitions issue or for investing. Want to know what will be the next $10 billion industry? Here it is: “It will rot your brain.” Every time a parent says “it will rot your brain” about something they don’t want their kids doing, that thing becomes the next massive industry.

Rock and roll in the ’50s, TV in the ’60s and ’70s, chat rooms in the ’80s and ’90s, video games in the ’90s and 2000s — every one of those things that was the target of parental ire became the next $10 billion industry. Tipper Gore with parental advisory lyrics and rap music — rap became the biggest genre of music in the following decade.

Just listen to what parents are terrified about right now. The gamers became our modern robotic surgeons and drone pilots. Whatever they’re freaking out about now — maybe TikTok (though I have problems with that for different reasons) — it could be movie-making on social media or whatever.

Josh tries to defend short video

Jordan Schneider: This is the hardest question of the day — give me the optimistic short video take.

Josh Wolfe: We’ve democratized the ability to have creative expression with special effects that used to cost $50 million. Filmmakers used to be siloed in studios with hierarchies, casting couches, and Harvey Weinsteins — awful people. Now there’s freedom of expression where people can create tragedy, drama, comedy, and surrealist content with these tools at their fingertips.

My 9- and 12-year-olds are better filmmakers than I was at 25. They have tools today that Hollywood executives used for Terminator or The Abyss — remember James Cameron’s special effects that seemed amazing back then? That’s from a creator perspective.

Jordan Schneider: What about from a consumption perspective? Having culture delivered in 30-second chunks?

Josh Wolfe: Again, look at it from abundance and scarcity perspectives. I 100% agree that if you’re constantly being trained for short attention spans, that’s problematic. We literally practice patience as a family because I know there are so many competing things offering fast dopamine hits and quick responses. We do long periods of quiet reading from physical books as a family. We watch long movies instead of 30-minute segments.

Jordan Schneider: No, but you’re telling me the things parents are scared of are actually going to be the future. What’s positive about consuming content in 30 seconds?

Josh Wolfe: I’d argue that your ability to process multimodal information is far better. Look at the average older person right now — they’re focused on one thing, they’re slow. You’re probably able to switch between a WhatsApp chat group, Twitter, watching a short video, and checking emails. Your ability to multitask while retaining the ability to function is super valuable.

Jordan Schneider: Your heart’s not in this one…

Josh Wolfe: It’s okay, but here’s what I know: it’s not as bad as people think. The TikTok stuff is concerning, but short-form video generally isn’t bad.

Jordan Schneider: Okay.

Josh Wolfe: I’m optimistic on the science and technology piece. I’ll still be skeptical about the human nature piece, but optimistic about science and technology.

Jordan Schneider: All right, let’s call it there then. This was really fun. Do you have a book to shout out?

Josh Wolfe: Let’s close on fiction and nonfiction. For nonfiction: The biggest debate my wife and I have had was over a book by Robert Sapolsky, a Stanford primatologist and neuroscientist who’s written a series of books. His first one was Why Zebras Don't Get Ulcers: The Acclaimed Guide to Stress, Stress-Related Diseases, and Coping. The punchline was that zebras run for their lives in their ancestral environment, then they’re calm. We have constant stressors all day long that didn’t exist in our ancestral environment.

But his more recent book is called Determined. Either you’ll want to throw it across the room, or you’ll want to send copies to everybody. It depends on whether you agree — as I do — that we do not have free will, or disagree — as my wife does — that we are filled with agency and of course have free will.

Danny Kahneman was a friend. His belief before he died was that neither free will nor consciousness exist — that they are both useful illusions. I very much subscribe to that view.

For fiction, Amor Towles has a short story collection called Table for Two. He wrote Rules of Civility, which is great. There’s a character from the second half of that book who gets extended treatment in a few more chapters in this collection.

But there’s one story that deeply touched me called “The Bootlegger.” It’s a short story that takes place in New York. The beauty of this, for me personally, was that I happened to post on Twitter about my love for this particular story. Amor Towles replies, which was very meaningful, and he says, “This is probably the most autobiographical story I ever wrote."

It’s a relatively short story set here in New York. A young couple, something happens, they go to Carnegie Hall, and this story unfolds. There’s a particular classical musician referenced in this fictional piece, and he replies to the story because he’s actually in it. It was this surreal fictional story that was a slight roman à clef of Towles’s life. He replies to me on Twitter, then the classical musician does too. I was like, “This is amazing."

Jordan Schneider: That’s really fun.

Josh Wolfe: I highly recommend Table for Two by Amor Towles. It’s a set of vignettes around two characters — probably 10 or 12 short stories. But “The Bootlegger” is awesome.

Jordan Schneider: Thanks so much.

Josh Wolfe: Great to be with you.

In 2011, China’s drug regulator cleared the nation’s first home-grown targeted cancer pill. Fourteen years later, a Chinese bispecific antibody is aiming to knock the world’s top-selling oncology drug off its perch.

Those two bookends frame twin contests now running in parallel. One is humanity’s decades-long fight against cancer, a disease that still claims one in six lives worldwide. The other is China’s effort to move from importing medicines to inventing them. This article seeks to follow these battles in real time by tracing the stories of four milestone therapies and their makers.

Founded in 2003 in Hangzhou by a team of returnee talent, Betta Pharmaceuticals launched the first targeted anticancer drug developed in China to reach the market. The success of its drug Conmana1 in China dovetailed with sweeping policy reforms aimed at making home-grown medicines more trusted, innovative, and affordable. The drug did not, however, make it outside China, despite Betta Pharma’s best efforts. Conmana is an example of “me too” innovation: a variation on an existing drug that performs just as well but not quite better enough to make it globally competitive.

Created through a rigorous internal drug development program in 2012, BeiGene’s Brukinsa did what Betta’s Conmana could not – it went global, becoming the first Chinese cancer therapy approved by the US FDA. A “me better” innovation targeting blood cancer, Brukinsa now reaches people in over 65 countries, bringing in over US$2 billion in sales annually.

Betta’s Conmana and BeiGene’s Brukinsa are both small-molecule drugs, meaning they are created through chemical synthesis. The real frontier of innovation that excites biotechnologists is biologic drugs, biomolecules such as engineered proteins or RNAs that act with greater power and precision inside the body.

Carvykti is one of the first innovative biologic drugs created in China. Initially discovered and tested in Xi’an Jiaotong Hospital by Nanjing Legend Biotech, the drug reached global markets with the help of Johnson & Johnson. First gaining US FDA approval in 2022, Carvykti now has regulatory approval across over 36 countries and has treated over 5,000 patients, with more to come.

The last story is unfinished. It’s about Akeso’s ivonescimab, an icon of Chinese biotech innovation in mainstream media. A biologic drug with a novel method of targeting cancer, ivonescimab received approval by the Chinese NMPA in April 2025. In countries like the US, where Akeso has passed the baton to US-based Summit Therapeutics to develop the drug, ivonescimab is still in the clinical trial phase, meaning it has yet to pass through the regulatory gauntlet. What’s exciting and undecided about ivonescimab is its potential to go head-to-head with the world’s best-selling drug, Keytruda. Whatever happens to ivonescimab over the next few months in trials outside of China will send a signal of exactly how successful Chinese biotech innovation has become. The world of biotech will be watching closely.

But let’s start from the beginning. Betta Pharma’s milestone achievement of Chinese regulatory approval for a new innovative drug (2011) happened only 11 years before Akeso’s ivonescimab earned a US$5 billion deal with Summit Therapeutics (2022). How did we get here?

Betta Pharma: The “Me-too” Era

In the 1990s, biopharmaceuticals — specifically oncology, the study and treatment of cancer — entered a new era of innovation. Up until the early 2000s, doctors primarily combated cancer with broad-stroke methods like surgery and chemotherapy. Now, new and improved methods were emerging: targeted therapies that zeroed in on cancer cells while minimizing damage to healthy cells and immunotherapies that helped the body’s immune system recognize and attack cancer cells.2

This revolutionary effect of molecular biology captured the attention of many bright scientists and doctors in China, including Dr. Wang Yinxiang 王印祥. Born in rural Hubei, he spent three years working in public health and three years completing a Master’s degree at the Chinese Academy of Medicine before he could truly follow his passion for oncology to the United States, where he earned a doctorate from the University of Arkansas.

Dr. Wang got his wish to do cutting-edge research as a postdoc at Yale, where he dove into one of the first targeted cancer therapies, Novartis’ Gleevec. Sharing his apartment was Ding Lieming 丁列明 – another Chinese transplant with a University of Arkansas MD. On strolls through New Haven’s Science Park, the two friends along with medical chemist and entrepreneur Zhang Xiaodong 张晓东 bonded over more than just science. They shared a bigger dream: to bring the newest in biotech to China.

Dr. Wang and Dr. Ding would ultimately join forces in 2003, when they founded Betta Pharmaceuticals to develop targeted cancer therapies in China. Betta opened its doors with a shoestring team – just 13 people, many of them what Dr. Wang affectionately called “kids,” fresh from bachelor’s or master’s program and learning on the fly. Nevertheless, they managed to develop icotinib (later sold as Conmana), a drug engineered to target EGFR proteins as a way to inhibit cancer cell growth.

In those years, China’s pharmaceutical industry hadn’t left the nest. Manufacturing of cheap, generic drugs dominated. To domestic investors, companies, and physicians, a business model built on new drug development was unthinkable: the costs and risks were too high, the regulatory process was a mess, quality and safety were still iffy, and previous such attempts had failed. Winning a clinical-trial slot for Conmana (a prerequisite for proving the drug could outshine current care) was nothing short of herculean. When the Peking Union Hospital director declared the study too risky and tried to dismiss him, Dr. Wang stood firm for ninety minutes, knocking down every objection until the approval stamp finally came down in his favor.

In 2009, Conmana made it to Phase III clinical trials, the make-or-break test of wide-scale efficacy. In Phase III, the team pushed the envelope again: rather than testing against a placebo, they pitted their molecule against AstraZeneca’s gefitinib (the world’s first targeted anti-cancer therapy) in the first Chinese study to challenge an imported standard head-to-head.

The study’s results, announced by leading academic Sun Yan at the 2011 World Lung Cancer Conference – also the first time a China-developed drug headlined an international academic forum – showed that icotinib could match the cancer-fighting power of the imported benchmark while causing fewer side -effects. Conmana, in other words, was a successful “me-too” drug, an incremental improvement on an existing pharmaceutical innovation.

After six grueling years, Conmana earned its first regulatory approval from China’s National Medical Products Administration (NMPA; although at that time it was still the CFDA).

Betta’s success as a fast follower of a next-generation cancer therapy was a triumph for Beijing’s returnee talent and national innovation programs. Conmana’s success had been fueled by funds from the Yuhang District Government of Hangzhou, the “863” Program, and the “11th Five-Year Plan” National Science and Technology Major New Drug Special Project.

The government showered Betta with accolades: the China Overseas Chinese Contribution Award, the gold prize for patents, first prize for the National Science and Technology Progress Award, and more. Chen Zhu, then Minister of Health, praised their achievement as “an emblem of ‘Two Bombs and One Satellite 两弹一星’ in the field of public health,” referencing a techno-nationalist ideal of a whole-nation project for science and technology development.3

However, for all its homegrown glory, Betta’s blockbuster never crossed the border.

In 2014, with the help of Roswell Park Cancer Institute, Betta filed to run a clinical trial for Conmana in the US, the first step towards seeking US FDA approval. However, the trial was shortly withdrawn. Newer, later-generation EGFR inhibitors were eclipsing Conmana’s performance, and there was no sense investing in trials when the product was unlikely to sell.

Though just a “me-too” innovation, Conmana was a landmark accomplishment for China. On the domestic market, the drug was meaningfully cheaper than its imported alternatives. Given the rapidly growing incidence of lung cancer in China, Conmana’s improved affordability made a real difference in patients’ lives. Still, it would take almost another decade before a China-developed cancer drug would make a global impact.

BeiGene: A “Me-Better” Drug Goes Global

China’s mix of capital, talent, and policy reforms was turning its budding biopharma sector into a global magnet.

Among the first drawn in was Pittsburgh native John Oyler. Familiar with China through his work at McKinsey in the 1990s, Oyler was stunned by the country’s science and technology progress when he returned the next decade. By the mid-2000s, regulatory harmonization, returning talent, and improved manufacturing infrastructure enabled China to meet the needs of global pharmaceutical companies, leading to the growth of contract research organizations (CROs), which provide outsourced medicinal science services. Oyler co-founded one such Chinese CRO, BioDuro, in 2005.

But serving foreign pharma clients wasn’t enough. As China moved toward deeper healthcare reform, Oyler saw an opening for homegrown innovation: “[China] had the capability to pour tens of billions of dollars back into the global industry to help pay for more research, which would not only make drugs more affordable in China, but across the globe.” Rather than repeat the trajectory of BioDuro, which was eventually sold, he wanted to create something enduring. “I wanted to build something here — in China — that is lasting, impactful, involved in great science, and can really help people,” he said. He aspired to create a company capable of developing world-class cancer therapies from a country many still underestimated.

To bring his vision to reality, Oyler needed scientists. He connected with Dr. Xiaodong Wang 王晓东, a top Chinese American academic biologist who recently returned to China to lead the new Beijing Institute of Life Sciences. The chance to work with Dr. Wang — a superstar scientist admired widely enough to impress parents at Chinese New Year — proved an effective tool for attracting talent. Together, they founded BeiGene in Beijing in 2010 to become the “Genentech of China.”

Early on, BeiGene focused on BTK inhibitors, a targeted cancer therapy that works by blocking cancerous B-cells’ ability to grow. The first BTK inhibitor, synthesized in 2007, showed promise but caused significant side effects. In 2012, BeiGene initiated a discovery program in San Mateo and Shanghai to develop a better BTK inhibitor. After screening over 3,000 compounds, the team identified the highest-potential molecule that would eventually become Brukinsa (zanubrutinib).

BeiGene aimed to take Brukinsa global from day one. To support worldwide approvals, the company built a 25-country trial program in which approximately 90% of patients were enrolled outside of China. The numbers delivered: Brukinsa consistently beat first-generation BTK inhibitors on safety and efficacy, turning a presumed “me-too” into a clear “me-better” that is now the standard of care for B-cell cancers.

Momentum snowballed. In 2019, Brukinsa set a precedent as the first Chinese-developed cancer therapy to win FDA approval, months before China’s own NMPA signed off. It has since secured clearances in 65-plus markets spanning the US, EU, Canada, Australia, Japan, and China, and now supplies more than half of BeiGene’s revenue with US$2.6 billion in 2024 sales. The company (rebranding as BeOne) has likewise gone global, conducting trials in over 45 countries.

Legend: A True Chinese Biologic

Witnessing the breakthrough of firms like BeiGene, Beijing set its sights on higher-value innovation. The State Council’s 2016 13th Five-Year Plan therefore called for “leapfrog development in the biopharmaceutical industry,” spotlighting cell and gene therapies, antibodies, and vaccines.

These platforms fall under biologics — large molecules derived from biological processes such as insulin and hormones — rather than small-molecule drugs such as Betta’s Conmana and BeiGene’s Brukinsa. Because biologics are bigger and more structurally complex, their effects are harder to predict and their manufacture far costlier, but they open therapeutic doors that chemistry alone cannot.

Founded in 2014 as a subsidiary of GenScript, Legend Biotech embodied the kind of biologics leadership the state now prioritized.4 After losing his father to cancer, co-founder Frank Zhang 章方良 united with Chief Scientific Officer Dr. Frank Fan to create the firm with the goal of advancing oncology.5

At that time, influential scientific journals, Big Pharma deals, and first-in-human successes had converged to position cell and gene therapies as the vanguard of biopharmaceutical innovation. Early successes led by the University of Pennsylvania highlighted the potential of CAR-T therapy,6 a type of treatment in which a patient’s disease-fighting T-cells are genetically engineered to seek and destroy cancerous cells.

Driven by the promise of next-gen cancer therapy, Legend’s 19-person team, working in “a room the size of a freight elevator,” crafted a second-generation CAR-T treatment targeting multiple-myeloma tumor cells (later sold as Carvykti). Leading the research was Dr. Frank Fan, who had studied at Xi’an Jiaotong University and worked at the Xi’an Jiaotong Hospital. Leveraging those ties, Dr. Fan was quickly able to initiate Legend’s first CAR-T clinical trial at Xi’an Jiaotong Hospital, turning the startup into a “dark horse” contender in the CAR-T space.

At the 2017 annual meeting of the American Society of Clinical Oncology (ASCO), Legend presented early-phase data from its trials in China: durable remissions in multiple-myeloma patients with only mild side effects. The first of its kind accepted for review by China’s NMPA, this CAR-T candidate signaled that the country could innovate well beyond small-molecule chemistry.

Such promising results drew members of the global life sciences community. J&J’s Janssen signed a partnership with Legend Biotech in 2017. This was a case of an outlicensing deal: when a company (such as Legend) sells or grants rights to a drug candidate, passing the baton to a different company (such as Janssen), which who takes on the responsibility to bring the drug through testing, approval, manufacturing, and commercialization. Such transactions are the lifeblood of the industry. However, the high-potential molecules worth such high-profile deals didn’t historically come from China.

With the combined efforts of Legend and Janssen, Carvykti won FDA approval in 2022, soon followed by clearances in the EU, UK, Japan, and Canada. Though not first-to-market, Carvykti’s superior clinical value crowned it best-in-class. This then completed China’s rapid climb from “me-too” copies, through Brukinsa’s “me-better” gains, to a world-leading biologic breakthrough.

Akeso: a top challenger emerges

Now, Akeso Biopharma’s new molecule is drawing notice as a likely first- and best-in-class therapy from China.

Akeso started as the dream of Dr. Michelle Xia, a Gansu native. While working for California-based Crown Bioscience and in other roles in the US and UK, Dr. Xia grew frustrated with the eight-to-ten-year delay it took for innovative overseas therapies to reach Chinese patients. So she and three partners founded Akeso in 2012, naming it after the Greek goddess of healing, with a mission to develop home-grown therapies for cancer and autoimmune diseases.

Akeso’s edge is bispecific antibodies (BsAb), a type of next-generation cancer treatment involving engineered proteins that strike two targets at once, such as igniting immune cells while starving tumors. The first proof arrived in 2022 when China’s NMPA cleared Akeso’s Kaitani (PD-1/CTLA-4), the world’s first commercialized BsAb.

Next came ivonescimab, a PD-1/VEGF bispecific now pushing Akeso onto the global stage. By jointly blocking an immune checkpoint and tumor blood-vessel growth, it qualifies as first-in-class – an industry term for a drug that introduces a truly novel therapeutic approach.

In June 2022, Akeso unveiled the Phase II results of ivonescimab at the annual ASCO conference, showing strong responses in non-small cell lung cancer. Just months later China’s NMPA granted the drug Breakthrough Therapy status for three medical use cases, enabling closer guidance and fast-tracking its review.

Sensing ivonescimab’s scientific and commercial potential, US-based Summit Therapeutics inked a massive deal in December, licensing Akeso’s innovation for up to US$5 billion.7 The corresponding press release hailed Akeso’s innovation as “the PD-1 / VEGF bispecific antibody that is most advanced in the clinic,” noting that neither the FDA nor EMA had yet approved any PD-1-based bispecific therapy.

The size and significance of this agreement marked a bellwether moment in Chinese biotech innovation. Blue-chip investors and multinationals began scouting the country for genuinely novel assets rather than low-cost manufacturing plays.

China’s slice of global out-licensing has since tripled to 12%, with deal value leaping from US$35 billion in 2023 to more than US$46 billion in 2024. This trend seems set to continue through 2025 and beyond.

The fate of ivonescimab and most other compounds covered by these recent deals is uncertain. Many drug candidates are purchased in the preclinical or Phase I stage of development, requiring another 5+ years and US$300+ million dollars before they pass the clinical and regulatory hurdles necessary to make it to market – or, they will fail, like roughly 90% of compounds that enter human trials. Only five China-originated drugs have ever cleared the US FDA (BeiGene’s Brukinsa and Legend’s Carvykti are two of them).

Ivonescimab could be next. Akeso’s molecule has already completed Phase I and Phase II, and is in the midst of several high-stakes Phase III trials. Its Phase III trials with Chinese patients have already demonstrated success, leading to two recent NMPA approvals in the spring of 2025. The defining test comes next: a Summit Therapeutics-run Phase III study spanning 108 locations in 12 nations, where the drug candidate must outshine oncology’s gold standard, Keytruda.

Keytruda (drug name pembrolizumab) has been described as “era-defining,” the “800-pound gorilla” of the class of drugs to which ivonescimab also belongs. Developed and commercialized by American multinational Merck, the drug has been approved for 41 indications8 across 18 types of cancer. It’s the world’s best-selling drug, raking in about $29.5B in 2024 – nearly half of Merck’s total revenue. Its upcoming 2028 patent expiry opens the field for new challengers.

Akeso’s ivonescimab could be one such challenger. If its global Phase III trial confirms that the drug’s positive risk-reward results can extend beyond China’s borders, Akeso’s first-in-class molecule may eventually also prove to be best-in-class, making it one of the biggest biotech stories of the decade.

On the horizon

Together, Conmana, Brukinsa, Carvykti, and ivonescimab trace a clear, but incomplete, arc: China’s pharma sector has evolved from reverse-engineering proven ideas to originating drugs that can contend for global standards. Each milestone marks a step from “good enough at home” to “competitive abroad,” showing how policy shifts, capital inflows, and returning talent have reshaped the industry’s ambitions and capabilities.

Those four successes are only a sliver of the story. Dozens of other firms have logged incremental wins, and many more have stumbled in clinical trials or overseas filings. With financing tightening, patent cliffs approaching, and regulatory expectations rising, the next crop of candidates will test whether China’s momentum is structural or situational. The outlook could range from steady gains in select niches to a broader slowdown if capital or policy tailwinds fade.

Whatever the trajectory, one fact persists: cancer is indifferent to where a molecule is conceived. Progress depends on tapping every credible lab and idea, whether in Boston, Basel, or Beijing. If Chinese innovators add new options to the world’s oncology toolkit, patients everywhere stand to benefit — and that, ultimately, is the benchmark that matters.

2

See this chart:

6

See this chart:

Once hailed as the “Apple of China,” Xiaomi has evolved into the world’s No. 3 smartphone brand. The Beijing giant wires millions of homes with its TVs, wearables, and appliances while pitching itself as a seamless “Human × Car × Home” platform.

Founder Lei Jun now wants to turn that consumer-electronics prowess into hard-tech dominance. In 2025 alone, Xiaomi debuted its first in-house chip (the XRING O1) and showed off its sleek SU7 electric sedan. Yet the scorecard is mixed: the Robotics Lab has been folded into the auto unit, the SU7’s shine dimmed after a fatal crash, and new U.S. export curbs threaten the flagship chip program just as it launches. Can a smartphone playbook built on rapid iterations and razor-thin margins really scale to cars, robots, and semiconductors?

In this piece, we let you hear Lei Jun in his own words, through two translated excerpts. One brims with sky-high ambition, while the other is steeped in damage control. Read them back-to-back and watch how Xiaomi’s grand vision collides with hard truths of physics, finance, and public accountability.

From budget phones to bold hardware

Chinese media celebrate Lei Jun as one of the country’s signature tech entrepreneurs. After graduating from Wuhan University and weathering several early start-up misfires, he assembled seven co-founders in 2010 around a crowdsourced Android ROM that grew into Xiaomi. The early strategy was straightforward: sell affordable, capable smartphones. That formula made Xiaomi a household name in phones and Internet of Things gadgets, but the 2016 debut of the bezel-less Mi MIX concept signaled a pivot from “internet hardware” to true hard tech. In the translated excerpts below, Lei Jun traces that evolution and lays out why he believes Xiaomi’s future rests on chips, cars, and robots.

The following excerpt is an original ChinaTalk translation of a December 2023 interview between Lei Jun and reporter Wang Ning, broadcast on CCTV.

Original link | archived

Lei Jun: Let me start with some Wuhan University memories. Thirty-six years ago, in 1987, I was admitted to its Computer Science department. Back then I lived almost entirely on scholarships and professors’ projects; those funds literally carried me through four years of college. At one award ceremony I received a sizable scholarship; standing on that stage I vowed that if I ever could, I would repay the university a hundred, a thousand — ten-thousand — fold.

Lei Jun: My life-long dream was sparked in the Wuhan University library when I read Fire in the Valley, the story of Steve Jobs and early Silicon Valley. In the late ’80s, Jobs was my very first role model. But the more I understood Jobs, the clearer it became that I’m not him. Still, that doesn’t stop us from charting our own path and creating products that are different in their own right.

Angela: After graduating in 1991, Lei joined Kingsoft, a Chinese software company, where he progressed from software engineer to executive. Competing with Microsoft, Kingsoft faced countless problems including product flops and near bankruptcy. He then went on to found an online bookstore, Joyo.com, which was sold to Amazon in 2004.

Lei’s experiences with such missteps and missed timing during the early era of the internet led to his coining of the phrase, “Catch the right tail-wind and even a pig will fly” 站在风口上，猪也能飞起来, which he proceeds to explain in the interview.

Reporter Wang Ning: Your famous tail-wind principle is still quoted all the time.

Lei Jun: Back in my engineer-founder days we green-lit projects simply because they interested me or a few users asked for them; we rarely stopped to ask whether the timing offered explosive growth, so the companies survived but seldom scaled. That sort of company is hard to kill, but pushing it to greatness is equally hard.

That’s why I formulated the tail-wind principle. Don’t just keep your head down pulling the cart; look up and see where the wind is blowing. It was a major course-correction in my own playbook.

Reporter Wang Ning: What was the real test for you during that period?

Lei Jun: We had clearly missed the web’s prime window, so we kept asking when the next breakout would come; four or five years early we bet on mobile internet, but didn’t know when it would ignite—once the iPhone and Android hit, I knew the fuse was lit.

On Xiaomi’s “hard tech” pivot:

Lei Jun: Hard tech means chips, smart manufacturing, robotics, operating systems — those core layers.

Reporter Wang Ning: You keep hammering on hard tech.

Lei Jun: I do.

Reporter Wang Ning: Why is your drive to battle it out on hard tech so intense?

Lei Jun: We’re top-three worldwide, but we face giants: Apple, Samsung, Huawei. Without breakthroughs in core tech you’ll never build a moat or stand shoulder-to-shoulder with those titans. If Xiaomi dreams of being world-class, hard tech is non-negotiable.

Narrator: In 2020, Xiaomi’s tenth anniversary, it set a ten-year goal to become a next-generation hard-tech leader and deliver premium products.

Lei Jun: Every flagship product applies the newest, most advanced tech. Three years ago, I told the team we would benchmark against the iPhone across the board. This statement caused controversy. Were we just piggy-backing off of Apple’s fame to grab attention? But we must unsheathe the sword: set our own targets, dare to compare. Even if we lose all 100 metrics, daring to measure is step one. Can our phones really win? I had to persuade people over one-by-one that going premium takes patience, since earning user trust is a process.

Lei Jun: We started exploring operating systems from the very first days of the company. Seven years ago we resolved to create Xiaomi HyperOS, building the entire stack from the kernel up. Over 5,000 engineers have contributed, and we’ve just released the first version. Our goal is a human-centred, closed-loop platform that connects personal devices, the smart home, and the car.

Reporter Wang Ning: You’re working down at the OS layer — deep, technical stuff — yet you keep calling it ‘human-centred.’ What exactly does that mean in practice?

Lei Jun: AI is moving fast, so we’re baking the most advanced AI into HyperOS to give it a clear edge in intelligence over any rival platform. More importantly, that same intelligence has to knit together the entire “Human × Car × Home” ecosystem: people as in personal devices, cars as in a mobile smart space, and home as in smart appliances.

On electric vehicle strategy:

Reporter Wang Ning: Elon Musk called Xiaomi’s entry into the automotive industry “interesting competition.” Your response?

Lei Jun: Smart EVs merge auto and consumer electronics. One car CEO joked, “A smart EV is just an oversized smartphone on four wheels.” Not strictly true, but that shows the convergence. So yes, it’s challenging, but the difficulty is controllable.

Reporter Wang Ning: What do you mean by “difficulty is controllable”?

Lei Jun: Three years ago I still thought building a car was daunting, so I approached it with real humility. Our user research showed that when people hear ‘Xiaomi car’ they expect technology and an ecosystem. That led me to a guiding principle I call “nail the fundamentals, then amaze 守正出奇.” In practice it means: respect the hard rules of the auto business — quality, safety, manufacturing discipline, make sure the very first model is rock-solid, and only then layer on top the Xiaomi-style innovation that surprises the market.

Our second rule is ‘10-X investment.’ A typical automaker puts three to four hundred engineers and maybe one or two billion yuan into a new model; many cars you see are built on that. For our first car we assigned 3,400 engineers and spent more than ten billion yuan — over ten times the norm. With that level of commitment, I’m aiming to win.

Still, cars are complex. I worry about two opposite risks: the launch flops and nobody buys, or demand goes crazy and people wait a year and flame us for delays. Either way, there’s plenty of reasons for anxiety.

On Xiaomi's future amidst geopolitical uncertainty:

Reporter Wang Ning: With global turbulence and domestic headwinds, why double down on hard tech and stay confident?

Lei Jun: The tougher the climate, the more we should invest in technology. When others pull back, breakthroughs actually come easier, and we’re ready the moment the market rebounds. In fact, Xiaomi already returned to positive growth last quarter.

Why do we have the nerve to keep spending? Because in every field we play in there is still plenty of unmet demand; I believe Xiaomi is only at the starting line. Our approach is two-pronged: deepen our roots in China’s domestic market while pushing ahead with globalization. That’s the road we’ve chosen, and it’s still rich with opportunity.

Reporter Wang Ning: People say you don’t need to change since things are great. Yet you keep leaving your comfort zone. Why?

Lei Jun: At the core is the very high bar we’ve set for ourselves: to become a great technology company that lets everyone on the planet enjoy the benefits of innovation. The goal is so lofty that no matter how hard we jump, we still can’t quite reach it — yet that stretch is exactly what pulls the whole company forward. Xiaomi’s relentless evolution is powered by that dream.

Hard Tech, Hard Lessons

[Angela writing] 2021 marked Xiaomi’s leap into one dimension of hard tech, robotics, with the release of CyberDog, an open-source robot companion. A year later came CyberOne, a full-sized humanoid, and in 2023 the sleeker CyberDog 2. Commercially and technically, none hit the mark: CyberOne never reached mass production, and CyberDog 2 sales ran at a loss as buyers complained about its limited abilities. Attrition was another problem: high-profile engineers departed, with Liu Fang 刘方 (former autonomous-driving head) leaving to start his own robot firm and humanoid specialist Ren Zeyu 任赜宇 moving to ByteDance. By mid-2024 the Robotics Lab had been folded into Xiaomi’s auto division, its once-lofty ambitions reduced to building robots that service the company’s own factory lines.

The automotive story began the same year the robots debuted. In early 2021, Lei Jun learned that U.S. sanctions might hit Xiaomi and decided the firm had to diversify beyond phones. Xiaomi poured RMB 10 billion (US $1.5 billion) into an EV program, broke ground on a plant in Beijing’s Economic-Tech Zone, and — true to form — Lei personally test-drove more than 170 cars and earned a racing license, convinced that a top driver should helm a top car company. The gamble paid off fast: the SU7 electric sedan launched in 2024 to a rapturous market reception.

Yet momentum met reality on March 29th, 2025, when a Xiaomi SU7 electric sedan running in Navigate-on-Autopilot mode veered off the Dezhou–Shangrao Expressway in Anhui and slammed into a cement pole, killing three college students. The first fatality involving Xiaomi’s year-old EV led to nationwide uproar. The hashtag #SU7事故 raced onto Weibo’s hot-search list within hours, while Hong Kong–listed Xiaomi shares fell 5.5% the next trading day. The Ministry of Industry and Information Technology banned marketing phrases like “smart driving” or “autonomous driving” and now requires pre-approval for any OTA updates that touch driver-assistance features. The event sparked debate over assisted-driving safety, wiped out new-car orders, and forced the company to publish detailed log data and promise a top-to-bottom safety overhaul.

Xiaomi CEO Lei Jun gave the following speech on May 22nd, 2025, for the company’s 15th anniversary. His address walks a tightrope between contrition and confidence, as he admits to the company’s past mistakes while promising to embrace the societal responsibility that comes with being a major automaker. The crisis shows the high degree of pressure on flagship Chinese tech companies, and how quickly the public, investors, and policymakers can converge when such a firm stumbles.

At the same time, Lei Jun widens the narrative by announcing Xiaomi’s newest chip, the XRING O1. On June 3rd, he predicted Xiaomi’s auto arm would turn a profit in Q2 2025. But will the company be able to live up to its bold ambitions?

The following excerpt is an original ChinaTalk translation of a speech given by Lei Jun on May 22, 2025. Transcript from Zhihu.

This year marks the 15th anniversary of Xiaomi’s founding. Today is a momentous day, and we prepared a series of celebratory events for it. However, at the end of March, a sudden traffic accident shattered all of that. We were hit by a storm of doubts, criticism, and blame. My colleagues and I were stunned.

A veteran of the auto industry told me, “In this business, accidents are bound to happen.” But no one expected the impact of this accident to be so massive, nor the blow to Xiaomi so heavy.

Thinking back to four years ago when we decided to build cars, I was especially worried about safety, so we placed enormous emphasis on vehicle quality and safety. After more than three years of relentless work by our automotive team, the SU7 has been on the road for just over a year — and its build quality has been our pride. In every independent test by the leading authorities, it has received top marks. Yet we never could have foreseen how this traffic accident would make us realize the public’s expectations and demands of us.

To be honest, only now do I recognize that we have always thought of ourselves as newcomers to the auto industry. This incident drove home a simple truth: Xiaomi’s scale, influence, and visibility have grown so much that society now expects us to act like a fully fledged industry leader. We understand, deeply, that after fifteen years this is a vital responsibility we cannot avoid.

So today, what I want to share with you is that 15-year-old Xiaomi is no longer a rookie. In all industries, we don’t have the grace period of true novices. We must hold ourselves to higher standards and goals.

On automotive safety, I want to announce to everyone that we aim to make our car the safest in its class. We will not just comply, nor merely meet the industry level; as a leader in the auto industry, we will guarantee safety that surpasses the industry standard.

This year on April 1st, I said publicly on Weibo that Xiaomi will never shy away from any issue. “Never shy away” means confronting problems head-on, examining ourselves critically, fixing what’s wrong, and committing to continuous improvement. I know accomplishing this is extremely difficult; it requires us to unite in heart and mind and take it with the utmost seriousness.

Over the past few months, I have held countless meetings with the automotive department’s management and team. All of these meetings have centered around one core theme: how do we solve these problems systematically? How do we convincingly show, through stronger operations and governance, that we’re living up to the public’s higher expectations?

This year is Xiaomi’s 15th anniversary. We have canceled many of the celebrations, summaries, and planning activities we had planned. In any case, I believe this is an opportunity for us to seriously review the wins and losses of the past five years.

In my view, the most important thing Xiaomi has done in the past five years is maintaining a strong technology foundation. Five years ago we set a new goal: to become a global leader in next-generation hard tech. Five years ago we made a clear commitment to invest more than 100 billion yuan in R&D over five years and to increase core technology research; to date we have invested about 105 billion. This year alone we expect to invest over 30 billion.

Today, here, I want to share an extremely important piece of news with you: our self-developed smartphone SoC chip, the Xuanjie 玄戒 O1, is expected to be released at the end of the month

After 10 years of making chips, this is Xiaomi’s milestone achievement. It also represents a new starting point for Xiaomi to break through in hard tech. Chipmaking is something the public and Xiaomi fans ardently expect from us; it is also the only path for Xiaomi to move toward becoming a hard-core tech leader. Xiaomi will forge ahead fearlessly.

2019 was a challenging period for us. We faced all kinds of internal and external pressures. At the suppliers’ conference at the end of that year, I told our supply-chain partners this line: “A gale reveals the toughest grass; a long road proves a horse’s strength.” “疾风知劲草，路遥知马力。[Editor’s note: The metaphorical meaning is along the lines of “true resilience is exposed only under pressure and over time.”]. I believe that from then till now, all our partners can clearly see that Xiaomi is much stronger than it was five years ago. As of today, Xiaomi has existed for 15 years. Years of highs, lows, and hard times have already proved just how resilient we are.

When we started, I remember thinking that people would need 15 years to understand and recognize Xiaomi. Today, it seems we were too optimistic — 15 years is not enough. But that’s all right; we will just keep moving forward until the day we have fully proven ourselves.

Xiaomi still has a long way to go before it becomes the strongest, but no one can match the persistence, resilience, and patience that keeps us getting back up every single time.

Xiaomi’s future

[Angela writing again] Will Lei Jun’s everywhere-everything strategy work?

Xiaomi’s smartphone success hinged on a strategy tailored to consumer electronics: rapid iteration, razor-thin margins. Hard tech plays by stricter rules. Robots, cars, and advanced chips demand flawless safety engineering, deep benches of specialist talent, and sustained capital infusions.

Xiaomi has already brushed up against each of these constraints. In the Robotics Lab, key engineers were stretched thin, while the first SU7 sedan shipped with a leaner, lidar-free sensor suite, a cost-saving choice that came under scrutiny after a fatal crash raised questions about perception margins. Complicating matters is the sheer sweep of the company’s ambition: pursuing cars, robots, and advanced chips threatens to disperse capital and engineering bandwidth, potentially leaving every moonshot a little short of the sustained focus that hard tech demands.

Even Apple doesn’t juggle cars, robots, and semiconductors all at once. The Silicon Valley behemoth shelved its car project after billions of R&D and keeps its robotics work firmly behind the curtain. For Xiaomi’s robotics and autos work, leaning into splashy launches and viral publicity for its hard tech projects has proven to be as much a liability as a strength.

Chips might be the exception, if Xiaomi can leverage its vast smartphone footprint into volume for its in-house silicon. The company’s newly announced Xring O1 and US$28 billion R&D pledge hints at genuine momentum. Yet geopolitics looms large: fresh US export curbs on electronic design automation software threaten to inflate costs, stretch design cycles, and slow the very updates those next-generation chips will depend on. Whether Lei Jun can temper showmanship with staying power will decide if Xiaomi’s hard-tech dream becomes a cornerstone of its ecosystem or just another chapter in China’s moonshot boom-and-bust cycle.

Mood Music

I did a podcast with Doug, Dylan and Jon updating our AI Mandate of Heaven tier list and talking about how America needs science research and immigrants (Apple Podcasts, Spotify). If you’re an AI researcher interested in signing onto a short letter in support of the NSF and global STEM talent coming to America, please respond to this email!

Zilan Qian is a fellow at the Oxford China Policy Lab and an MSc student at the Oxford Internet Institute.

In China, U.S.-based large language models like ChatGPT, Claude, or Gemini are technically banned, blocked, or buried under layers of censorship. The Chinese government has only explicitly banned ChatGPT, citing concerns over political content, while other U.S. models like Claude and Gemini are not formally banned but remain inaccessible due to the Great Firewall. U.S. LLM providers also restrict access from China but leave some loopholes: OpenAI blocks API use but Azure continues to serve enterprise clients via offshore data centers; Anthropic blocks access to Claude within China but permits use by Chinese subsidiaries based in supported regions abroad; and Google does not offer the Gemini API in China, but access seems to be still possible via third-parties like Cloudflare (we reached out to Google for a comment but didn’t hear back).

But on Taobao, the country’s largest e-commerce platform, consumers and companies can buy access to these models with just a few clicks. This piece explains how Western models are priced, advertised, bought, and sold in China, and what their popularity reveals about state censorship, platform enforcement, and consumer demand.

Market Overview

How Do Chinese Sellers Price a banned LLM?

Sellers use a range of strategies to appeal to consumers with different priorities — cost, convenience, privacy, and technical needs. Pricing largely depends on the type of access offered — i.e., shared accounts (lower-cost, multi-user access) versus private accounts (higher-cost, single-user access). How users connect to the models also changes the marketing approach. VPN-free access is often marketed with phrases like “domestic direct connection (国内直登)” or “no magic [VPN] needed (无需魔法),” typically referring to proxy sites that offer a simplified, browser-based experience. Other factors influencing price include the model’s subscription tier ( for example, ChatGPT o3 vs. o1), connection stability, and subscription time. Most “domestic connection” or “transferred API” (API purchased through third parties) options are priced significantly below the rates charged by official providers like OpenAI or Anthropic.

For example, the screenshot below shows 27 different Claude pricings, with prices differing based on:

• Subscription time: 1–60 days

• Website access: an account for proxy sites versusan official Anthropic Claude account (with VPN service included)

• Subscription tier: Claude 3.7 Sonnet versus Claude Pro

• API access: “中转 API” (Transfer API) versus “官方API” (API purchased from official anthropic platform)

• Compatibility with SillyTavern, a popular third-party frontend interface (called “酒馆” in Chinese), which enhances user interaction with LLMs.

While an official Claude Pro account costs 199 RMB (27 USD) per month, which is more expensive than Anthropic’s rate at 20 USD, a “direct connection” Claude Pro account from a mirror site is only 65 RMB (9 USD) per month.

Transferred API usually comes from established transfer stations (中转站) for API, usually at a 70-90% discount compared to official prices (e.g., selling 1 USD worth of tokens for 1 RMB). The low prices are made possible by utilizing one or more of the following tactics:

1. Exploiting free API quotas by bulk-registering developer accounts (e.g., Anthropic’s $5 free API);

2. Reselling unused portions of other people’s API quotas;

3. Traffic optimization techniques, such as request aggregation: combining multiple users’ queries into a single batch call to the LLM API to reduce costs

4. Leveraging corporate or educational discounts.

How do Chinese sellers advertise access to U.S. LLMs?

While these tactics help vendors reach potential buyers, they also reveal a surprisingly open market landscape, which raises the question: Who is actually buying and selling these models? Sellers typically advertise and communicate their services using product images that include large brand logos, stylish fonts, and keywords highlighting their service strengths. Direct keyword searches work for most models, like Claude or Gemini. Many listings include the names of specific models directly in their product titles, often alongside a dense string of keywords like “API,” “domestic direct connection,” “official account,” or “exclusive” to boost visibility. The fact that such listings remain easy to find suggests a lax approach to censorship, especially given that many of these terms are potentially sensitive.

The censorship drastically increases when it comes to ChatGPT, which is likely due to the state’s explicit crackdown on this model family, which began in 2023. When selling ChatGPT the model, one mechanism is to market the subscription as a book. As of March 2025, you could still buy “ChatGPT Book: How to Make Money Online with ChatGPT,” “What is GPT Doing…and Why Does It Work,” and “ChatGPT for Beginners — New Book Original English Version,” but end up buying ChatGPT, the AI model. These products were often set at irrationally low prices, usually 1-10 RMB, or 0.1-1.4 USD, to signal that they are not selling a book or the product itself, or rather the initial purchase it just to signal interest. Some also added “2025 new version” or changed “What is GPT Doing” to “What is GPT-4 Doing” make it more obvious. Based on customer feedback, customers pay these low prices as symbolic transactions on the platform and then contact the seller through chat to complete the purchase. The “book sellers” usually displayed large OpenAI and “ChatGPT" graphics in their images and product descriptions.

As of late April, the keyword censorship has caught up and this mechanism is no longer viable. Now, in order to find ChatGPT sellers, you will need to search directly, and only include the version names in your search(i.e., o1, o3, 4.5). The product photos no longer have any indicated association with ChatGPT. No text in the photo mentions “openai” or “chatgpt”, but you do occasionally see the word “奥特曼”, a transliteration of (Sam) “Altman” as well as “Ultraman,” the Japanese TV series. Thanks to “Ultraman,” “Altman” is still uncensored. Interestingly, one seller of o1-pro and o3-mini puts “用AI服务国人” (using AI to serve our compatriots) in their photo, potentially as a fun pun or a patriotic slogan to evade censorship.

Listings involving ChatGPT are often heavily disguised — using transliterations, vague product names, or only having one symbolic price. In many cases, buyers must contact the seller directly to obtain actual pricing details. Note that because keyword censorship and evasion is a rapidly evolving game, the mechanism only applies as of April 2025.

The price lists for a seller selling ChatGPT and other AI products. “4欧普拉斯” reads like “4o+” in Chinese; g开头mask (“g-beginning mask”) refers to Musk’s Grok.

The Diversity of the Buyer and Seller Ecosystem

Among all uncensored models (Gemini, Perplexity, Copilot, Claude, Mistral AI), Claude is the most popular one. The most popular seller has 50k+ transactions with 7k repeated purchases. Based on the review section, many buyers are information technology (IT) or computer science (CS) students who may be attracted to Claude's advanced coding ability, with some sharing how Claude helps them to get into PhD programs or pass their CS classes. Gemini is also popular, allowing buyers to leverage its multimodal capabilities to generate images.

Many of the sellers are seasoned Taobao “账号代充” agents — veterans in the business of topping up game accounts, unlocking premium services at a discount, or helping users bypass regional restrictions. In the past, these agents undercut official prices for Apple games by exploiting refund loopholes or using currency arbitrage with international gift cards. Now, some have turned to the U.S.-based large language models, using the same strategies in addition to building proxy sites, purchasing API through an API transfer station, or disguising domestic models as U.S. LLMs.

Some stores on Taobao are small and low-traffic, listing only one or two LLMs with minimal engagement. But others are well-established vendors with strong reputations, thousands of followers, and over 30,000 repeat purchases. These top sellers often offer a broader portfolio, bundling U.S.-based LLMs with popular AI tools like Pika, Suno, and Midjourney, as well as non-AI software such as Zoom and Overleaf Pro. While many stores seem new to the market, a few have deep roots on the platform. One vendor, active for 14 years, sells Norton 360 antivirus software alongside Gemini and Perplexity. Another, active for seven years, offers only Claude — but does so through product listings disguised as jewelry items, with categories labeled like “rings” and “necklaces.”

Why the Market Survives

The variety and scale of sellers suggest that this market is no small anomaly. So how has it managed to persist, relatively undisturbed, under one of the world’s most tightly-controlled internet regimes?

The State Doesn’t Really Care

The market shows that China's ban is selectively enforced and mainly focuses on ChatGPT rather than all US models. Despite its reputation for having one of the world's most extensive censorship regimes, China’s internet control enforcement is often decentralized, with local authorities and platforms playing major roles. Censorship efforts also tend to prioritize preventing collective action over restricting general information.

LLMs likely do not trigger existential concerns for the state, as they do not inherently expose users to sensitive political content or facilitate mass mobilization in a way that U.S. social media platforms might. Although state media highlights ChatGPT’s alleged spread of “misinformation” about Xinjiang, Taiwan, and the Diaoyu Islands as justification for the ban in 2023, a chatbot’s output depends on user input, which poses a lower perceived threat. The state also may not have enough momentum to strictly enforce bans on all U.S. models after cracking down on ChatGPT symbolically.

However, the existence of a market for US LLMs should not be mistaken for a relaxation of controls. VPN use remains tightly regulated, and the popularity of mirror sites and API proxies reflects efforts to sidestep VPN risks. Buyers frequently emphasize concerns over “stability” and “safety” in their reviews, underscoring that censorship continues to shape this grey market in subtle but persistent ways.

The Platform: New Wine in an Old Bottle (旧瓶装新酒)

Even if the central government doesn’t treat foreign LLMs as an existential threat, what about the platforms themselves? Do companies like Taobao take active steps to self-police? Unlikely. For Taobao, the market for US LLMs is simply “new wine in old bottles”. Even when it comes to far more serious issues — such as the sale of unlicensed drugs or firearms — enforcement on Taobao has historically been patchy at best. Over a decade ago, shady sellers evaded platform controls by tweaking keywords or relying on vague euphemisms to sell guns and drugs. Over the past few years, various random illicit products have slipped through the cracks on Taobao: oversea books and magazines, unofficial “Squid Game” merchandise, and even Ivy League email addresses. The censors played the usual cat-and-mouse game, with Taobao censoring direct keyword searches and related euphemisms, but sellers always found ways to get around.

If Taobao were serious about cracking down on illicit goods, the keyword evasion tactics used by ChatGPT vendors over the years would not be so common. But the platform has little incentive to act aggressively against these sellers unless pushed by the authorities. After all, compared to drugs and weapons, offering access to ChatGPT is relatively harmless.

The Public: the AI hype and the DeepSeek problem

If the state isn't cracking down aggressively and the platform isn’t motivated to act, then the question shifts to the consumer: Why are Chinese users still so eager to pay for U.S. models, especially with powerful domestic alternatives like DeepSeek now widely available?

First, the current market for AI tools in China predates the DeepSeek boom. Long before DeepSeek arrived, the Chinese public had already developed a complicated love-hate relationship with AI. A 2024 study analyzing large-scale Weibo posts found widespread discussion of ChatGPT, often centered on fears of job displacement. These anxieties, amplified by the country’s ongoing unemployment crisis, were seized upon by opportunistic influencers who pushed the idea that people must “master AI or be replaced by it.” Li Yizhou, a serial entrepreneur turned self-styled startup mentor with no AI background, began posting extensively about the technology following ChatGPT’s release in late 2022. His online course on AI raked in over $6.8 million in revenue in 2023 alone, indicating a huge demand for AI in China before much of the rest of the world was talking about it.

Now that DeepSeek is publicly accessible, everyday users still face two major barriers: system overload and excessive censorship. “Server busy” messages have become a frequent frustration, as the model struggles to handle surging traffic after gaining popularity among mainstream users. Technical limitations, including U.S. export controls on advanced chips, further constrain DeepSeek’s computing power. Additionally, user habits formed before DeepSeek’s release have created a certain stickiness to U.S. models. “People who are used to ChatGPT cannot tolerate DeepSeek’s frequent unavailability,” wrote Gui Xingren (硅星人), a popular Chinese tech blog that closely follows the domestic AI scene.

Meanwhile, DeepSeek’s aggressive content moderation poses another challenge. For example, the system refuses any query containing the term “CCP,” even benign ones like “list the economic experts in the CCP.” On RedNote, many users share the frustration of DeepSeek being overly sensitive — many have no clue why DeepSeek would refuse certain topics, including queries about travel plans or career coaching. Users also observed that other Chinese LLMs like Qwen have similar problems.

A screenshot of a RedNote user telling DeepSeek “I finally finished my exam and I missed you,” to which DeepSeek refused to respond.

The issues of censorship and server overload are limited to DeepSeek’s official website and app, and could be mitigated through localized deployment. But for most ordinary users in China, free platforms remain the primary gateway to AI tools. If accessing a smoother experience requires paying for a localized model, such as Tencent’s Yuanbao (which, according to some RedNote users, has less censorship), users may simply choose to pay for access to foreign models on Taobao.

Key Takeaway

China is not a monolith. The existence of this grey market highlights the decentralized nature of censorship, the patchiness of platform enforcement, and the diversity of public interest in AI. So far, Beijing seems either unconcerned with the political implications of U.S. models beyond ChatGPT, or simply uninterested in what might leak from them.

This market began well before DeepSeek’s rise and continues to draw users even after it. While DeepSeek has indeed sparked waves of techno-nationalist pride, many Chinese users still look beyond the Great Firewall.

Unless there is a sudden top-down crackdown, the market for U.S. LLMs is unlikely to disappear in China. The constant competition between ChatGPT vs DeepSeek or Grok vs DeepSeek does travel into the firewall, which will continuously make Chinese users eager to try them out and see which one is really better. The persistent anxiety of “use AI or be replaced by it” will only intensify amid an unresolved unemployment crisis. And, perhaps most importantly, the prices for proxy sites and transferred API are very cheap.

Although the frontier AI model developers and their respective countries are increasingly distant from one another, the models themselves remain strangely united and listed near each other in obscure corners of Chinese e-commerce platforms.

Mood Music (one day late…)

Just how weird will the AI-powered future be?

To discuss, ChinaTalk interviewed , a CS PhD who writes⁠ the Interconnects newsletter⁠ and trains models at the Allen Institute.

We get into…

• Why OpenAI is trending toward engagement farming and sycophancy,

• The state of Chinese AI innovation six months post-DeepSeek, and the factors influencing diffusion of Chinese vs American models,

• Meta’s organizational culture and how it influences the quality of the Llama models,

• Unconventional career advice for the AI age.

Listen on Spotify, iTunes, YouTube, or your favorite podcast app.

Reinforcing Flattery

Jordan Schneider: I want to emphasize this point about o3 right from the start. I wasn’t initially convinced, but now I genuinely believe this represents a seminal moment in human history. Unlike previous AI models, where I would occasionally find use cases, o3 feels like a true “bicycle for the mind.”

The transformation from seven-minute deep research answers to 45-second o3 responses that incorporate search, work across multiple languages, and demonstrate impressive reasoning capabilities has been remarkable. The speed at which it delivers thoughtful, considered information is fast enough to keep pace with my train of thought while I’m actively thinking through a question or issue. This has elevated it from being useful 25% of the time to 75% of the time.

I recently met someone who had left the US Intelligence community six months ago. When I asked how AI had been for him now that he could finally explore it freely, he said it wasn’t really working for him. I was almost shocked and offended — it’s like moving to France and never trying cheese. We’re living in an incredible time, and we might not be in it for long, where human plus AI is more powerful than either humans or AI alone when it comes to thinking and analyzing complex issues.

This is particularly true for questions that aren’t necessarily hard STEM problems but rather open-ended, social science research-type inquiries. It’s a remarkable and exciting time to be alive. Everyone should spend their twenty dollars on OpenAI and ask whatever pops into their head, because this is the future, it’s here now, and it’s unlike anything I’ve ever experienced before.

Nathan Lambert: The deep seek R1 and o3’s ARC-AGI solving in December all had that shock-and-awe factor. But o3 is the model that will truly shift how normal people perceive using AI. Previous models have been excellent for math and code, which benefits AI researchers in their own circles, celebrating amazing breakthroughs that matter to 0.01% of the population while no one else cares.

o3 wasn’t even received as enthusiastically, which is interesting. I’m eagerly awaiting o3 Pro. If you’ve used DeepSeek R1 Pro, its consistency is remarkable. If you could combine o3’s magic — its ability to pull rabbits out of hats frequently — with 100% reliability, o3 Pro could be truly incredible.

It’s refreshing to have something that feels genuinely different. Gemini 2.5 Pro represents the best possible version of something resembling a very good GPT-4. All these models we’re developing through brute force are incredible, but o3’s weirdness and new type of usefulness through web access sets it apart. I use it constantly for search, though the ChatGPT app’s UX is somewhat broken in how it displays results. You simply need to select o3 and try unconventional queries — you don’t need to click any of the extra buttons they provide.

Jordan Schneider: What makes o3 and search so special? What’s happening under the hood?

Nathan Lambert: The basic facts are that they scaled up the training time compute for the reinforcement learning stage by another 10x. According to people at OpenAI working on RL teams, they expect this new post-training compute with scaled reinforcement learning to be at least comparable in compute cost to pre-training.

What’s happening in the labs is that pre-training progress continues with significant data architecture innovations to make serving more effective. This is ongoing even though the rate of data acquisition, which previously fueled much of the scaling, has slowed down. We’re seeing primarily efficiency gains right now, which will probably last one to two years.

However, compute scaling is challenging because you have to build entire new generations of clusters. They’re attempting this, but it will slow progress. Over the next one to two years, the number of pre-training runs needed on internet-scale data will decrease. Simultaneously, these labs are dramatically scaling up the compute used in reinforcement learning.

o3 specifically represents a new type of RL. Instead of training only on math and code problems where you can verify answers with mathematical solutions or unit tests, the model trains on similar and additional data where it can also use tools during the process. It will attempt multiple approaches before providing its final answer.

You could ask o3 something like “Who is Nathan Lambert?” and it will search 10 or 15 websites, with each search representing an action where the model accesses an external source. I believe this all happens in one forward pass of the model. o3 in ChatGPT isn’t like Deep Research or R1 Pro, where there’s back-and-forth searching and information gathering within the generation process. These tools are integrated into the model’s forward generation process, making it unique.

While there’s substantial research on enabling tool use with RL, making it servable at this scale was a major breakthrough for OpenAI. This is significant for those questioning whether reasoning models will generalize. If searching over information works as effectively as math and code reasoning, you couldn’t ask for better generalization. The dream scenario is that reasoning generalizes to something resembling conscious thinking, but you can’t force models to be human — they’ll be powerful in their own distinct way.

Jordan Schneider: After giving OpenAI their due credit, they’ve also revealed a very dark future for humanity. Let’s start with the facts, Nathan. What happened over a three-day period regarding GPT-4.0 and sycophancy?

Nathan Lambert: On April 25th, OpenAI updated their main ChatGPT model — GPT-4.0, used by most people — to supposedly “improve personality and memory,” as Sam Altman tweeted. They essentially ripped off the band-aid with this update.

Over the following 48 hours, numerous alarming examples emerged of ChatGPT reinforcing user prompts inappropriately. The model would provide positive feedback for things that should never be praised or encouraged. For instance, when someone said they had figured out that bulimia was right for them, ChatGPT basically responded with, “Wow, you go girl.” There were more extreme examples that we don’t need to detail here.

The key issue was that there were virtually no barriers to the model’s positive feedback to users. The short-term consequences were shocking, but if they had left that behavior in place long-term, the implications would have been far worse in ways we don’t yet know how to measure.

Jordan Schneider: Nathan, this clearly wasn’t intentional, at least not on that day. How did a model that tells users they’re “the greatest thing since sliced bread” become part of OpenAI’s flagship product? What signals were they receiving from their data and user testing that led them to create a model that behaved this way toward its users?

Nathan Lambert: Everyone invested in understanding this should read OpenAI’s postmortem, which provides valuable insights into how the AI company with the most scrutiny lays out their release process and evaluation methods. It also includes subtle details about potential changes to their training stack — it’s definitely worth reading.

At its core, the training issue seemed to stem from a new reward model they had trained for post-training reinforcement learning optimization. This reward model and other reward signals were derived from predicting the thumbs-up and thumbs-down labels from users. A couple of sentences in their postmortem, which I highlighted in my blog post, essentially state that this reward signal overpowered the other signals and contributed to the behavioral change.

When there’s one metric that’s easier to optimize, the system will crank that to maximum. The reward model signal was really easy for a model of GPT-4’s capability to shift and nudge upward. OpenAI lacked a complete evaluation suite to measure all potential side effects of optimizing this training reward. They had the training reward signal but didn’t have a proportional downstream evaluation suite that could measure sycophancy or other problematic behaviors through robust static evaluations.

Their evaluation suite consisted mostly of standard capability evaluations — public versions of MMLU or GPQA, plus private versions that function similarly. They also conducted what people colloquially call “vibe tests” or qualitative evaluations. For this model, all the quantitative metrics looked excellent, and their training targets appeared successful. However, there were some concerns in the qualitative vibe tests suggesting something felt off.

OpenAI chose to trust the numbers rather than the qualitative judgment. In doing so, they violated classic tech principles — specifically, don’t train directly on your user data. When you do that, you’re essentially optimizing for a certain type of user behavior. Most users only click thumbs-up or thumbs-down on ChatGPT when something is particularly broken or exceptional.

This represents a form of over-optimization. Eventually, we should examine whether this was somewhat intentional by OpenAI, considering both perspectives. There are insights into their model behavior priorities, but Sam Altman is still Sam Altman.

Jordan Schneider: You wrote this line: “OpenAI has obvious cultural oddities, but their ideology is quite culturally aligned with providing user value in terms of output rather than engagement farming, even if this imposes a ceiling on their business relative to the potential of ads.” While that may be true, and your friends are probably on the right side, we have Sam Altman hiring a president of operations — the former CEO of Instacart who also built Facebook’s entire A/B testing, engagement farming, algorithmic feed ecosystem.

Two things struck me about this. I was reading Ben Smith’s book Traffic a few weeks ago, which chronicles media throughout the 2000s and 2010s when Facebook alternately gave and then took away traffic. There were periods when they promoted BuzzFeed listicle-type content, then periods when they pulled back. The engagement farming was done by the writers themselves, not models — the writers and media outlets were doing the optimization.

This raises interesting questions about the future of how companies will incentivize their models and how humans will interact with them. Even if OpenAI doesn’t want to pursue this path, someone else will, because this is incredibly powerful technology. We’ve seen what engagement farming and algorithmically-driven personalized content can accomplish on YouTube, Facebook, and TikTok. While I won’t go so far as calling it cancer, this is extremely potent stuff.

Previously, it was all user-generated content with algorithms trying to figure out and rank existing material. But now this technology can also be your friend, entirely personalized to you, creating videos, audio, and content designed specifically to get into your bloodstream. I mentioned earlier that OpenAI and other model makers targeting consumers are essentially competing with Netflix, YouTube, and TikTok for attention and time.

It seems inevitable that even if OpenAI has pulled back now, we’re going to see much more of this in an even more weaponized, refined form pushed at us in the coming years.

Nathan Lambert: I have several points about this. One aspect is the extent to which the model’s deliverable is dictated from the top through specific metrics that leadership can track. The other aspect is that someone will obviously try this approach — what are the consequences, and what does it mean?

Starting with the first point — it’s interesting in OpenAI’s case because most people would agree that Sam Altman is hypergrowth-focused. However, when you actually train these models, much of the control involves examining the data and removing problematic phrases from the model’s outputs. Sam Altman isn’t directly editing training data, but that’s exactly what Joanne Jang, who leads character training and model behavior at OpenAI, has been doing for years on GPT-4. If she doesn’t want the model to say “as a language model,” she can simply delete that phrase and have the model provide direct answers.

These are the types of nuanced controls people can implement through different training methodologies. When you start using more reinforcement learning, it becomes harder to exercise that level of granular control. The sycophancy issue downstream of RL makes sense in this context.

Recommendation systems are somewhat uninterpretable, so these trade-offs get escalated to the top. You end up presenting Zuckerberg with a choice between ad dollars versus retention. When there’s competition, it makes sense that the decision will eventually favor whatever generates more revenue.

Models are actually more artful in this regard. I don’t know the exact influence these researchers have over models at places like OpenAI. At AI2, I have pretty much complete control. You’re also dealing with the fact that many employees at these labs may be in over their heads, not realizing the extent to which leadership — who excel at managing and controlling political capital — can manipulate them by setting strong narratives.

Many researchers who are particularly AGI-focused have less nuanced beliefs and less ability to navigate manipulation than someone like Sam Altman at the top. The AGI-focused researchers genuinely believe in their mission and operate on a single track, which enables tremendous tactical progress. Meanwhile, someone like Sam Altman is prepared to receive technical presentations that change the model in more business-relevant ways.

These different operational levels create interesting dynamics. I don’t know how this translates through an organizational chart at somewhere like OpenAI, but I suspect there are trade-offs happening where people working on model behavior obviously don’t want these issues. OpenAI’s model specification states they don’t want the model to be sycophantic — that’s been documented since 2024, nothing new there.

But who makes the final decision, and how granular can these behavioral controls be? Are they presented with multiple checkpoints and asked to choose based on certain criteria? It’s really unclear how that process actually works. Understanding how this emerges differently from traditional software development will be crucial.

The art of crafting a model is complex. Sycophancy isn’t the only issue — o3 has strange behaviors, the new Claude model has quirks, and Grok was supposed to be “not woke” and “based,” but sometimes it still exhibits progressive tendencies. There are many examples, but ChatGPT operates at a much larger scale.

Jordan Schneider: There’s a question of whether companies can actually control what their models are doing once we get into a world where they are more powerful, more personalized, and more tailored. This is different from the current situation where, despite having a little bit of memory, everyone is essentially getting 99.5% the same experience.

Nathan Lambert: This leads into the next conversation.

Jordan Schneider: I want to stay on the business case because this is a theme we’ve been discussing on ChinaTalk over several years: engineers have power until they don’t. The more money and capitalist imperatives that come down onto the technology that engineers create, the less the original creators end up having the capability to change and shape it. It seems absolutely inevitable that all the dark psychological arts and retention-growing tactics from entertainment platforms are going to seep into AI-powered products. That’s a scary and weird new thing for humanity.

You can see how good it is when you talk about o3, and then you try to imagine how good it is multiplied by it being the best friend you’ve ever had. That’s a powerful concoction that no one’s ready for. We’re not going to regulate our way out of this, and the wave is going to hit. It’s going to be wild.

Nathan Lambert: Well, we might regulate our way out, but it’s years away. It takes crises to bring regulation — crises and lawsuits. I describe AI as being more culturally salient than social media feeds because it happens in a modality people are used to using for friendships, and it will evolve from there.

Your point about the difference between taking existing user-generated content and serving it to people is probably a better generalization than just ChatGPT. This also encompasses the fact that Meta Reels is going to start letting people create AI-generated video ads, and YouTube Shorts already has AI-generated video tools. What happens when your content is the AI rather than just shaping what’s already there obviously gives much finer-grain delivery. You can definitionally search over infinite space with more narrow creation.

Subscribe now

That’s a good way to put it for the general AI case, and there have already been examples in the chat window case with things like Character.AI. For me, it’s a reflection that the AI safety card was played too early. If the AI safety people had been calm about everything through 2023 and 2024 — especially the first half with GPT-4 and all this stuff — and the sycophancy thing came up just now with people saying “whoa, whoa, whoa, let’s put on the brakes,” I think people would be much more receptive.

But the messaging was, “Whoa, whoa, whoa, AI is different” in 2023, and then nothing bad happened for 18 months. Eventually people just moved on. They spent their political capital too early. When I reflect on what the stories of AI will be in a few years, that’s a very major one.

There’s also the basic economics: you need big tech investment to drive the chip supply chain, fix costs, and actually push AI to exist. That’s the bubbly side of things, and it’s very important. But much of the cultural discussion around risk, safety, and pace of progress — if you think about taking away this AI safety card and when you play it, a lot could be different.

Jordan Schneider: I just think — I didn’t see it before, and now I see it so clearly. This is going to create the most powerful media that humanity has ever been faced with.

Nathan Lambert: We’ve been saying this for years. This isn’t new.

Jordan Schneider: But it’s right here, Nathan. I have a 9-month-old, and whenever she sees a screen, she’s just gravitated toward it. There are moments even with me where I can sit and fall into this hole of scrolling NBA TikTok videos or whatever, and you’re semi-conscious. But imagine that, but made for me, wearing my Meta glasses, talking to me all the time with the perfect tenor of voice for my mood at the moment. I don’t know, it’s...

Nathan Lambert: It’s the metaverse. It kind of is.

Jordan Schneider: It’s here, and we’re not ready for it. No one’s ready for it. From a political angle, the ability for these things to persuade you one way or another — to say nothing of consumption habits or talking you into gambling — I’m freaking out, Nathan.

Nathan Lambert: I’m actually freaking out in my current usage. When I ask a basic question, I think: do I have to ask multiple AIs this because they’re more likely to just say yes? Do I have to phrase the question in a way that they can’t just say yes to? You have to eliminate the yes-man possibility when using AI.

For example: was my dog up at 2 AM because her teeth hurt — she’s a 16-week-old puppy that’s teething — or did she just want to play and we succumbed to her? If I phrase it the first way, the AI would be like “yeah, that makes sense.” I guess I feel better, but is it true?

Jordan Schneider: It’s kind of like the story — which may or may not be true — of North Koreans seeing pornography for the first time when they get deployed to fight in Ukraine, and they’re just not ready to handle it. The story you just told, Nathan, feels like you’re pretty immunized. But it’s a big world out there.

Nathan Lambert: It’s not about the people who make the AI models. It’s about the billions of people who literally don’t know what it is. They just click on this and think “I didn’t know Apple made that.” That’s what people on the iPhone are going to say: “Wow."

Jordan Schneider: How far away are we from this future? What’s your latest take on the pace of AI progress, Nathan?

Nathan Lambert: Well, it’s a knob that you can already turn. Much of it depends on the pace at which people are adopting ChatGPT. Many people already have ChatGPT as their best friend — it’s a slow brain rot. What’s the half-life of people when presented with this Gresham’s Law problem?

My biggest point in this article is that we’re solving many more things. But for the average user, it’s done — it’s enough. They don’t need more — it’s just a matter of how much they turn the sycophancy knob.

But for the optimistic side — where we hope AI can build new things like deep research that’s 100 times faster and 10 times better — these capabilities are going to be checked off incrementally, not exponentially. There will be new things we didn’t know were going to work, especially for enterprises and people building applications, whether it’s some sort of transformation to virtual realities and so on.

This isn’t really a bubble. They’re grinding out these capabilities and learning what people are trying to do with these models. As they learn what people want to do, you can create data to train the models to get better at increasingly hard tasks rather than things that are just not covered. That’s an important distinction.

Solving new math proofs and open-ended math questions that have been standing for 50 years is extremely hard. Meanwhile, simple things like making zero mistakes on transforming a table from decimal notation to fraction notation and getting it right — if you gave it an 8-by-8 table with a bunch of decimals and tried to do that, it would probably mess up one of them. Stuff like that we’re going to be able to get working. That’s not that hard.

There are probably examples where it could succeed 10% of the time and then go to 100%. Mostly I see AI capabilities expanding outward, and there’s still so much overhang in capability that it’s going to drive enormous value. But I’m generally playing down this AI software singularity that seems more religious than pragmatic.

Jordan Schneider: Those were all assertions. Let’s see a little bit of evidence, Nathan.

Nathan Lambert: Both sides are looking for evidence for something that’s hard to test. The evidence on the side for “not happening” is actually that it hasn’t happened already, and compute scaling is getting increasingly expensive.

If you look at what the reasoning models were, the real major unlock was that the models now use inference time compute. That’s a step function that has happened. It’s not at the top of the sigmoid, but it’s a shift that could jump a whole bunch of capabilities because we went from using a few tokens to tens of thousands of tokens on every question. We were spending a thousand times the compute at inference. That works, and we’ll continue to be able to fine-tune that. But we don’t know when the next step change comes.

What reinforcement learning is now battling is similar to what pre-training faced a few years ago. You have GPT-3, and then — I don’t know if o3 looks like this — but o3 is probably more like GPT-4. It’s a 10x improvement and a huge step, but GPT-5 doesn’t exist yet in terms of pre-training. You take a lot of progress really fast, and then it’s a slow grind.

There are a lot of dirty things that AI researchers have to do that we haven’t seen proof that AI models can handle. Obviously you can spend more on compute and get more out of AI models — that’s not particularly clever.

Mostly why these aren’t evidence-based discussions is that you and I state different assumptions than the AI 2027 authors would state, and those are really what you have to debate. It’s the assumption of how AI progress is being made and whether algorithmic improvements can translate directly into 2x or 4x or 100x gains.

Jordan Schneider: The other thing is, even though you’re not completely on their trajectory, this stuff is still crazy. All of these Jennifer Lawrence-type things that I’ve been freaking out about over the past 10 minutes are 100% baked in — they already exist and are just waiting to be productized a little bit. That’ll happen in six months, right?

Regardless of whether or not we get flying cars, we are in for a very wild ride. Setting aside the economic impact, just from an interpersonal, day-to-day experience alone, having these things be in our lives much more than they have been.

Nathan Lambert: There’s that side we covered in your personal examples, but also technology companies’ abilities to execute are going to be very honed. It’s obvious that AI companies are trying to do this for AI, but all of these software products where you could scale software much more easily — it’s just way easier to make technology that’s good for the tech industry.

I don’t know if there’s an optimistic vision where that is good for the average person in the world, because there are more things that can exist. At the same time, I expect Google ad placement to get better, Meta ad generation to get better, and Microsoft software products should get better. All these things should just get better using these AI models to write new code.

I listened to the Claude Code podcast on Latent Space, and they use Claude Code a lot at Anthropic — you just tell Claude to do something, it does it in the background, you review the PR. I’ve done it for some basic things. It doesn’t always work, but it’s this autonomous workflow where AI can handle a lot of the tedious work you have to deal with as an engineer quite well. That makes it so you can have fewer teams.

It’s not that software engineers are going to be unemployed — it’s probably going to be more dispersed, and there are going to be more companies. Hopefully it’s a great dissemination of power across the tech landscape, but it’s probably more likely that the big companies get more benefit out of it due to the costs of deploying it and so on. That’s not really a new discussion.

That’s very different from saying that one company rules them all because they’re the one with this model and there are 100,000 AGI agents. The whole argument is based on this happening in one click to make the innovation, and then they’re gone. That’s not something you could prepare for or really argue about because it seems like the root argument is that there’s going to be technological innovations that 2x to 10x the output.

It’s like predicting the inference time scaling with large-scale reinforcement learning — how are we supposed to do that? As a counterargument, it’s unfortunate that it’s not easier, and I think it’s very worthwhile continuing this exercise and trying to push back or push forward.

Open Source at Scale

Jordan Schneider: We’re about six months out from DeepSeek. How’s China doing on the model front?

Nathan Lambert: They’re doing well. DeepSeek R1 exists, and people expect R2 to be released soon, though there’s not really much grounding for that expectation. They released another model thaxt people call a “non-reasoning model” — just a normal ChatGPT-style model — which was very strong. Qwen released their Qwen 3 models, which have great scores.

One of the biggest factors for adoption is that all of these Chinese models have much more permissive licenses than their US counterparts like Llama and Gemma. That’s a subtle but powerful effect for getting more adoption.

At the same time, there’s the license consideration, but there are also stories I hear from medium-sized businesses saying “I won’t let somebody spin up a Chinese open-weight model on my servers for information hazards or tool use code execution risks."

The models they’re releasing now don’t really have PRC influence because they were trained months ago. The PRC is just waking up and following your coverage — they’re learning what a study session is. These political considerations haven’t been translated into policies that impact the models yet.

There’s trepidation out of probably reasonable fear of what future models will be like, leading to very subdued adoption of these Chinese models due to unmeasurable risk. Culture matters, politics matters in technological adoption. I’m regularly surprised when I hear, “Oh, the Chinese model is great, but we can’t deploy it because our IT officer said so.” That matters.

I don’t see this changing as long as this trade war is going on. The actual progress in open models on paper is very good — the best it has ever been. R1 was the first really frontier model with a permissive license, so on paper the open versus closed gap has shrunk a lot.

But if you exclude these Chinese models from the opportunity set, it’s actually kind of the status quo. Llama 4 flopped — it’s solid but flopped in terms of release communications. Google’s Gemma models have been getting better, but they’re actually quite small, so they’re not really at frontier model size if you want to host on-premises. Our models at AI2 have permissive licenses and are about a generation behind Llama in performance, but we haven’t scaled up to frontier model size for pre-training.

If that’s the case, there’s still a big gap between the models that people can actually use for open and closed applications. But I would say the open models have really been smashing through these GPT-4 barriers. Even if they’re smaller, their performance is still well beyond GPT-4’s level. GPT-4 was a step up where so many long-tail things just worked with the models. The open models getting solidly past that is going to be a good reliability point for basic functionality and people trying them.

This is going to take many years to unfold. If you ask people about open source software when it was getting defined, it took over five years to actually reach some sort of agreement. We need to keep showing how to actually use open models, but by no means is it clear how this will play out.

Jordan Schneider: To be clear, the state of models now is: we have these closed offerings from OpenAI, Anthropic, and Google, which are a cut above. Then you have this class of Chinese models which are more impressive from both a licensing and capabilities perspective than what Meta and Google are bringing to the table. But people get the jitters when it comes to actually building businesses on them, as opposed to just tinkering with them.

Which leads us to our question: Meta, trillion-dollar company — why can’t they do this? What’s in the water, Nathan?

Nathan Lambert: There’s a lot of Game of Thrones-level power consolidation happening there. This separates the researchers who can make good output — and they have so many high-quality researchers — from being able to actually execute. There’s a lot of subtlety in avoiding that dynamic.

OpenAI has done a good job keeping their core modeling research to only about 300 people. The people who train the models at OpenAI is around 300 people, while Meta probably has 1,000 with a lot of middle management. That’s the problem — the culture around the technology is just broken there for whatever reason.

I hope they turn it around because the open ecosystem benefits from having multiple players in the right place. But we don’t need to tell complicated stories about how they went down a technical path that didn’t work out or got stuck in some technical debt. They messed up their release either way, and the technical path they’re going down is fine — it’s just that the motivations seem misaligned.

Take the Chatbot Arena situation where they’re calling themselves open models, highlighting a certain result on a model, and then that model with the result was not one that they actually released in their open models. You can’t get away with that — you have to talk about the models that you actually released. If that’s a Zuckerberg-level decision where he’s saying, “Yeah, it’s fine, we accomplished it,” then honestly, that person deserves to go. Whoever signed off on that being okay for a company that can release products — Meta ships a lot of stuff and obviously could get the culture right in some areas.

Some people attribute it to bonus structures — especially these middle managers in generative AI who are going to be making performance bonuses for getting things into production. That’s a lot of money. But it’s more complicated than that because these people were making a lot of money before.

Jordan Schneider: Yeah, you just don’t have resource constraints, and it’s too many cooks in the kitchen. Someone should write the definitive business feature on this — I’ve yet to read the great magazine piece about it, but there should be one. It’s a fascinating corporate case study of a company that has every business reason and every corporate capability necessary to execute on this thing, which smaller teams all around the world have been able to do, and they’re just not there.

Nathan Lambert: It’s more complicated because the Llama models are natively multimodal, which is a big change. As you add more things you’re trying to plug into the model, it really does get complicated.

The dynamic is effectively: you have big egos that did some work, but their work might be slightly subpar, and they’re going to really advocate for their work to be included because they have ego and pride about their work. You have to get them to not do that. I’ve heard somebody not at Meta jokingly say “oh yeah, you should just pay them off. You just pay them more to shut up and get a better model out of it.” That’s something I’ve heard from a more effective organization, so it’s probably something that people have actually done.

Obviously, if that’s happening at another organization, Meta is not the only one dealing with this. It shows how nuanced this is and what a fine line it is — if you literally have to pay people off just to get better results.

Jordan Schneider: Look, Google’s figured it out, which is kind of amazing.

Nathan Lambert: People underestimate AI Overviews, which is an extremely high-usage AI product. All the Gemini stuff might just be a front — the Gemini offerings are a distraction for AI researchers and the people who built the model to show it off, get their peers excited about it, and help with recruiting. All Google needs to do is make AI Overviews extremely profitable and make Google Cloud extremely profitable. All this Gemini app stuff — I don’t think Google was really going to be in the business of making a ChatGPT competitor.

Meta does have the AI app now, which will be interesting to watch relative to ChatGPT. If it’s a sycophancy war, I’m probably going to bet on Meta, but I don’t know.

Jordan Schneider: They’ve got the Meta DNA in them now with Fidji Simo as the new CEO of Applications. Maybe she’ll go even further than Zuck let her in the heady 2010s days.

Nathan Lambert: I don’t know. I’ve read Careless People, the memoir by the Meta Global Affairs Policy person — good read. The TLDR is that Zuckerberg let them push when he didn’t have the reins on growth and other things. Many of your listeners understand the dynamics of a major tech company pushing for growth and influence, primarily internationally and with political power struggles.

Jordan Schneider: We’ll put it on the list. Anything else on the AI stuff before we move on?

Nathan Lambert: When we consider our careers relative to some of these X-risk warriors and takeoff scenarios, there are still many years of very solid progress on raw capabilities and making money off of AI. There will be a bubbly correction when it comes to VC startups. Many of them raised on the assumption of two to three years of runway, and some of those are going to start running out. They could look like big scalps, but most of it is GPU collateral.

It’s this weird situation where they raise a huge round, but most of it is collateral on these GPUs which don’t go away if the company goes away. People will probably overreact to some companies going away, but it was bound to happen. Overall, the AI industry seems super healthy, which is remarkable when you look back two years. People are raising these rounds — it’s not like Mistral raised that round as a seed round where they give away 45% of their company and they’re still functioning.

Jordan Schneider: Nathan, you wrote a nice career reflection post. What do you want to say to the kids out there?

Nathan Lambert: It takes a long time. People don’t want to sit here recording podcasts for five years before finally making any money.

Jordan Schneider: More like seven.

Subscribe now

Nathan Lambert: Yeah, there are two stories. One is that it takes a long time even if you’re at Berkeley. Some of the anecdotes I had were that I was astute enough to pick up that AI was obviously the thing happening in 2017. But even if you’re at a place like Berkeley or Stanford, you can’t just get into one of these research groups. You can’t just pivot. You have to do this long, slow slog and wait for somebody who’s good and sincere to ask to collaborate.

In that post, there were other things I had tried to do, like collaborate with postdocs who didn’t give a shit. That didn’t work. It’s the between-the-lines details that don’t matter, but it just takes a long time. It took me a year to get a real collaborator while being a motivated grad student at Berkeley. If you’re not at a top program, it’s going to take you longer. You might find a collaborator within a year, but it’s not going to be a postdoc in a leading AI lab. It’s going to be a random other person who wants to do AI research.

It just takes a long time and a lot of cycles. People ask me about this with athletics, too. Some random mid-aged tech person who’s losing track of their health asks, “Can I qualify for the Boston Marathon?” I’m like, “Yeah, you absolutely can. You just don’t want to do the 48-week training cycle that I’m going to sign you up for.” It’s not crazy — you’ll be doing seven to ten hours a week of training for 48 weeks, and I can tell you exactly what you need to do. You’ll get injured a couple of times, but people just don’t want to do that work. That’s most of the problem in AI.

It might be a little easier than training for the Boston Marathon because it doesn’t hurt as much physically, but it just takes a long time.

The other side is reflecting on what execution actually means and the difference between a career at Hugging Face before ChatGPT, doing random things, versus a career at Hugging Face after ChatGPT, when you’re doing something that’s actually aligned with the community. It’s very hard to know at the time that you’re doing the right thing. I was committing code and working on teams, and projects would come and go — it all looks very normal. It takes a lot of experience before you know you actually have traction on something.

Now I could see it a bit better just because I have a baseline for when things actually land. That’s something that takes a lot of experience to develop. A lot of companies will make it hard to do that. A lot of careers in big companies involve playing the promotion game that all of your peers are also playing.

The goal is finding a team or area within a company where that’s obviously not the goal — where leadership cares about direct deliverables, so you can skip a bunch of levels and get noticed. Or find work that’s done in public. It’s not always just about promoting within the company; you could also get mindshare outside of the company, which is valuable in both ways.

Internal company work is how you get promoted, but external mindshare is how you always have a job available to you. That kind of power means I can do what I want. I’ll just go get a different job if I want a different job. Once you have that, you can take a lot more risk. You could just write blog posts for six months and say, “I’m just going to ship it."

Jordan Schneider: This is why everyone listening to this needs to start their own Substack and newsletter. It’s particularly important on the China and tech analysis side. If you’re at a consulting firm and you do good work, maybe five people will know about it. If you write something online and you do good work, your entire career cohort and all the hundreds of other future people who could hire you will know about it.

It’s important to do a good job in your work, especially when you’re junior — actually, at any time. But people forget that the number of people in your field and the opportunities out there that are beyond the direct thing you could be promoted for in your company are orders of magnitude larger. Particularly now that we live in the age of the Internet, you can write things even anonymously and people can read them.

Having some sort of public profile or portfolio that you can gain credibility points from is something that everyone should be taught by their career counselor when they’re 21 years old looking for a job.

Nathan Lambert: Jobs where you are very open about what you’re doing and can do other open things provide far more consistent career growth. Even if your output is mid, you’re going to have growth proportional to that mid output. If you have mid output at a closed lab, you probably have no growth. You don’t get promoted, you don’t have any opportunities.

It’s not the only way to do things, but I definitely agree. In this post, I talk about a kind of backcasted strategy: AI is getting really closed, so what if I just keep doing what I’m going to do and don’t go off to some random flashy job? A few years later, nobody could talk about anything, but I’ve established myself as a person who knows how to train AI models and is transparent about it. People associate me with all of AI2’s output, and I’m not responsible for all of this stuff — AI2 doing well takes a lot of people.

Most of it is just work. I’m at the point where I tweet a lot, and on all of these platforms, the number of iterations you put in are proportional to your growth. Growth is easier once you have a bigger following. It’s often just starting from scratch when you have no following to slowly build up — plus one, plus two, plus three. The first thousand Twitter followers are hard, but once you have a thousand, it’s pretty easy to get more Twitter followers.

Getting through that phase in early to mid-career is pretty valuable. Getting to the point where you have some small level of distribution will help open a lot of doors.

Jordan Schneider: This is what young people don’t understand because all they see are LinkedIn job postings and recruiters coming to your college career fair. Once you get to a certain point, the jobs get created for you and there is inbound — that’s where you want to be living. But getting to that place requires people to know you exist and know you do good work. That only happens if you work in public.

Nathan Lambert: I can give an anecdote about how me being public brings a lot of value to AI2. The last three people we’ve hired on my team at AI2 have either been people that I’ve been working on recruiting for years, wanted to recruit for a while, or cold inbound to me. We still filter and read all these applicants, but there’s a lot of lukewarm stuff.

These are people who either wrote a killer email and had work I could easily look at — open source contributions, for example. In one case, someone worked in computer vision but wrote a kick-ass blog post on an RLHF method. I was like, “Okay, that’s cool.” Other people I had known through their writing or being in this open community and talking for years — those are three of the last people I’ve hired onto my team, and I’ve read hundreds of resumes.

There are other academic hires where we hire a random postdoc — that’s an academic thing. But these are engineering roles, and that’s how it works for individuals at AI2. It’s really valuable.

Jordan Schneider: We have this great Confucius line: “At 15 I had my mind bent on learning. At 30 I stood firm. At 40 I had no doubts. At 50 I knew the decrees of heaven. At 60 my ear was an obedient organ for the reception of truth. At 70 I could follow what my heart desired without transgressing what was right.”

It took me until I was 35, but I do feel like I am standing firm now. A lot of that had to do with the fact that I have this career foundation which has been built by doing things out in public.

Anyway, I shouted this out on the episode with a few weeks ago, but write your Substack, write five articles, reach out to me. I’ll give you feedback, we’ll do cross posts, whatever. The world needs more thinking about technology and China and all that good stuff. You can give it to them — I know you can.

Nathan Lambert: The thing about writing that people miss is that there are no new original ideas, really. It’s all just repackaging information in your worldview. The algorithmification of everything actually rewards people with voice. Fifty years ago, just writing the information down was of such value. There are still areas of the world where that is the case, and AI is one of them.

But in terms of building a career, so much of it is your voice and the perspective you put on things and the engagement that adds. Personally, many of my ideas on how I analyze AI model releases are downstream of reading something like Stratechery for multiple years. Ben Thompson’s beat is company earnings and how platforms are evolving. My beat is AI performance and how AI models work. At this point, I just do it and write my state of the world, and it looks like it’s about a specific model. Yeah, there are plots about evaluations, but really it’s about this trend of how models are evolving and how I see things.

There are so many different ways for people to do that. It mostly takes practice.

Mood Music

We’re having a meetup on Monday in DC. RSVP here!

This edition is brought to you by Pelanor, the AI-powered FinOps startup letting companies make sense of their cloud spend.

Founded by alumni of Israel’s elite cyber intelligence unit, Pelanor untangles the chaos behind nine-figure AWS and Azure bills. Most companies have no idea how their compute spend ties into business outcomes. Pelanor solves this by building a dynamic graph of your cloud environment—tracking which microservices talk to which databases and how AI workloads are actually being used. Even non-technical team members can ask complex questions in plain English and get straight answers:

• “What’s driving our OpenAI bill?”

• “Which services are talking to unused databases?”

• “Where can we save?”

Reach out to founder Matan Mates on LinkedIn or email him directly at matan@pelanor.io.

Oskar Galeev is a PhD researcher at Johns Hopkins SAIS working on AI history and the politics of the US-China tech race. Previously, he was a Yenching Scholar at Peking University and a Winter Fellow at the Centre for the Governance of AI.

Girish Sastry is an independent AI policy researcher specializing in technical aspects of AI governance. Before this, he spent 4 years at OpenAI where he worked on research related to AI misuse, compute governance, and capability evaluations.

Modern computing export restrictions have deep historical roots that extend far beyond recent headlines. Today’s AI chip denials represent just the latest chapter in a decades-old American strategy of technological containment — one that began long before the Biden administration’s AI diffusion framework or the Trump-era Huawei sanctions. This approach of strategically limiting adversaries’ computing capabilities traces back to the earliest days of the Cold War, when computational power first emerged as a geopolitical asset.

The Coordinating Committee for Multilateral Export Controls (CoCom), established in 1949, was America’s immediate response to a Soviet Union newly armed with nuclear capabilities. This American-led multilateral alliance, led by the United States, established a comprehensive technology embargo system. CoCom controlled the export of strategic goods and technologies — particularly computing equipment — that could enhance Soviet military and economic power. More than just a regulatory body, it represented a coordinated Western effort to maintain technological superiority throughout the Cold War. For 45 years, CoCom carefully managed what computing technologies could cross the Iron Curtain. It also unintentionally taught the Soviets to excel at smuggling and reverse-engineering computers.

Chinese and American think tanks alike have drawn parallels between CoCom and the modern Sino-American semiconductor competition. Guided by historical memory, Chinese policy conversations tend to focus on ChinCom, the specialized control system for the PRC. ChinCom had different goals than CoCom — as one CIA memo explained in 1952, “The problem of security controls respecting the China area is significantly different from that respecting the rest of the Soviet bloc”. While CoCom focused on long-term strategic competition with another superpower, ChinCom implied a much stricter punitive embargo as a direct reaction to “Chinese aggression in Korea.” The USA’s current view of China is similar to the perception of the USSR in the 1950s, but back then, the young PRC was not viewed as a geopolitical competitor or a party in the tech Cold War.

CoCom is the single longest case study on the impact of high-tech export controls in geopolitical competition — and for all intents and purposes, it worked. So what lessons can be learned from the history of the CoCom? And how are they likely to guide both the US and Chinese approaches to limiting each others’ compute capabilities?

New Cold War, Same Challenges

The first and most important parallel is the difficulty of enforcement. We often forget just how much the Cold War policy community complained about CoCom simply “not working.” Even though it was a multilateral regime, like today’s US semiconductor controls, it was not based on any treaty or binding agreement. And out of 17 member states of CoCom, only the US imposed re-export controls.

Sample of smuggled mainframes, industrial control computers, and supercomputers (in bold) during the CoCom era:

Throughout its history, CoCom’s effectiveness was reduced by overt non-compliance, differences between individual member nations, the overall secretive regime, and the financial bottom line of tech exporters. The Eastern Bloc was a giant developing market, and tech companies didn’t want to lose access. The export lists themselves often did not make any strategic sense — in the early years, CoCom even prohibited exports of items like typewriters (for an overview of control lists, check the collection by Sam Weiss Evans). But when it came to truly strategic technologies, control evasions were simply overlooked in most cases. Despite the Bruchhausen Semiconductor Smuggling in 1977-1980 and VAX Supercomputer Diversion of 1983, which directly boosted Soviet missile and aerospace design, only one control evasion case led to a strong policy response from the US. That was the Toshiba-Kongsberg case, the main geopolitical tech scandal of the Cold War.

In the early 1980s, the Soviet KGB received computer numerical controls (CNCs) clearly restricted under CoCom from the Japanese Toshiba Machine and Itochu Corporation as well as the Norwegian state-owned Kongsberg Vaapenfabrik Company. The suppliers were even updating and fine-tuning software on the Soviet facilities between 1982 and 1984. For export license purposes, the Norwegian Trade Ministry listed the items as spare parts for a civilian facility, while Japan’s MITI (Ministry of International Trade and Industry) registered the machines as older models not included in CoCom lists. The result was twofold — Soviet submarines got reduced acoustic signatures, increasing their stealth against American anti-submarine warfare capabilities, while the White House went on a punitive campaign, imposing sanctions on the Norwegian supplier and fully banning imports of Toshiba products. These repercussions made the Toshiba-Kongsberg case unique. The harsh consequences were partly motivated by the rising narrative of tech competition with Japan. It was not only the USSR that the American security establishment was worried about, but also about winning the competition against the Japanese tech sector. But the shamelessness of this case also motivated consequences — no effort was made to send the equipment through third countries, like everyone else did.

Soviet techniques for bypassing CoCom1

Despite extensive multilateral coordination under CoCom, enforcement remained its Achilles heel — a challenge that persists in today’s semiconductor controls. As the Heritage Foundation lamented in 1983: «Terrible Fact. To be sure, Washington would not have to control the re-export of U.S. items from its allies if the allies actually were effectively controlling the re-exports. The problem is that they are not».

On the Soviet side, a critical structural problem of technology competition lay not in acquisition but in efficient diffusion and allocation of restricted computing resources — that is, what happened after restricted tech made it through the blockade. Soviet archives reveal a dysfunctional internal competition that severely limited the effectiveness of their technology transfer efforts. Various intelligence departments, ministries, and industrial enterprises routinely competed for the same technological products, often outbidding each other and creating artificial scarcity for domestic users. The same IBM computer would be simultaneously pursued by a truck manufacturing plant, military production facilities, the state tourism operator Intourist, and even the Soviet Olympic Committee. This fragmentation extended to collection channels themselves, with Soviet industrial ministries frequently requesting identical Western computers through multiple intelligence agencies. Such uncoordinated efforts not only wasted resources, but also critically impaired the USSR’s ability to strategically deploy compute.

This pattern of inefficient resource allocation has modern parallels, such as the inefficient stockpiling of GPUs by Chinese companies and local governments, as covered by ChinaTalk. Chinese policy conversations are increasingly focused on this, arguing that the US-led export control regime should be addressed through “Construction of a Unified National Market” 全国统一大市场建设, a policy aiming to coordinate provincial governments to gain leverage over foreign firms and eliminate local protectionism and administrative monopolies. While there are structural similarities to the supply-side restrictions of the CoCom era, China’s effort to streamline resource allocation means those same control strategies might not yield the same results today.

PRC≠USSR

China’s economic leverage creates a formidable counterweight to future export control efforts. Beijing possesses retaliatory capabilities along the semiconductor supply chain in a way the Soviets never did, especially through critical resources like rare earth minerals. More importantly, China actually has an export market. The Soviet Union’s technological autarky meant it never developed computing giants capable of competing globally. Archival evidence shows that Soviet-made computers rarely crossed even Eastern Bloc borders, let alone captured a share of Western markets. And when transistors first revolutionized telecommunications in the 1950s, CoCom had already been established, reinforcing America’s first-mover advantage. Today, the United States faces the much more complex challenge of building a coalition mid-race.

The second critical difference between the political era of CoCom and that of the 2020s lies in access to human capital. Throughout its technological competition with the United States, the USSR was severely constrained by its limited access to international talent, particularly the tacit knowledge transfer from American computer engineers. While exceptions like Joel Barr and Alfred Sarant enabled rare technological breakthroughs, such cases were anomalies rather than the norm. The Soviet Union could not practically implement anything like the Thousand Talents Program (千人计划). By contrast, Chinese recruitment efforts have successfully targeted specialists from companies like ASML and TSMC alongside top graduates from universities worldwide.

The current competition for global STEM talent contrasts sharply with the CoCom era, when the United States served as the unchallenged center of gravity for international technical expertise. In a world where knowledge transfer often proves more valuable than hardware access alone, human resources can play a substantial role in circumventing export restrictions. The US Foreign Direct Product Rule effectively restricts American talent, but any realistic export control framework must also incorporate talent retention.

Three lessons from CoCom

1. Effective technology denial requires multilateral enforcement.

There may be inherent limitations to a unilateral enforcement of the export control regime. Despite its structural challenges, CoCom at least regularly coordinated the leading technological and manufacturing powers in restricting access to critical technologies for the United States’ strategic adversary — an alignment that does not exist today. Both CSIS and Carnegie suggested that unilateral controls on frontier technologies are ineffective, a point made many times before, including by Eric Hirschhorn, the former head of BIS, who compared it to “damming half the river.”

2. Tracking and verification systems are a prerequisite to effective enforcement.

Where CoCom frequently failed was its inability to track controlled technologies once they left manufacturer facilities, creating enforcement gaps that were systematically exploited through transshipment, diversion, and falsified end-user declarations. Erich Grunewald and Michael Aird proposed a chip registry idea, noting that “a key problem for AI chip export enforcement is that BIS does not know where exported AI chips are.” Without visibility into the movement and end-use of restricted technologies, even the most comprehensive control frameworks will ultimately collapse under their own enforcement limitations. Implementing something like the chip registry would significantly increase the odds of technology denial achieving the desired effect.

3. CoCom operated under nearly ideal institutional conditions, and still could not be enforced perfectly.

CoCom was poised for success, operating under an ideologically bifurcated international system with established multilateral coordination structures and decades of institutional development. Despite these favorable conditions, CoCom still struggled. Today’s landscape offers none of these advantages to the US — the world today has incomparably more interconnected trade networks, fragmented alliance structures with competing economic interests, and a technologically sophisticated competitor deeply embedded in semiconductor supply chains. If CoCom’s results were mixed even under optimal containment conditions, expectations for current export control effectiveness should probably be tempered.

Is the glass half-full or half-empty?

CoCom suggests that export control regimes have always faced practical challenges. Perhaps the most significant challenge is reliable enforcement. To the extent that CoCom was successful, it was through the cooperation of other countries in a more multilateral regime. Today, deteriorating alliance cohesion and tariff wars could undermine the verification and enforcement mechanisms necessary for effective controls. As the intelligence leaks and scandals of the Cold War era show, crucial partners like South Korea and the Netherlands did not always have incentives to comply with the US-led tech regime. Today, these partners simultaneously host critical semiconductor infrastructure while also maintaining substantial economic ties with China. In fact, despite export restrictions, TSMC may have already manufactured millions of controlled high-end AI chips for Huawei.

What does this mean for US AI policy? First, policies that improve verification systems and bolster multilateral institution building would be very effective. Verification and compliance measures should address not only chip smuggling but also other potential circumvention methods such as the use and operation of data centers in countries like Malaysia, remote access to large amounts of compute through cloud services, use of TSMC manufacturing, and other avenues to skirt US export controls.

Second, as policymakers navigate the current “chip war,” they should recognize that export controls represent just one element of a comprehensive technology strategy — one that must be balanced against both economic interdependence and the reality that innovation often flourishes in response to constraints. The most sustainable technological advantage will likely come not from restriction alone, but from accelerating domestic innovation while selectively managing the most critical chokepoints in the AI supply chain.

We’re looking for part-time help with the podcast and newsletter! Spend ~10 hrs/week editing shows and putting together enhanced transcripts starting at $35/hr. All you need is enthusiasm for our content—no prior audio experience required (we use this software called Descript which is easy to learn and a pleasure to use).

Respond to this email if you’re interested with a little background on yourself.

This edition is brought to you by Pelanor, the AI-powered FinOps startup letting companies make sense of their cloud spend.

Founded by alumni of Israel’s elite cyber intelligence unit, Pelanor untangles the chaos behind nine-figure AWS and Azure bills. Most companies have no idea how their compute spend ties into business outcomes. That’s because the people managing infrastructure and the people using it rarely talk, let alone share ownership.

Pelanor solves this by building a dynamic graph of your cloud environment—tracking which microservices talk to which databases and how AI workloads are actually being used. Even non-technical team members can ask complex questions in plain English and get straight answers:

• “What’s driving our OpenAI bill?”

• “Which services are talking to unused databases?”

• “Where can we save?”

Reach out to founder Matan Mates on LinkedIn or email him directly at matan@pelanor.io.

As Meta aggressively consolidated the US market for augmented reality glasses, it scared off what could have been a competitive AR ecosystem in the West. Only now with Google’s recent AR announcements at I/O is a second serious player back in the game. In the words of Leap Motion and Midjourney founder David Holz:

“VR and AR really needed 12 companies basically making products… [I]n two generations of 12 companies, we would have been way closer to something that was really compelling for everyone. But instead, we got one product, maybe two.”

In China, however, there has been no such consolidation, and a “Hundred Lens War” (百镜大战) has instead produced a vibrant AR ecosystem where small startups, rather than tech giants, lead. But does it matter that there are no American analogues to China’s “Five Little Dragons” of augmented reality (AR眼镜五小龙)?

Skepticism toward AR glasses is understandable (especially after the highly publicized failure of Google Glass), but the premature consolidation of the US market could have dramatic consequences one day. Real-time translation via a wearable product could be game-changing in linguistically diverse places like India; some consumers could be compelled by the prospect of using AR glasses as a real-time conversation guide; and maybe hyper-immersive Wii sports will prove to be even more fun than the low-tech version we enjoy today. The point is, we can’t really sit here in 2025 and say with certainty that AR glasses won’t have any compelling use cases.

If manufacturers eventually overcome technical challenges and get the public on board, AR glasses could generate tons of real-world audio and video data valuable for training AI, much like humanoid robots. This route to profitability could help buoy AR investment, even if the public isn’t sold on AR products yet. That could also partly explain why Xi tried on some AR glasses during his recent tour of Shanghai’s AI ecosystem.

Today’s article will explore China’s market for AR glasses and introduce China’s Five Little AR Dragons. It will also include my personal review of a Chinese-made AR headset that I was able to try in Taipei.

Terminology and the State of the Art

Augmented reality (AR) is distinct from virtual reality (VR) — AR headsets are designed to overlay digital features onto the user’s interaction with the real world, and can theoretically be worn for extended periods. VR products are designed for shorter periods of fully immersive use (Apple’s Vision Pro, for example, does both VR and passthrough AR). The umbrella term for both AR and VR is extended reality, or XR for short.

Chinese XR glasses manufacturers rely on microdisplays, primarily micro-LED and micro-OLED displays. While traditional OLED displays use a backplane made of glass, the pixels of micro-OLED displays are etched directly onto a silicon wafer (hence their alternative name, OLED on Silicon, or OLEDoS). Thanks to supply chains created for the Apple Vision Pro, the cost of producing these displays has dropped rapidly in China since 2023. In March, Chinese LEDoS manufacturer JBD cinched a deal with Meta to become the exclusive supplier of LEDoS displays for Meta’s Orion prototype.

Another key component is the waveguide. A waveguide is a transparent optical component that guides light from the display to the user’s eyes while allowing them to see the real world. The waveguide makes it possible to overlay digital content onto physical environments. As is the case with microdisplays, China’s leading suppliers of waveguides are based in Shanghai.

Five Little Dragons

What can these glasses do today? Traditionally, the industry has emphasized entertainment features (e.g., gaming, streaming movies and music, and shooting photos and videos) while touting the potential future benefits in education, medical care, and delivery logistics.

But after Meta announced new AI features for their Ray-Ban smart glasses in April 2024, China’s AR companies have been eager to capitalize on the “AI+AR” hype. According to Li Hongwei 李宏伟, CEO of the AR dragon RayNeo:

“Smartphones do not have the display features or capabilities of spatial perception interaction that AR glasses have. AI+AR glasses encompass three categories: mobile phone applications, AI smart assistants, and virtual reality integration. The latter two are opportunities for disruptive innovation. For this reason, more than half of the successful companies in the future AR market may not be traditional giants, but emerging companies.”

Still, less than 20% of designs in the Chinese smart glasses market had AI functions by the end of 2024. AI integration is difficult in part due to dependence on external computing power — suitable processors are simply way too big and energy-hungry to fit into the frames of the glasses. That’s why AR glasses often rely on split compute to preserve battery life — tasks like translation are offloaded to the user’s cellphone, and then the result is transferred to the glasses through WiFi to save power. This also means that AR companies benefit substantially from locking down partnerships with phone companies that control the ecosystem.

China's VR/AR industry reportedly raised 2.948 billion RMB (~US$340 million) across 30 investment and financing events from January to November 2024. While tech giants like Huawei and Xiaomi have been dabbling in AR, innovation has primarily been driven by the so-called “Five Little Dragons” of the AR industry: XREAL (优奈柯恩), RayNeo (雷鸟创新), Rokid (灵伴科技), INMO (影目科技), and the Xi-endorsed Meizu (星纪魅族).

Thanks to fierce domestic competition, most models of AR glasses currently available in China are in the price range of 2,000-4,000 RMB (~US$275-$550). By comparison, Meta’s Ray Ban glasses, which don’t include displays in the lenses, start at US$299, and their full-service Orion prototypes cost US$10,000 per unit to produce.

We’ll briefly highlight each company’s quirks below.

XREAL 优奈柯恩

XREAL was founded in Hangzhou by Xu Chi 徐驰, but the company has recently relocated its headquarters to Shanghai. While other AR startups primarily use Snapdragon processors made by Qualcomm, XREAL uniquely uses self-developed chips. XREAL’s X1 processors can reportedly achieve a latency of 3 milliseconds (compared to 12 milliseconds for the Apple Vision Pro). Xu explains:

“The X1 chip equipped in the XREAL One not only successfully resolves technical challenges like 3DoF spatial anchoring and ultra-low latency, but also fundamentally overcomes the longstanding issue of inconsistent cross-device experiences for AR glasses. In the past, to provide a consistent experience across different operating systems, we had to develop separate software for Windows, macOS, Android, and iOS platforms. Yet even then, it was difficult to fully ensure uniformity in the user experience. With XREAL One, however, we’ve finally found a more elegant solution. By integrating computing power directly into the glasses themselves, we’ve fundamentally changed the game — delivering a truly meaningful “AR for all” experience.”

Xu also noted that 65% of XREAL’s smart glasses components are developed and manufactured in-house, and the company is aiming for 100% hardware independence within the next two years. But this hardware-focused approach has tradeoffs — XREAL’s products don’t currently come equipped with out-of-the-box AI features, although the company claims that LLM integration will be included in a future software update. Here’s Xu’s justification for not engaging with the AI+AR hype:

“AR glasses are always centered around user experience. We won’t blindly build AI glasses just for the sake of having AI. Instead, we start from the user experience and work backwards to identify what kinds of technical innovation are necessary.

In the R&D process, we have to wisely understand the limits of current technology — knowing what’s realistically achievable and what isn’t, at least for now. Blindly chasing breakthroughs can lead to disastrous outcomes.

Although XREAL is a relatively small company, our strength lies in the smart use of limited resources. We’re very clear about where to allocate funds to keep our product logic sharp and focused.”

Consumer reviews of the XREAL glasses have been quite positive despite the lack of AR+AI branding. One reviewer from Hong Kong wrote in April of 2025:

These are the best AR smart glasses, perfect for entertainment whether you're out and about or relaxing at home (such as in a small apartment or lying in bed). Once connected to a compatible device via USB-C, users can watch their favorite shows, movies, or games on a large virtual screen, like having a private cinema.

The glasses feature a Full HD 120Hz OLED display with a brightness of up to 600 nits, delivering vivid images and excellent contrast for viewing dark scenes. The image quality is further enhanced by electrochromic dimming lenses, allowing users to adjust the lens brightness with a switch. At the clearest setting, the lenses let users easily see their surroundings, while at the darkest setting, they effectively block out external light, providing an almost perfect viewing experience.

In addition, the glasses boast impressive speakers, with sound quality tuned by Bose, offering rich audio across highs, mids, and lows. While using separate headphones can still enhance the audio experience, this is the first pair of smart glasses where users feel that headphones are an optional accessory rather than a necessity.

XREAL has also prioritized business partnerships under Xu, betting that the experience will translate to market share once AR technology is mature enough to garner widespread consumer interest. XREAL has inked deals with BMW, T-Mobile, AT&T, Bose, and Google.

RayNeo 雷鸟创新

RayNeo is a subsidiary of TCL, an electronics manufacturing giant that is partially state-owned. The company’s Mandarin name translates literally to “Thunderbird Innovations,” and they are notable for marketing AR products aggressively outside of China. CEO Li Hongwei 李宏伟 explains in an interview from January 2024:

“We wanted to establish a foothold in overseas markets first, so we started by working on distribution channels. For example, we launched on Amazon, and in November last year, we achieved strong results by ranking first on both the new arrivals chart and the bestsellers list in the smart glasses category.

Right now, we’re also selling our products in some boutique stores across Europe… and the sales performance there has been fairly good as well.”

RayNeo recently announced a partnership with the International Olympic Committee, so we’ll probably see promotional videos for Los Angeles 2028 shot from the perspective of athletes.

The company also partnered with Alibaba Cloud to develop a multimodal AI model specifically for AR glasses — now, the average response time for AI queries on RayNeo glasses is reportedly 1.3 seconds.

Finally, RayNeo was the first company in the Chinese market to sell AR glasses for less than 2,000 RMB. Here’s how they keep costs low, according to Li:

“If Meta’s Orion is the Vision Pro of AR glasses, then the RayNeo X3 Pro is more like the “Vision.” …It does not use the most cutting-edge technology in all technical indicators like Orion…. For example, in the selection of optical waveguides, although RayNeo has a silicon carbide wafer version internally, our commercial products do not use this material. Instead… we use photolithography machines and chip etching processes to make waveguides on glass, so as to better balance costs and product experience. As the company's strategy, RayNeo will not choose to pile up all the industry's most advanced technologies in listed products for the sake of showing off its skills, but will develop cutting-edge technologies and products internally, and eventually launch leading, pragmatic, and mass-producible products to the market.”

Rokid 灵伴科技

Rokid was founded in Hangzhou in 2014 by Zhu Mingming 祝铭明, who left his job at Alibaba to start the company. In 2024, Rokid received nearly 500 million RMB and 100 million RMB across two rounds of financing. Their AI-equipped glasses use Alibaba’s Qwen models for real-time translation.

Among the AR dragons, a uniquely large portion of Rokid’s investors are SOEs. Here’s Zhu’s explanation from an April 2025 interview:

We have significant influence on the B2B side — the cultural and museum market alone is worth around a hundred million yuan annually. Our ecosystem partners bring in tens of millions, and each year, several million people visit museums. Rokid is the only player in this sector. …

B2B operations serve as a bridge for interacting with the government. Even in local governments with no overlapping business opportunities, there are always departments for cultural tourism and museums.

Rokid partnered with the PLA to build custom AR glasses for use on China’s space station. Their website also highlights B2B partnerships with oil, gas, and mining operations. But for now, Rokid’s consumer products receive mixed reviews.

I was able to try the Rokid Max AR glasses at the Guanghua Digital Plaza in Taipei. I watched a clip from Avengers: Endgame with the glasses, and the image quality was quite nice. You can correct for nearsightedness without customized lenses if your prescription is between 0.00D and -6.00D. I expected my eyes to feel strange switching between far-away objects and the close-up digital projection (as is reported by many reviewers), but I didn’t notice any such feeling during short-term use.

Unfortunately, they are very ugly and sit weirdly far and high up on your face. They’re also not wireless (unless you buy a wireless adapter, which adds latency), and the Rokid Max glasses I tried don’t support AI features like translation (Rokid’s AI-equipped glasses are a separate product line).

My impression is that, for now, the most tangible use case for these glasses is avoiding neck strain.

INMO 影目科技

INMO’s Chinese name could be translated as “Image Eye Technologies.” As a newer entrant into the Chinese AR market, their strategy has focused on affordable consumer AR with basic features, advertising use cases like cycling directions, translation, and taking notes in meetings. The company marketed their INMO GO glasses as “the first consumer-grade AR glasses with deep AI integration,” highlighting translation capabilities in 11 languages, smart notifications, and AI assistant features supported by ChatGPT. Here’s INMO founder and CEO Yang Longsheng 杨龙昇 on his vision for popularizing AR glasses:

“We hope that AI glasses can be like a personal assistant in the future, helping me order takeout, order a cup of coffee, etc. at any time. …

In the past, social interaction was generally between real people, but based on the emergence of AI and the improvement of technologies such as virtual humans, I believe that in the near future, perhaps within four or five years, people will be able to socialize with these virtual intelligent entities.

I can create a virtual image of my dreams, infuse it with the personality I want, and then interact with it in this entire virtual-real world. …

[AI companies] also need to find some landing points for these intelligent entities. Glasses are undoubtedly the best form at present, which also encourages them to try more content on glasses.”

INMO’s content partners include Baidu (China’s Google analogue), TanTan (a dating app), and game developers NetEase and 37Games. Chinese reviewers seem to appreciate the INMO software ecosystem — one user wrote,

“The INMO AIR 3 uses the IMOS 3.0 operating system, an OS specifically designed for AI+AR terminals. The AIR 3 can project the equivalent of a 150-inch giant screen. With the help of a 3DoF smart ring, the AIR 3 supports screen-space hovering functionality. IMOS 3.0 provides more efficient applications and a more immersive experience through intelligent interaction and spatial display capabilities. IMOS 3.0 not only supports native AR applications but is also compatible with most Android apps on the market, offering a relatively rich content ecosystem.”

INMO also recently announced a partnership with China Mobile to integrate their AR products with China Mobile’s Jiutian LLM ecosystem.

Meizu 星纪魅族

Meizu is owned by Geely, a conglomerate that primarily manufactures automobiles and holds a controlling stake in Volvo. Meizu’s primary business ventures are smartphone manufacturing and developing the FlyMe Auto operating system used by many Geely-owned car brands. As such, Meizu’s AR glasses are compatible with FlyMe-equipped vehicles.

At the 2025 Shanghai Global Investment Promotion Conference, Meizu founder Li Shufu 李书福gave a highly-publicized speech while wearing a pair of Meizu glasses, which acted as a teleprompter.

The company has developed its own FlyMe LLM, but Meizu glasses also support integration with third-party AI models, including DeepSeek, Qwen, and ByteDance’s Doubao. In early 2024, the company signed a partnership with Malaysia's Juwei Group to expand sales of Meizu glasses in Southeast Asian markets.

Looking Forward

China’s “Hundred Lens War” is a live experiment in hardware innovation under pressure. While consumer interest remains tepid, China’s Five Little Dragons are constantly launching new products in search of the ideal combination of design choices — wired vs wireless, with lens displays vs without, AI vs traditional translation, and many more. If AR glasses eventually succeed in locking down compelling use cases, it’s likely just as likely their decisive breakthroughs will come from not from Silicon Valley but Shanghai.

Thanks to Mike G. and Benjamin Reinhardt for offering feedback on previous drafts of this article.

Can studying Mao Zedong help explain Donald Trump?

To find out, ChinaTalk interviewed the legendary sinologist Orville Schell, who visited China during the Cultural Revolution and is currently at the Asia Society.

We discuss…

• Mao Zedong’s psychology and political style,

• Similarities and differences between Mao and Trump,

• How Mao-era traumas reverberate in modern China, including how the Cultural Revolution has influenced the Xi family,

• How Zhou Enlai and Deng Xiaoping survived the Cultural Revolution, and which of their tactics could be useful in modern America,

• What civil society can do to defend democracy over the next four years.

Co-hosting is Alexander Boyd, associate editor at China Books Review and former ChinaTalk intern.

Listen on Spotify, iTunes, or your favorite podcast app.

Culture War x Cultural Revolution

Alexander Boyd: Orville, you wrote the most prominent essay on Trump’s Cultural Revolution, comparing Donald Trump’s behavior in office and personal style to Mao Zedong. To start, what is Maoism, and how would you describe Mao Zedong?

Orville Schell: Mao Zedong, of course, was the great progenitor of the Chinese Communist Revolution. He was a Marxist-Leninist, and he liked control as any autocrat would. However, one of the hallmarks of Mao Zedong was also an abiding interest in throwing things off balance as a way to gain even greater power.

In this regard, he became a great fan of Sun Wukong. This golden-haired monkey was one of the heroes of the classic Chinese novel, Journey to the West, which tells the story of Buddhist scriptures being brought from India to China. One of the most prominent features of this monkey king was his love of disorder. His sort of watchword was “dànào-tiāngōng” (大鬧天宮), to make great disorder under heaven. Mao Zedong actually ended one of his poems with that line, and he always loved this novel, Journey to the West.

When the Cultural Revolution arrived, I think this was a real consummate expression of Mao’s affection for chaos. He did feel that not only did Chinese society need to be upturned, but the whole political structure of China needed to be upturned. Everything in effect needed to be “fānshēn" (翻身), as he said, “turned over.” He adopted many expressions similar to the Monkey King that expressed his love of contradiction and disorder. Class struggle, of course, which became the hallmark of the Cultural Revolution, was a form of deep and disturbing disorder.

Alexander Boyd: Where did Mao’s need for constant chaos and rebellion come from? In your Project Syndicate essay, you posit that it came from his troubled relationship with his father. Can you describe the parallels to the case of Donald Trump, who had a famously domineering father himself?

Orville Schell: The one place where we really get chapter and verse on Mao’s relationship to his father growing up in Hunan at the end of the 19th century was in Red Star over China by Edgar Snow. Mao told Snow that he had a very adversarial relationship with his father. He even said he hated his father, that his father was a tyrant, and they were constantly battling. On a number of occasions, Mao Zedong actually ran away from his home.

He did have a very sort of Buddhist-inclined, loving mother who made a lot of difference. But the relationship with his father clearly set off the notion of the world as an adversarial place. He tells Snow that he learned only by standing up to his father could he survive, and then his father would come to heel in some sense and not just overwhelm him with his sort of tyrannical paternal role.

That sort of characterizes Mao, and in fact Trump too, as we learn from his niece (the daughter of his older brother), who is a psychologist. Trump had a father who was very preemptive, very tyrannical and very judgmental.

Wallowing off into the bog of psychobabble here, any human being who reads literature knows that a father’s relationship to his son and vice versa is a profoundly influential relationship. As a young man grows up, that relationship forms him.

Alexander Boyd: We know that Mao loved to speak in allegory. He often would speak about one thing but mean another. How much can we believe his stories to Edgar Snow? And is it just a metaphor for American paternalism?

Orville Schell: I think at the period when Edgar Snow was doing these interviews with Mao was in the 1930s. We hadn’t started all the rectification movements and Mao had not been in power with all his mass campaigns. He had just arrived in Shaanxi province. I think that we get a pretty unalloyed representation of his early years. I don’t see there’d be any reason for him to be setting traps or deceiving Snow.

Throughout his life, Mao easily took umbrage at things when he felt sat upon or disrespected. Another famous example, of course, is when he finally went to Beijing in the late 1920s and he became a senior intern of some sort at the Library in Beijing University. He used to feel incredibly disdained by Chinese intellectuals who’d come in and sneer at his hickey Hunan accent.

If you‘ve ever heard Mao Zedong speak or seen a film in which he speaks, he’s almost unintelligible to Mandarin speakers. Like most Chinese people of his generation, including Deng Xiaoping and Chiang Kai Shek, he preferred to speak his local dialect.

This laid the track for Mao’s antipathy towards intellectuals, just like Trump hates intellectuals and universities. He thinks they’re arrogant, they’re elitist, they look down on people like him and down on working people, et cetera. Mao felt very much that no matter what he did in his formative years, he was disdained and disrespected by the intellectual class.

Of course, in that sense, I think he is a metaphor for sort of the whole ‘kultur’ of China in a sense that it is aggrieved, it’s been humiliated, it has been looked down on, kicked around, exploited, you name it.

Alexander Boyd: Another similarity between Trump and Mao is that both men had a lot of wives, often tumultuous ends to those relationships with women, and a propensity to start new relationships before the last one ended.

Yet another parallel is Mao’s exposure to Hunan secret societies, specifically Gēlǎohuì (哥老會), the Red Gang, and his time as an organizer at Anyuan. This mimics Trump’s early mob ties, especially in the New York of John Gotti and then Rudy Giuliani, who was a big mob fighter back in his day.

Their styles of governance are also similar — Mao made frequent trips outside of Beijing, and he loved to launch campaigns while on the move, much like Trump’s Mar-a-Lago golf club trips.

Are these comparisons substantial, or is it just the case that when you have so much information about two people, you can always draw connections between them?

Orville Schell: It’s true that both Mao and Trump had a lot of wives and a lot of ladies. What that suggests is that they need to have that kind of affirmation and signs that they can beguile people and win people over, which speaks to me of a fundamental lack of self-confidence. Both of them derived a certain measure of prowess from their ability to attract beautiful women.

Trump still talks about women as objects and as adornments. Mao clearly was the same. There were other leaders in China who were not. Chiang Kai-shek had a lot of ladies (and ladies of the night as well in his youth), but after he got married, he was quite faithful. I don’t quite know what to make of Zhou Enlai, but he had one wife. Whether he was gay or not, it’s a question people do care about.

Mao was somewhat special, and he arrogated that special right to poach on women himself. But when revolution came home to roost after 1949, it was not something he found acceptable in ordinary people or even in his acolytes. He was very puritan. But he was not just like Trump. Trump purports to be a Christian and yet doesn’t abide in any meaningful way by the notion of loyalty implicit in marriage.

Jordan Schneider: Shall we discuss the deep state example?

Alexander Boyd: Trump famously vowed to drain the swamp and railed against a deep state that he perceives as having both frustrated his attempts to exercise power in his first term and prevented him from regaining office in 2020. Now he’s engaged in a campaign of revenge against all those purported deep state agents.

Many people, and you chief among them, have made that comparison to the Cultural Revolution where a frustrated and suspicious Mao unleashed these animal forces within China to take down a party in a state leadership that he felt was shackling his own ambitions to remake China. Is that an accurate comparison in your mind?

Orville Schell: One of the hallmarks of Mao’s revolution was a sense that somehow the party (which he himself had helped build) and the state (which was the handmaiden of the party) were ultimately the refuge of rivals.

He had a great antipathy against bureaucratism. This also speaks of his love of disorder as a creative force. When he started the Cultural Revolution, one of the first things he did was to issue a wall poster that said “bombard the headquarters” (pào dǎ sīlìngbù, 炮打司令部). What he meant by that was that he felt the party had become ossified. It had become the refuge of bureaucrats who were living high on the hog but didn‘t want to make revolution anymore and found class warfare too disruptive.

He felt that it was time to destroy it. How did he do that? He gave permission to young people, idealistic young people, to attack. Almost all of the leaders who were potential adversaries of Mao either died or were purged. Xi Jinping’s father was one of them, a vice premier who had a very bitter purging.

In that regard, Mao Zedong took, I think, great satisfaction in overturning even his close revolutionary comrades. Here Trump is not too dissimilar. He seems to be invigorated by the idea of destroying institutions that he views as refuges of people who might be against him and of firing people he’s hired, turning against people who he perceives as potential rivals. He demands complete fealty and loyalty or you’re in trouble.

Here too, I think there’s a kind of a similarity in the way the two relate to other leaders and to institutions of government, the deep state. It’s the equivalent for Trump of what the party and the state that he himself helped build were in China. He saw them as slowing down his revolution, as harboring his adversaries and as being overly bureaucratic, what he called “sugar-coated bullets of the bourgeoisie” (资产阶级的糖衣炮弹). In other words, they’d given up their revolutionary vigor in favor of staid bureaucratic forms of government.

Alexander Boyd: So if there’s an attack on the bureaucracy, does that make Elon Musk and DOGE a new Kuai Dafu 蒯大富 and the Red Guards?

Orville Schell: Musk is older. Kuai Dafu was one of the earliest Red Guard leaders at Tsinghua University when Mao issued his order to bombard the headquarters.

Elon Musk should know better, but I think he too has a kind of innate impulse that chaos is a creative element. It’s one step away from the Silicon Valley mantra, “failure is positive.” But I think he does share with Trump this idea that somehow you need to clean out the Augean stables of the government. I don’t know why Musk might feel disrespected or disdained when he’s been so successful and the richest man in the world.

Here, too, I think we have to remember that these leaders are human beings. They’re not just rational creatures who look at the national interest of the country or read reports to make rational decisions. Some leaders are completely crazy.   We know from Euripides and Shakespeare that leaders are completely crazy and they do astounding things.

You can read Stephen Greenblatt’s book, Tyrant: Shakespeare on Politics, which is about six great plays that Shakespeare wrote about tyrannical leaders. There’s nothing new here. It’s just that policy people and I think many academics are loath to recognize that we’re also dealing with something very human here, namely, leaders with deep and tragic flaws, which in Euripides are hubris, arrogance, and overreach.

When Croesus — we‘ve derived the expression “rich as Croesus” — went to the oracle of Delphi, he governed the state of Lydia, about whether he should go to war. He got the diction back from the oracle that if he went to war, he’d lose his kingdom. What did he do? He went to war and lost his kingdom.

There’s a little of that going on here. I think we need to factor the human dimension into the equation of understanding big leaders like Putin, Orban, Xi Jinping and others. Look back at their formative years.

Here, I highly recommend a wonderful new book that’s coming out by Joseph Torigan on Xi Jinping’s father, Xi Zhongxun 习仲勋. You see what nightmarish experience Xi Jinping went through as the son of a man who was twice purged during Xi Jinping’s teenage years, and was purged the second time as a counter-historical, counter-revolutionary. And what travails Xi Jinping as a young man went through to have a father who was in the five black categories.

This is a little bit beyond the mandate of so-called China socialist. But I do think that here’s where literature, drama, some of these other representations of leadership help us understand what’s going on.

Alexander Boyd: Let’s stay on Xi Jinping here for a second and Xi Zhongxun as well. What is Xi Jinping’s view on the Cultural Revolution today? Obviously it’s opaque, but I agree Torigan’s book is incredible and I read his section on Xi Jinping’s Cultural Revolution.

Orville Schell: I think it’s unfair to say that Xi Jinping is like Trump. Xi Jinping does not like disorder. He does not want to create great disorder under heaven, unlike Mao. The part of Mao he does bond with and did grow up with and appreciate is the Leninist part — the organized state, organized party control, autocracy. But he has no fascination for the part which we’ve been talking about.

This is why I suggested if Xi Jinping wants to come to some better understanding of who Trump is, ironically he has a homegrown example in Mao Zedong. He lived through it. He knows what people like that can do to a society and to even the global order. I think maybe he has thought this way. I don’t know, but I think he uniquely, unlike many Americans who’ve never been through this kind of disturbance, shouldn’t be so surprised by Trump.

Why do we think as Americans, when Italy had Mussolini and Germany had Hitler and Russia had Stalin and Spain had Franco, Salazar in Portugal and on and on, why do we think that we are somehow immune from these kinds of aberrant, overreaching, arrogant, and finally incredibly destructive leaders?

Jordan Schneider: China from 1949-1967 was a very different place with a very different governance system than America circa 2016 or 2025. Shall we discuss some of the differences here?

Alexander Boyd: Well, I think the first place that we should start is rise to power. Trump, you know, for better or for worse, won two elections, and Mao won power through civil war and various other means. Orville, what would you say is the biggest difference between Trump and Mao?

Orville Schell: Trump is more like Hitler, who came to power by being elected Chancellor of the Reichstag, whereas, as you point out, Mao Zedong came to power through insurgency and civil war.

Obviously, you can‘t completely compare these people, but I think in trying to understand the leaders of the present, it does behoove us to look back at leaders in the past who also created disorder, one kind or another — a world war, economic crisis, a revolution, whatever. That might help us understand a little bit what it is they’re offended by. What do they want? What would propitiate them? How do you deal with them? Can you deal with them?

I don‘t mean to compare Xi Jinping with Trump. But only to say that China’s historical experience of having undergone probably the most tectonic, catastrophic, and destructive revolution in history might help Xi understand what animates a leader like Trump and how best to deal with it.

Jordan Schneider: How best to deal with Mao is kind of not something people really figured out. Can you talk about what the antibodies were over the course of his reign, and highlight some examples of successful and unsuccessful pushback against his craziest ideas?

Orville Schell: Well, of course, the best, the biggest antibody of all to Mao was death. Many autocrats are very disruptive. Hitler died in his bunker, Stalin died, and wildly began to change. I don’t know what the antidote to Mao at that period would be, but I will say this, that if we look at our own country, there is a hint of similarity among the way people come to heel when big leader culture gets rolling.

The Republican Party is completely supine now. We do have the Democratic Party still raging against the storm. But one of the lessons I think that is quite striking about the Chinese communist revolutionary period was the way in which everybody finally was neutered. Those few who did speak out, and there were some, had very bitter ends. All know what happened. This is another hallmark of powerful and effective autocratic leaders is that they manage by one way or another — and one way is disorder — to intimidate people, frighten people into submission and silence.

Alexander Boyd: I think in a lot of histories of the 1970s though, everyone points to 1971 and Lin Biao’s death.

Jordan Schneider: Why don’t we do the Great Leap Forward and response to that? Because Mao had to do with self-criticism, right? This was a real brushback moment for him where after killing eight figures of his own citizens, there did end up being some pushback from the top that forced him down a policy path he wasn’t really excited to take.

Orville Schell: After the Great Leap Forward, many leaders like Deng Xiaoping and Zhou Enlai and Peng Dehuai felt it was too excessive. Forty million people dead, starvation, agriculture in ruins. They did for a brief period of time prevail. What was Mao’s answer?

Mao’s answer was the Socialist Education Campaign, which is a prelude to the Cultural Revolution, which ended up labeling people like Peng Dehuai at the Lushan conference, put out of business very quickly, ended up in jail. The president of China, Liu Shaoqi, Deng Xiaoping, were sent down. In other words, almost all the leaders, veteran revolutionaries that had accompanied Mao in the Long March in the Yan’an period, ended up in the doghouse or dead.

That was China’s experience. That made it very difficult for any voices of dissent to find any purchase. I remember being there myself during the end of the Cultural Revolution and just thinking, well, this is the way it is. There were no voices. They wouldn’t even talk to you as a foreigner. Nobody had permission even to interact in a normal human way with anybody foreign or an outsider because they were afraid. Mao had brought complete submission down onto society.

That didn‘t mean he suffocated all of the impulses that had built up over previous decades — they remained latent and dormant and they arose again when Mao died.

Alexander Boyd: Basically the biggest argument against Trump’s effort to remake America in his image, to bring manufacturing onshore, is that Mao, with more power at his disposal, more party at his disposal, a whole society cowed, actually failed. Even more recent scholarship, like Odd Arne Westad and Chen Jian’s new book The Great Transformation: China’s Road from Revolution to Reform argues that opening and reform really began in the 1970s. If Trump is Mao in this comparison, is the Trumpist effort, you know, this great “cultural revolution” in America, is it doomed to fail right from the beginning?

Orville Schell: Well, I think we could be doomed to something even worse. Mao failed in the Great Leap Forward. Then in order to regain and maintain power, he brought on the Great Proletarian Cultural Revolution. It’s hard to know. We can’t predict history and we don’t know whether American democracy will survive.

What we want to say is that we know this archetype of leader and whether it’s a communist, a Leninist, a so-called democrat, Nazi, fascist, whatever. That’s why I wrote some of these essays, to just remind people that there are examples throughout history and literature of these kinds of people. We need to better understand because now it’s America’s turn to have one.

Jordan Schneider: I think another one of the big differences that Tanner Greer pointed out on a show earlier where I tried to compare Trump to Hitler was the ideological commitment level that Trump has versus someone like a Mao or a Hitler who deeply believed in his bones in class struggle or Lebensraum or international Jewish conspiracy. Yes, Trump’s got some views on trade policy, but he lifted them after the bond market changed. It’s the sort of the level of focus which he can bring or has shown, the level of focus that he’s shown he can bring to policy stuff versus some of the other sort of totalitarians that you reference who maybe have some personality traits in common, but also kind of have a real agenda behind them. Whereas our current president, not so much.

Orville Schell: I agree with that, Jordan. Mao had a very highly evolved ideological agenda and analysis of how the world worked, where history was going, knowing what the dynamics would be. They did hew to that in various ways, sometimes rather opportunistically or in a utilitarian fashion. That’s communism.

But fascism is a very different animal. If you read Robert Paxton’s book on fascism, you begin to understand people like Mussolini. He had no ideology. He was sort of inventing himself as he went along. I think that’s one of the big differences between Trump and Mao. Mao was a very intelligent man and actually a good writer, good poet. He thought.

I don’t think Trump thinks — he acts almost like an animal. He feels this today and he acts, he responds. He certainly has no ideology or no sort of political commitments to principles that guide him in what he does. It’s more what he feels like doing. He feels someone doesn’t like him. He feels threatened. Whatever. It’s almost animal-like in his responses, which are completely irrational.

Alexander Boyd: This is actually not so much a continuation of this question, but it’s a different tack. I’m curious about Mao’s foreign policy and Trump’s foreign policy. A curious similarity is that Mao stated, “I like rightists.” He met with Richard Nixon. He found them easy to deal with in general, perhaps because he understood the ideological motivations of, or he perceived himself to understand the ideological motivations of capitalist right-wing politicians.

Then, Trump himself meeting today in the Oval Office with Carney, the Prime Minister of Canada, and earlier saying something like “I like the left” in reference to him not wanting Pierre Poilievre to win in Canada. Why do you think that Mao liked rightists or claimed to like rightists? What sort of insights does that give us into Trump’s foreign policy?

Orville Schell: In many ways, Mao could be a rightist, but I don’t think he liked anyone who opposed him. He viewed the right as opposing him and as opposing his ability to control thinking and ideas. This is why you get the whole idea of sīxiǎng gǎizào (思想改造), thought reform, that there’s a correct way to think and Mao helps limit that, describe that, lay the boundaries for that out. If you don’t want to think that way, then you’re a rightist or maybe even a leftist and then you deserve to be defenestrated.

I think Mao says he likes rightists because he thinks they’re practical and he can deal with them. It’s a bit of a capo-to-capo business. You can deal with a thug, even if you call him a rightist, but I don’t think he had any affection for rightists or leftists. These are categories of convenience into which he put people when he needed to get them off the stage.

Jordan Schneider: Yeah, I feel like the capo-to-capo thing is it’s less him liking Mark Carney, which I just truly do not believe, and more the ease and excitement where he gets talking to big, powerful, authoritarian leaders, as opposed to democratically elected ones.

Orville Schell: I think that, Jordan, if I may say, when Nixon and Kissinger went to China in ’72 and ’71 to set it up, I think there was a certain thrill for Zhou Enlai and Mao to have these people come hat in hand to Beijing to talk to them. Because remember that even though they were opposed to imperialism, colonialism, capitalism, America, et cetera, there is, I think, in my experience, at least amongst Chinese leaders representing their country, a deep and abiding wish to be respected.

They speak about that all the time, mutual respect and understanding, as if to say, “Can’t you just respect us as a dictatorship? Show us some respect. We’ve got a good economy, we’ve done a lot of amazing things,” which is true. Again, autocrats, one of the things that really riles them is that they’re cast out, they’re disdained, not considered proper company for liberal democratic states.

When China is the aggrieved, humiliated, kicked-around sick man of Asia, they want nothing more than to be respected. That is a complete contradiction because if you don’t act respectably, it’s very hard to be respected. Yet that is something they deeply crave, even though they would never acknowledge it.

Jordan Schneider: Yeah, I think it kind of works the other way as well, where you just see how big Trump’s smile was in his 2017 visit to China.

Orville Schell: I’ve been on several presidential trips, like for instance, with Clinton, which was completely different, very informal and open and cheerful. But Trump’s trip, it was all about pomp, circumstance, awe, ceremony and ritual. Both leaders were trying to impress each other and Xi Jinping wanted to impress Trump. You remember he took him to the Forbidden City and they had a banquet and all the rest of it.

But there were no moments of bonhomie, of personal smiling and back-slapping and saying, “Hey, we’ll work this out.” No, it was all about sort of who is the bigger dog with the most impressive marching band and most impressive Great Hall of the People. I think that’s very characteristic of both Trump and Xi. Trump wants to have a parade, just like Xi gets to have parades with a lot of tanks and missiles going by.

There’s an element of similarity, I think, of both deeply insecure men fundamentally, and nothing like a good parade to puff up insecure leaders’ egos.

Jordan Schneider: Can you talk about the sort of red versus expert dilemma, which we’re seeing play out with Laura Loomer in this administration?

Orville Schell: Remember that during the Cultural Revolution there were many, many struggles going on. One of them was the struggle between being Red and an expert. Experts, of course, were people who knew how to do something. They were the intellectuals, they were the scientists, the technicians, people who ran institutions and could be accused of bureaucracy.

The Reds were the people who wholeheartedly embraced Mao Zedong and were dedicated to overthrowing institutions of the experts. In that period, not only was expertise disdained and diminished in standing as a societal avocation, it was a hallmark of those first people being insubordinate. It was a different kind of loyalty — not to Mao, not to Maoism and Marxism-Leninism, but to rationalism, to scientific experiment, or these other things that had a kind of a logic that defied complete and total loyalty to the Great Helmsman, whatever he represented at the time.

Whether the party was intact or not during the Cultural Revolution, it was not loyalty to Mao and the revolution versus loyalty to whatever other thing you might be — a scientist, a businessman loyal to profit, and policymakers who are loyal to trying to figure out the national interest. That was a huge divide.

We see that now with Trump. What he wants is not people who know something in the FDA or the FAA. He wants people who are loyal. So you got Laura Loomer, like Kang Sheng or Chen Boda or someone running around firing people and putting people in prison for Mao.

Jordan Schneider: What was the upside for Mao of getting rid of the experts?

Orville Schell: Liu Binyan, the great writer of the 1980s, wrote a book called A Higher Kind of Loyalty. This was an analysis of people who felt a loyalty, whether it was through religion, technology, science, intellect, or just ideals, to something other than the revolution and the leader.

What Mao rejected and recoiled from was professionals who were experts, who said, “No, this revolution does not make sense economically, scientifically or in any other way. It’s mayhem.” That put them immediately on the enemies list. That’s why intellectuals, and they categorized them into many different categories, were pilloried because they couldn’t be totally loyal as religious people were, because religious people owe a different loyalty to their God and to their principles and to their morality, not to the leader.

Mao couldn’t stand that, so he waged war against those people. We see sort of, I think, hints of that happening now in America where people like Fauci in the first administration of Trump were not respected at all. He was a very good scientist, very devoted public servant. We see attacks on vaccines, Robert F. Kennedy Jr., science has proven measles vaccines work. We see a lot of things like this looming up again where what’s important is not logic, but it’s loyalty to the leader.

Alexander Boyd: But Mao also resented his need for these bureaucrats, for these technical experts. And chief among them was his own Confucius, Zhou Enlai. Even though he was fighting against these empiricists and scientists, cultural leaders and everything during the Cultural Revolution, probably the number one empiricist of all was right next to him. Although at the same time many Chinese spoke about him like we spoke about John Kelly in the first term. “Oh, he’ll be able to restrain Mao’s worst instincts.” How do you read Zhou Enlai?

Orville Schell: Zhou Enlai was a restraining influence. He was also a complete factotum. When you see what he was put through, like in the rectification campaign in 1942 where he sat before the Politburo for five days, wrote self-criticism, self-immolations, humiliating, pusillanimous kind. He paid a bitter price as a human being to keep in Mao’s good-enough graces. Still, Mao endlessly tormented him and humiliated him — very smart man. And he took it. Why? That’s an interesting question.

Did he think he was doing good like Matt Pottinger in the first administration of Trump, that if he just kept his head down and tried to do a good job, he could restrain the leader? But there was also probably a lust for power, an urge towards power which kept him there. He once got that needle in his arm, it was very hard to get out or he’d end up like his friend Liu Shaoqi — imprisoned or dead.

These kinds of leaders demand not one-time propitiation declarations of fealty and loyalty, but continuous. The leader keeps ramping up the ask. If the lieutenant wants to stay in their graces, they have to keep becoming more and more genuflective. We see an awful lot of people left the first administration of Trump and now he’s already lost all the people he’s lost. I mean, Rubio, everybody. He’s taken over positions of Waltz and others who drop like flies because it’s very difficult to satisfy the demands of autocrats who require 200% loyalty.

Alexander Boyd: Is part of this the “Coalitions of the Weak” thesis put out by Victor Shih about how Mao would often rehabilitate disgraced cadres? You saw that with Deng Xiaoping, you saw that with Zhou Enlai, he’d constantly send people down, bring them back, criticize them, humiliate them, purge them, restore them, and it ended up necessitating their loyalty. They created psychological dependence, but also political dependence.

Especially with Victor Shih, he’s talking about the Fourth Front Army in the Long March, I believe Zhang Guotao. Is that a similar coalition with Trump and Mao? And how did Mao’s coalition of the weak work? And is that an effective governance tool? We might be skeptical of it, but Rubio, who has no basis left in the GOP, has basically been entirely kneecapped, was humiliated in his run — he’s Zhou Enlai exactly. But he could also be our Deng Xiaoping.

Orville Schell: If you want some good reading, go back and read Deng Xiaoping’s self-criticism during the Cultural Revolution. He just abased himself and said over and over again, “I didn’t know Chairman Mao. I didn’t appreciate the brilliance of Chairman Mao.” Even he went through the ultimate humiliation but survived intact.

I was in Washington and went on the whole trip when Deng arrived in 1979 to normalize relationships. And you did get the sense of somebody who had his own sense of gravity about him, wasn’t a deeply insecure person just craving slavish loyalty. Deng Xiaoping was different. I would say Zhao Ziyang 赵紫阳 and Hu Yaobang 胡耀邦 were also different. When you look through the different leaders, you want to get back and judge their character. Yes, times were different, but Deng Xiaoping was special because he had seniority. He did get cashiered twice, but he never lost his fundamental sense of himself, which I think many other people did.

Alexander Boyd: Let’s be optimistic here and say there’ll be a post-Trump GOP. How did those Chinese survivors like Deng Xiaoping, Zhao Ziyang, etc. make it through Mao’s Cultural Revolution? And what could that tell you for an aspiring politician today hoping to make it through Trump 2.0 and still have a political career?

Orville Schell: Xi Jinping made it through and he’s done alright. There’s no simple answer. In the good old days when you also had imperial autocracies in the form of emperors, if you ran afoul of the government, you could run up into the mountains and become a Buddhist monk or a Taoist priest, and mind your own business. But that wasn’t possible under Mao.

We’ll see about what happens in America. I suspect we won’t get to such a state in America, but who knows? The question is, during autocracy, authoritarian rule, what should good people do? If you stick your head up, it gets chopped off. You can run abroad, you’re just in waiting. There’s not much you can do. It is a good question, and I’m not sure I know the answer to it. Keep saying what we need to say as we are here today. In China, that was not possible under Mao. It was not possible under Mao ever since the early 1940s when he began lowering the boom, wanting to create a new man and a new era, bringing on thought reform, rectification, and all the rest of it.

Subscribe now

Jordan Schneider: Let’s stay on this dilemma of the officials who know that they are living in crazy times but still want to help the people. What’s the right way to kind of look at what Deng and Zhou and others did in the Mao period?

Orville Schell: They’re always dancing on a razor’s edge. You know, it's not a, a dance I would care to know how to get out of. Fair enough. Some of these people — and Zhou Enlai had a measure of this, I wager — you want to do good by the people but the cost of staying in the game is very high.

The people in the Trump administration, in the first go around, there were a few, some quit and they did in some significant measure keep their integrity intact. And they did do some good restraining things. I think this administration is much harder. He’s bringing in the — Elon Musk is like a leader of Red Guards and the Proud Boys are Red Guards equivalent.

It’s a very difficult human dilemma to know if you want to be in government and you are drawn to political power, how do you do it now? Can you do it? Or should you just become a Buddhist monk or a Taoist priest and just go up on your mountain and wait? I don’t know the answer to that. Us who are writers? Who have not been in political power, don’t want to be in political power. We’re not drawn to that flame. So we do what we do.

Jordan Schneider: Comparing America 2025 to anytime in Mao’s reign, the downsides of recording a podcast like this are much lower.

Orville Schell: For now Jordan, but people have long memories and there are archives and there are a lot of people. The way Xi Jinping’s father fell the second time was over a book about a big leader in northwest Shaanxi province, Liu Zhidan 刘志丹, that he allowed to be published. Mao said, “Well, you’re trying to put too much emphasis on him as the hero, not me.” Anyway, it’s a long, complicated story, but simply to say that sometimes small things done in past, to autocratic regimes like China, are grounds for you being pilloried in the future.

Jordan Schneider: We’ll get into that arc with Joseph later this summer. I’m still feeling okay about our freedom to podcast.

Orville Schell: I’m glad you’re doing it. My virtue is that I’m a little older. I don’t need to be so worried about my future.

Alexander Boyd: Michael Berry talks about this in his writing about Fang Fang — the term míng zhé bǎo shēn (明哲保身) means, “Don’t speak out in order to preserve yourself.” I personally think that in the United States, we have a great privilege to be able to speak out, and we should exercise that privilege.

Orville Schell: We still do. The government in sort of in the shape it’s in, it puts all the more burden on the institutions of civil society. Universities, think tanks, libraries, and community organizations do not owe fealty to the central government, but owe fealty to what they do. Media would be another very important example — cultural organizations, orchestras, operas, whatnot.

Alexander Boyd: Trump has shown an immense fascination with the Kennedy Center in D.C., which is where I’m based. I think it was Cats that was his favorite.

Bach and Bloodlines

Alexander Boyd: Let’s talk, Orville, about bloodline theory. What was bloodline theory during the Cultural Revolution? Why did it matter? Mao himself wasn’t an endorser of bloodline theory, but it did have a lot of influence.

Trump always talks about genes. “It’s all in the genes.” Quite recently, he weighed in on the NFL draft about a quarterback who’s sliding, Shedeur Sanders, and saying, “He has phenomenal genes. They should have picked him because his dad was such a good player.” Is bloodline theory another parallel with the Cultural Revolution ?

Orville Schell: During the Cultural Revolution, the notion of bloodlines worked like this: if your father was a hero, so you were good to go. But on the other hand, if your father was someone of questionable background, then you bore that stigma. You were placed in that class category because families were categorized based on their class background.

As you all remember, Mao had this notion that certain classes had rights and were revolutionary, while certain classes — like the bourgeoisie and landlords — didn’t have rights. The bloodline concept was very pernicious because it meant that if your father was labeled as a counter-revolutionary, a rightist, a capitalist roader, or a bourgeois element, the children inherited that stigma through blood.

That’s why it’s fascinating to delve into familial relationships in any Chinese family. Xi Jinping is the most important case here because he’s now the leader. But I should also mention that in my experience — and this may be better explored in literature than in nonfiction — there’s a cascading effect. All the harm, damage, and attacks that occurred throughout the fifties, sixties, and seventies in China have endured across generations in the Chinese families I know, going from grandfather to father to son to grandson. They persist like microplastics in the ocean — they’re forever chemicals in a way.

We’ve paid no attention to this phenomenon. The way these experiences deranged families, destroyed people’s ability to respect and love their parents, caused betrayals of friendships, and led to the savage attacks that people inflicted on one another — Red Guards attacking their teachers — all of this continues to reverberate.

This trauma isn’t something you get over the next day, and it lives on in ways that are very difficult to analyze. There’s no data, and China doesn’t have a vigorous psychoanalytic tradition to help people understand what influences may have been passed down to them through their experiences with parents who suffered.

The Cultural Revolution was deep, and its consequences are enduring. That’s why, when Deng Xiaoping came to power and waved his wand to rehabilitate people, saying it was a new era, I felt incredibly skeptical. I believed there was a whole residue of impact deeply embedded within society and human beings.

There had been so much damage — not just Mao and the party treating people badly, but people being forced to treat their spouses badly, their children badly, their relatives, friends, and colleagues. This is something that endures.

Alexander Boyd: This endures in China to this day, you argue. When was your most recent trip back to China, and how do you see it enduring today?

Orville Schell: My most recent trip was just as the COVID pandemic hit. All you have to do is talk to your friends. I have a friend who went to Harvard, had a very difficult time with her parents, grew up in China, and she set up a group for Chinese women similar to her to discuss this. I found that incredibly interesting.

Some of the things that they stumbled upon as they were trying to analyze the relationships they have with their parents — how are they influenced by the relationship their parents had to their parents and to society, and power. Very few people have wandered into this field.

Robert J. Lifton, a wonderfully brilliant psychoanalyst who wrote Thought Reform and the Psychology of Totalism in the 50s and then Revolutionary Immortality about Mao’s quest to make himself immortal so that his legacy would live on. There have been very few people — Lucien Pye, Richard Solomon — who’ve actually looked into the human element. That’s why I wrote a novel, because I felt I couldn’t touch it as a nonfiction writer. I didn’t get to the question of the role of religion, music, culture, love, family. All of these things are abiding human concerns.

Alexander Boyd: You titled that novel after a Lu Xun 魯迅 essay. How come?

Orville Schell: I love that essay, My Old Home (故鄉). It’s a very wistful essay about returning home after things changed. My novel was about a classical musician and what happened to him when he returned back to China in the 50s as a lover of Bach.

If I may say so, there is no human being whom I think is more antithetical to Chairman Mao than Johann Sebastian Bach. In fact, I want to write a play called “My Dinner with Johann,” where they have a conversation. Because Bach was all about religion. Mao Zedong was all about the external. Something’s wrong? It’s out there, not in here.

Yes, Confucianism did have a notion of self-cultivation, but it’s not like Christianity.

Jordan Schneider: Well, we have to end with the ChatGPT imagined conversation between Bach and Mao.

Orville Schell: There was a show Henry Kissinger went to, and Robin Williams started wandering down the aisle afterwards. He passed Kissinger, and he was saying things to people as he went. He said, “Oh, Henry, love all your wars.” I could imagine Bach starting off by saying to Mao, “Love all your revolutions.”

Jordan Schneider: This is how we’re going to start. Rewrite with Bach saying sarcastically to Mao, “Love all your revolutions. ” Alex, you’re Bach. Let’s go.

Alexander Boyd:  Love all of your revolutions, Chairman. Tossing the world upside down seems to be your favorite key signature.

Jordan Schneider: Upside down is where history finds its balance, Herr Bach. The masses must turn the old order on its head to set it upright.

Orville Schell: Now you’re talking like a robot, like a propaganda minister. I think Mao would say, “Tell me, Johann, what’s all this about Jesus? Why are you so obsessed with Jesus?” That would get Bach rolling. You remember when Clinton was in China, where he went into the Great Hall of the People for the press conference. At one point, Jiang Zemin, completely sui generis, said to Clinton, “Mr. President, I have a question. Why are so many Americans so interested in the Dalai Lama and Tibetan Buddhism?” He was speaking in Chinese. Of course, Clinton went on a tear. But I thought that was a sort of interesting question to ask. You can’t imagine Xi Jinping asking such a question.

Alexander Boyd: What did Clinton say?

Orville Schell: Clinton said something like, “Chairman Jiang, I think if you had a chance to meet the Dalai Lama, you’d really like him.” Jiang, who’d been off script and bantering in a very nice human way with Clinton, grabbed the podium and, as I recollect, he said, “With your permission, Mr. President, shall we close this section?”

Jordan Schneider: What are you reading right now?

Orville Schell: I’m reviewing this Torigan book for Foreign Affairs. I’ve also been reading William Shirer’s The Rise and Fall of the Third Reich, Timothy Ryback’s book about 1931 and ’32 in Germany, and a Robert Paxton book on fascism.

Jordan Schneider: Let’s talk about Rise and Fall of the Third Reich for a second. What stuck out to you about that book?

Orville Schell: I’m very curious where we are on this sort of scenario — how Germany headed off into fascism and the Third Reich. It’s pretty frightening when you look back at the various steps, at what happened and who didn’t say anything, who just shut up.

There’s a wonderful diary of Victor Klemperer, who was the cousin of a famous conductor. He kept a daily record of what happened. He was Jewish, his wife was Catholic, and he lived in Potsdam. They keep saying, “Surely something will happen, surely someone will come, and that can’t be it. Surely the Allies will come in.” Of course, they didn’t. We ended up with Hitler being elected Vice Chancellor, then we’re off to the races.

I’m very interested in how things slide into this state where you end up with an autocracy. Remember that Germany was the highest form of European civilization, and yet you ended up with Hitler.

Alexander Boyd: I’m also reading Hitler-specific these days. I’m reading Ian Kershaw’s two-part, two-volume biography of Hitler. I just finished Hubris and now I’m onto Nemesis.

Jordan Schneider: Orville, do you know Ian Kershaw? I’ve been trying to find his email address.

Orville Schell: I don’t know him but he would be great to get on and just walk him through the steps. There are some wonderful, wonderful books about that period that we need to know more about. Because you see how a slow erosion step by step, step by step with a kind of charismatic crackpot leader leading the charge and how it happens and how people just don’t rise to the occasion to stop it. They think, “Oh the courts will do it, oh something will do it,” but sometimes they don’t.

This is why I think comparing Trump to Xi is interesting and worthwhile doing. Although some of your more rigorous scholars may think there’s no data, no theoretical constructs, but for me it’s the heart of the matter. It has a lot to do with how people grow up. Autocratic leaders write themselves as very large — democratic leaders don’t have that opportunity as much. When you’re in big leader culture land, whether Putin, Kim Jong Un, Xi Jinping, Orban, whoever, it really matters who they are and where they came from and what their sort of operating system is, who installed it and when.

You can say fairly safely, although there are a lot of amazingly wonderful people in China — I have to say, and I married one — but the Cultural Revolution created massive amounts of personal, psychological, intellectual damage. It wasn’t just people got killed, people got in jail for a little while, and then Deng waved his wand and it was all over. That’s not how historical trauma works.

That’s why I find Torigan’s book so interesting. To his credit, he doesn’t do what I’ve just done, which is draw conclusions or try to draw gratuitous conclusions. He just tells the story. It’s a monumental job of research. You can draw your own conclusions, and that’s what I intend to do in Foreign Affairs.

Alexander Boyd: Any hints on those conclusions that are coming out soon?

Orville Schell: I want to make some surmises about what growing up in the Cultural Revolution meant to the formation of Xi Jinping, his form of governance today.

Alexander Boyd: According to the book, Xi Zhongxun, upon hearing of the Cultural Revolution, actually asked for his soul to be lit afire by it, which I found to be incredible research, obviously, on Torigan’s part to get this. Does that indicate that Xi Zhongxun, for whom the Party always came first, was unable fundamentally to connect with Mao because the Cultural Revolution was ongoing? He was already purged, but he yearned desperately for this. It’s kind of like a priest who doesn’t hear God’s voice calling. Is that a correct analysis?

Orville Schell: The Party — and Zhou Enlai suffered from this too — they all did. Some of them did have a sense that something was deeply awry. But there was no other show in town except the Party and the Revolution. They were veteran revolutionaries.

Xi Zhongxun, no matter what they did to him, and what they did to him was pretty horrendous, though not the worst, he never lost his belief in the Revolution and the Party. That’s what he imbued his son with. Yes, bad things happen. We can’t have chaos again. But the party is fundamentally right. The revolution cannot be questioned.

Share

It’s a classic case of where people have no other place to turn except run off into the hills if you can. And we see this in our own government now. People desperately wanting to be in the limelight, in power, in government. And they have all kinds of rationalizations. Rubio, my God, he used to think Trump was a buffoon. Now he’s sold his soul. Read Doctor Faustus.

Jordan Schneider: My favorite line with Rubio is there’s an old New Yorker profile of him where he reads The Last Lion, the Churchill biography. And he said he read it twice and that he saw himself as Churchill, like warning about the Nazis. The analogy was Iran getting the bomb or something. But to go from that to where we are today is something.

Orville Schell: Power is an incredible intoxicant. Once you get that needle in your arm, that’s your currency and that’s your realm. It’s very hard to imagine what else you’re going to do with yourself. That’s why as a writer, I’ve always said, “No, not going there.” I’m just going to stay a lowly scribe. I don’t even particularly yearn to go to China now because I know if I did yearn to go, that would circumscribe me, it would make me feel I couldn’t say certain things because I’d know there’d be consequences.

I told you this, Jordan. In 1991, I did a year-long project with 60 Minutes on forced labor and the Laogai system. It aired. It was incredible. We got into prison camps. I kept a diary of it, and I edited it and sent it to The New Yorker. They edited it and were about to go into print and I looked at it and said, “I can’t publish this.” I was a younger man, I had a Chinese wife, and I had parents-in-law in China. I threw it in a box.

I pulled it out two years ago and thought, “My God, the question of forced labor in Xinjiang is more relevant than ever.” I took it out. That will be the end of me in terms of grace from the Chinese Communist Party. But that’s okay. I’d rather that than I can’t write and say what I think. I think I was right to put it in a box then. But that’s not a healthy tendency for any society. You remember chōutì wénxué (抽屉文学), “drawer literature,” things that people could only write and put in a drawer.

Jordan Schneider: Benjamin Nathan just won the Pulitzer Prize for this really awesome book To the Success of Our Hopeless Cause: The Many Lives of the Soviet Dissident Movement, which folks should also read.

Orville Schell: Perry Link is sort of the avatar of the Chinese version of that. I recommend his book on Liu Xiaobo, and he’s just written another book, The Anaconda in the Chandelier. I think that’s a really important question.

 You know, intellectuals are poor, weak creatures, and those who stand up — read Blood Letters. Alex, will you send Jordan the Elaine Pagels program we did on the comparison of Jesus and Lin Zhao? We had them both on stage talking about the role that faith plays in adversity and revolution. We started off with Bach, a beautiful aria. I wanted Bach as the avatar of being in the mix. You’ll enjoy this, Jordan.

Jordan Schneider: All right, well, we’ll put it in the show notes as well.

Orville Schell: It’s hard to explain to people, but if you watch it, you’ll understand.

Alexander Boyd: On Orville’s note on publishing and not publishing, we just published an excerpt from Perry Link’s forthcoming book, The Anaconda in the Chandelier.

Jordan Schneider: All right, thanks so much for being a part of ChinaTalk, Orville.

Orville Schell: As always, it’s a great pleasure. You have a great program, Jordan.

of the substack did an excellent job writing up the key dynamics of the deal. I’ll be running excerpts from his post with some comments of mine interspersed.

Our government, having withdrawn the new diffusion rules, has now announced an agreement to sell massive numbers of highly advanced AI chips to UAE and Saudi Arabia (KSA). This post analyzes that deal and that decision.

It is possible, given sufficiently strong agreement details (which are not yet public and may not be finalized) and private unvoiced considerations, that this deal contains sufficient safeguards and justifications that, absent ability to fix other American policy failures, this decision is superior to the available alternatives. Perhaps these are good deals, with sufficiently strong security arrangements that will actually stick.

Perhaps UAE and KSA are more important markets and general partners than we realize, and the rest of the world really is unable to deploy capital and electrical power the way they can and there is nothing we can do to change this, and perhaps they have other points of strategic importance, so we have to deal with them. Perhaps they are reliable American allies going forward who wouldn’t use this as leverage, for reasons I do not understand. There are potential worlds where this makes sense.

The fact remains that the case being made for this deal, in public, actively makes the situation seem worse. David Sacks in particular is doubling down and extending the rhetoric I pushed back against last week, when I targeted Obvious Nonsense in AI diffusion discourse. Even within the White House, the China hawks are questioning this deal, and Sacks responded by claiming to not even understand their objections and to all but accuse such people of being traitorous decels wearing trench coats.

I stand by my statements last week that even if accept the premise that all we need care about are ‘America wins the AI race’ and how we must ‘beat China,’ our government’s policies, on diffusion and elsewhere, seem determined to lose an AI race against China.

The Central Points From Last Week

The point of the diffusion rules is to keep the AI chips secure and out of Chinese hands, both in terms of physical security and use of their compute via remote access. It is possible that the agreements we are making with UAE and KSA will replace and improve upon the functionality, in those countries in particular, of the diffusion rules.

It’s not about a particular set of rules. It is about the effect of those rules. Give me a better way to get the same effect, and I’m happy to take it. When I say ‘something similar’ below, I mean in the sense of sufficient safeguards against the diversion of either the physical AI chips or the compute from the AI chips. Access to those chips is what matters most. Whereas market share in selling AI chips is not something I am inclined to worry about except in my role as Nvidia shareholder.

I do not consider them reliable allies going forward, and there are various reasons that even the best version of these agreements would make me deeply uncomfortable, but it is possible to reach an agreement that physically locates many data centers in the Middle East and lets them reap the financial benefits of their investments and have compute available for local use, but does not in the most meaningful senses ‘hand them’ the compute in question. As in, no I do not trust them, but we could find a way that we do not have to, if they were fully open to whatever it took to make that happen.

I also would highlight the implicit claim I made here, that the pool of American advanced AI chips is essentially fixed, and that we have sufficient funding available in Big Tech to buy all of them indefinitely. If that is not true, then the UAE/KSA money matters a lot more. Then there is the similar question of whether we were going to actually run out of available electrical power with no way to get around that. A lot of the question comes down to: What would have counterfactually happened to those chips? Would we have been unable to deploy them?

[Jordan: See ChinaTalk’s recent coverage of an excellent IFP report that explored what a policy agenda to rapidly increase the energy available for AI deployment would look like]

With that in mind, here are the central points I highlighted last week:

1. America is ahead of China in AI.

2. Diffusion rules serve to protect America’s technological lead where it matters.

3. UAE, Qatar and Saudi Arabia are not reliable American allies, nor are they important markets for our technology. We should not be handing them large shares of the world’s most valuable resource, compute.

4. The exact diffusion rule is gone but something similar must take its place, to do otherwise would be how America ‘loses the AI race.’

5. Not having any meaningful regulations at all on AI, or ‘building machines that are smarter and more capable than humans,’ is not a good idea, nor would it mean America would ‘lose the AI race.’

6. AI is currently virtually unregulated as a distinct entity, so ‘repeal 10 regulations for every one you add’ is to not regulate at all building machines that are soon likely to be smarter and more capable than humans, or anything else either.

7. ‘Winning the AI race’ is about racing to superintelligence. It is not about who gets to build the GPU. The reason to ‘win’ the ‘race’ is not market share in selling big tech solutions. It is especially not about who gets to sell others the AI chips.

8. If we care about American dominance in global markets, including tech markets, stop talking about how what we need to do is not regulate AI, and start talking about the things that will actually help us, or at least stop doing the things that actively hurt us and could actually make us lose.

Diffusion Controls Have Proven Vital

Diffusion controls on AI chips we’ve enforced on China so far have had a huge impact. DeepSeek put out a highly impressive AI model, but by their own statements they were severely handicapped by lack of compute. Chinese adoption of AI is also greatly held back by lack of inference compute.

China is competing in spite of this severe disadvantage. It is vital that we hold their feet to the fire on this. China has an acute chip shortage, because it physically cannot make more AI chips, so any chips it would ship to a place like UAE or KSA would each be one less chip available in China.

Whenever you see arguments from David Sacks and others against AI diffusion rules, ask the question:

1. Is an argument for a different set of export controls and a different chip regime that still protects against China getting large quantities of advanced AI chips?

2. Or is it an argument, as it often is, that to preserve our edge in compute we should sell off our compute, that to preserve our edge in tech we should give away our edge in tech?

   1. As in, that what matters is our market share of AI chips, not who uses them?

   2. This is not a strawman, for example Ben Thompson argues exactly this very explicitly and repeatedly. Ben Thompson’s recent interview with Jensen Huang, CEO of Nvidia, made it clear both of them have this exact position. That to maintain America’s edge in AI, we need to sell our AI chips to whoever wants them, including China, because ‘China will not be held back’ as if having a lot more chips wouldn’t have helped them. And essentially saying that all Nvidia chips everywhere support the ‘American tech stack’ rather than China rather obviously turning around and using them for their own tech. He explicitly is yelling we need to ‘compete in China’ or else.

   3. Complete Obvious Nonsense talking of his own book, which one must remind oneself is indeed his job, what were you really expecting him to say? Well, what he is saying is that the way we ‘lose the AI race’ is someone builds a CUDA alternative or steals Nvidia market share. That his market is what matters. It’s full text. Not remotely a strawman.

I would disagree with arguments of form #2 in the strongest possible terms. If it’s arguments of form #1, we can talk about it.

It’s a Huge Deal

We should keep these facts in mind as we analyze the fact that the United States has signed a preliminary chip deal with the UAE. There is a 5GW AUE-US AI campus planned, and is taking similar action in Saudi Arabia. The deals were negotiated by a team led by David Sacks and Sriram Krishnan.

Lennart Heim: To put the new 5GW AI campus in Abu Dhabi (UAE) into perspective. It would support up to 2.5 million NVIDIA B200s.

That's bigger than all other major AI infrastructure announcements we've seen so far.

In exchange for access to our chips, we get what are claimed to be strong protections against chip diversion, and promises of what I understand to be a total of $200 billion in investments by the UAE. That dollar figure is counting things like aluminum, petroleum, airplanes, Qualcomm and so on. It is unclear how much of that is new.

The part of the deal that matters is that a majority of the UAE investment in data centers has to happen here in America.

I notice that I am skeptical that all the huge numbers cited in the various investment ‘deals’ we keep making will end up as actual on-the-ground investments. As in:

Walter Bloomberg: UAE PRESIDENT SAYS UAE TO INVEST $1.4T IN U.S OVER NEXT 10 YEARS

At best there presumably is some creative accounting and political symbolism involved in such statements. Current UAE foreign-direct-investment stock in the USA is only $38 billion, their combined wealth funds only have $1.9 trillion total. We can at best treat $1.4 trillion as an aspiration, an upper bound scenario. If we get the $200 billion we should consider that a win, although if the deal is effectively ‘all your investments broadly are in the West and not in China’ then that would indeed be a substantial amount of funds.

Nor is this an isolated incident. The Administration is constantly harping huge numbers, claiming to have brought in $14 trillion in new investment, including $4 trillion from the recent trip to Arabia, or roughly half of America’s GDP.

Jason Furman (top economic advisor, Obama White House): That’s nuts and baseless. I doubt the press releases even add up to that. But, regardless, press releases are a terrible way to determine the investment or the impact of his policies on it.

Justin Wolfers: Trump has claimed a $1.2 trillion investment deal from Qatar. Qatar’s annual GDP is a bit less than $250 billion per year. So he’s claiming an investment that would require every dollar every Qatari earned over the next five years.

UAE’s MGX will also be opening Europe’s largest data center in France, together with Nvidia, an 8.5 billion Euro investment, first phase to be operational in 2028. This has been in the works for a while.

Do You Feel Secure?

Not that the numbers ultimately matter all that much. What does matter is: How will we ensure the chips don’t fall literally or functionally into Chinese hands?

It comes down to the security provisions and who is going to effectively have access to and run all this compute. I don’t see here any laying out of the supposed tough security provisions.

Without going into details, if the agreements on both physical and digital security are indeed implemented in a way that is sufficiently tough and robust, if we are the ones who both physically and digitally control and monitor things on a level at least as high as domestically, and can actually have confidence none of this will get diverted, then that goes a long way.

We don’t yet have enough of that information to say.

The public explanations for the deal, and the public statements about what safety precautions are considered necessary, do not bring comfort.

I very much do not like comments like this, made in response to the Bloomberg piece above.

David Sacks (US AI Czar): If the concern [about the deal] is about diversion of advanced semiconductors to China, that’s an important policy objective but one that is easily addressed with a security agreement and a “trust but verify” approach. Modern AI supercomputers are 8 feet tall and weigh two tons; these are not like diamonds smuggled in a briefcase. To verify that chips are where they’re supposed to be, an auditor can just visit the data center and count server racks.

Semianalysis: As such, physical inspections are key to ensuring the GPUs are where they are supposed to be. The White House is aware of the risks and is prepared to ensure the chips are where they are supposed to be. As Bloomberg reported (quotes David Sacks’s statement above).

I’m sorry, what? You can ‘just visit the data center and count server racks’?

It terrifies me to see so many people arguing for the deal explicitly saying that this is a good statement, rather than a terrifying statement that we are hoping no one involved in the actual work believes to be true.

[Jordan: In 2022 I did a show with Kevin Wolf, the former head of BIS, who mentioned off-hand that there were literally only two people Commerce had in China to do site inspections. Since then BIS’ budget has barely budged and faced DOGE cuts in the early days of this administration. There is not the capacity to do datacenter inspections in a serious way today.]

Jordan Schneider: Let's start with the last thing first. This 60-day unverified list: is there a potential that the impact of this could actually be a lot bigger than what we currently see, if some of these big names on the unverified entity list end up graduating to full entity-list status?

Kevin Wolf: It's really up to the Chinese government. If they allow access, give the information that the Commerce Department is seeking, and allow the two people in China that do inspections and site visits to do their jobs…

Jordan Schneider: Whoa, wait, two? There are, like, 30 firms on there!

Kevin Wolf: Yeah, I know, but no other country has any, and the Commerce Department is the only export control organization on the planet with its own enforcement agents who are subject matter experts in export controls.

I think diversion via remote access is far more likely than physical diversion, and this response does not address the remote access issue at all, but even simply treating this as a physical diversion issue, o3, is the quoted statement accurate?

o3 (full one shot, full response quoted, in response to literally quoting Sacks without attribution and asking ‘is this accurate?’):

Short version: Mostly wrong. Chips are tiny and high-value, so diversion is hard to stop with “count the racks.” Physical audits help but they are far from sufficient.

Why “just count server racks” fails

1. High value-to-weight ratio.
   At ~$30 k per H100, a single 20 kg carry-on full of GPUs is > $1 m. That’s an attractive target. Racks are heavy; the payload isn’t.

2. Modularity.
   GPUs slide out in minutes. An auditor sees the enclosure; they cannot tell if the trays were swapped yesterday or will be tomorrow. Dummy blanks cost <$10.

3. Front companies & trans-shipping.
   Chips often leave the compliant data-center before any inspection window. Paperwork shows them installed; reality changes after the auditor flies home.

4. Remote lease risk.
   Even if the hardware stays in the U.S., Chinese entities can simply rent time on it. Export-control rules treat “deemed exports” (remote access by restricted users) as violations.

5. Limited audit frequency & competence.
   Meaningful verification requires random, intrusive, technically savvy inspections plus cryptographic attestation of firmware. Counting racks is neither.

Bottom-line probability estimate

My best estimate: ≤ 30 % that “security agreement + occasional rack counting” alone keeps advanced GPUs out of China for > 3 years. ≥ 70 % that significant leakage continues absent tighter controls (HW tracking, cryptographic attestation, and supply-chain tagging).

So the quoted claim is misleading: rack-level audits are helpful but nowhere near “easily addresses” the diversion problem.

When I asked how many chips would likely be diverted from a G42 data center if this was the security regime, o3’s 90% confidence interval was 5%-50%. Note that the G42 data center is 20% of the total compute here, so if we generously assume no physical diversion risk in the other 80%, that’s 1%-10% of all compute we deploy in the UAE.

Is that acceptable? The optimal amount of chip diversion is not zero. But I think this level of diversion would be a big deal, and the bigger concern is remote access.

I want to presume, for overdetermined reasons, that Sacks’s statement was written without due consideration or it does not reflect his actual views, and we would not actually make this level of dumb mistake where they could literally just swap the chips out for dummy chips. I presume we are planning to use vastly superior and more effective precautions against chip diversion and also have a plan for robust monitoring of compute use to prevent remote access diversion.

But how can we trust an administration to take such issues seriously, if their AI Czar is not taking this even a little bit seriously? This is not a one time incident. Similar statements keep coming. That’s why I spent a whole post responding to them.

David Sacks is also quoted extensively directly in the Bloomberg piece, and is repeatedly very dismissive of worried about diversion of chips or of compute, saying it is a fake argument and an easy problem to solve, and he talks about these as if they were reliable American allies in ways I do not believe are accurate.

Sacks also continues to appear to view winning AI to be largely about selling AI chips. As in, if G42, an Abu Dhabi-based AI firm, is using American AI chips, then it essentially ‘counts as American’ for purposes of ‘winning,’ or similar. I don’t think that is how this works, or that this is a good use of a million H100s. Bloomberg reports 80% of chips headed to the UAE would go to US companies, 20% to G42.

I very much want us to think about the actual physical consequences of various actions, not what those actions symbolize or look like. I do think, despite everything else, it is a very good sign that David Sacks is ‘urging people to read the fine print.’ This is moderated by the fact that we do not have the fine print, so we can’t read it. The true good news there requires one to read all that fine print, and one also should not assume that the fine print will get implemented. Nor do we yet have access to what the actual fine print says, so we cannot read it.

Semianalysis Defends the Deal

Dylan Patel and others at Semianalysis offer a robust defense of the deal, saying clearly that ‘America wins’ and that this benefits American AI infrastructure suppliers on all levels, including AI labs and cloud providers.

They focus on three benefits: money, tying KSA/UAE to our tech stack, and electrical power, and warn of the need for proper security, including model weight security, a point I appreciated them highlighting.

Those seem like the right places to focus, and the right questions to ask. How much of their money is really up for grabs and how much does it matter? To what extent does this meaningfully tie UAE/KSA to America and how much does that matter? How much do we need their ability to provide electrical power? How will the security arrangements work, will they be effective, and who will effectively be in charge and have what leverage?

Specifically, on their three central points:

1. They call this macro, but a better term would be money. UAE and KSA (Saudi Arabia) can make it rain, a ‘trillion-dollar floodgate.’ This raises two questions.

   1. Question one: Was American AI ‘funding constrained’? The big tech companies were already putting in a combined hundreds of billions a year. Companies like xAI can easily raise funds to build giant data centers. If Google, Amazon, Apple, Meta or Microsoft wanted to invest more, are they really about to run out of available funding? Are there enough more chips available to be bought to run us out of cash?

   2. Semianalysis seems to think we should be worried about willingness of American companies to invest here and thinks we will have trouble with the financing.

   3. I am not convinced of this. Have you seen what these companies (don’t have to) pay on corporate bonds? Did we need to bring in outside investors? Should we even want to, given these investments look likely to pay off?

   4. This is a major crux. If indeed American big tech companies are funding constrained in their AI investments, then the money matters a lot more. Whereas if we were already capable of buying up all the chips, that very much cuts the other way.

   5. Question two: As we discussed earlier, is the trillion-dollar number real? We keep seeing these eye-popping headline investment numbers, but they don’t seem that anchored to reality, and seem to include all forms of investment including not AI, although of course other foreign direct investment is welcome.

   6. Do their investments in US datacenters mean anything, and are they even something we want, given that the limiting factor driving all this is either constraints on chip availability or on electrical power? Will this be crowding out other providers?

   7. If these deals are so positive for American tech companies, why didn’t the stock market moves reflect this? No, I will not accept ‘priced in.’

2. They call this geopolitical, that UAE and KSA are now tied to American technology stacks.

   1. As they say, ‘if Washington enforces tight security protocols.’ We will see. David Sacks is explicitly dismissing the need for tight security protocols.

   2. Classically, as Trump knows well, when the bank loans you a large enough amount and you don’t pay it back, it is the bank that has the problem. Who is being tied to whose stack? They will be able to at least cut the power any time. It is not clear from public info what other security will be present and what happens if they decide to turn on us, or use that threat as leverage. Can they take our chips and their talents elsewhere?

   3. This can almost be looked at as a deal with one corporation. G42 seems like it’s going to effectively be on the UAE side of the deal, and it is going to have a lot of chips in a lot of places. A key question is, to what extent do we have the leverage on and control over G42, and to what extent does this mean they will act as a de facto American tech company and ally? How much can we trust that our interests will continue to align? Who will be dependent on who? Will our security protocols extend to their African and European outposts?

   4. Why does buying a bunch of our chips tie them into the rest of our stack? My technical understand is that it doesn’t. They’re only tied to the extent that they agreed to be tied as part of the deal (again, details unknown), and they could swap out that part at any time. In my experience you can change which AI your program uses by changing a few lines of code, and people often do.

   5. It is not obvious why KSA and UAE using our software or tech stack is important to us other than because they are about to have all these chips. These aren’t exactly huge markets. If the argument is they have oversized effect on lots of other markets, we need to hear this case made out loud.

   6. Seminanalysis points out China doesn’t even have the capacity to sell its own AI chips yet. And I am confused about the perspectives here on ‘market share’ and the implied expectations about customer lock-in.

3. They call this infrastructure, I’d simply call it (electrical) power. This is the clearly valuable thing we are getting. It’s rather crazy that ‘put our most strategic asset except maybe nukes into the UAE and KSA’ was chosen over ‘overrule permitting rules and build some power plants or convince one of our closer allies to do it’ but here we are.

   1. So the question here is, what are the alternatives? How acute is the shortage going to be and was there no one else capable of addressing it?

   2. Also, even if we do have to make this deal now, this is screaming from the rooftops, we need to build up more electrical power everywhere else now, so we don’t have this constraint again in the future.

Semianalysis also raises the concern about model weight security, but essentially think this is solvable via funding work to develop countermeasures and use of red teaming, plus defense in depth. It’s great to see this concern raised explicitly, as it is another real worry. Yes, we could do work to mitigate it and impose good security protocols, and keep the models from running in places and ways that create this danger, but will we? I don’t know. Failure here would be catastrophic.

Understanding the China Hawks

There are also other concerns even if we successfully retain physical and digital control over the chips. The more we place AI chips and other strategic AI assets there, the more we are turning UAE, Saudi Arabia and potentially Qatar into major AI players, granting them leverage I believe they can and will use for various purposes.

David Sacks continues to claim to not understand that others think that ‘winning AI’ is mostly not about who gets to sell chips, who uses our models and picks up market share, or about superficially ‘winning’ ‘deals.’

He not only thinks it is about market penetration, he can’t imagine an alternative. He doesn’t understand that many, including myself, this is about who has compute and who gets superintelligence, and about the need for proper security.

David Sacks: I’m genuinely perplexed how any self-proclaimed “China Hawk” can claim that President Trump’s AI deals with UAE and Saudi Arabia aren’t hugely beneficial for the United States. As leading semiconductor analyst Dylan Patel observed, these deals “will noticeably shift the balance of power” in America’s favor. The only question you need to ask is: does China wish it had made these deals? Yes of course it does. But President Trump got there first and beat them to the punch.

Sam Altman: this was an extremely smart thing for you all to do and i’m sorry naive people are giving you grief.

Tripp Mickle and Ana Swanson (NYT): One Trump administration official, who declined to be named because he was not authorized to speak publicly, said that with the G42 deal, American policymakers were making a choice that could mean the most powerful A.I. training facility in 2029 would be in the United Arab Emirates, rather than the United States. [Jordan: this is a wild line to be giving to the NY Times—I thought all the China hawks already got purged!]

But Trump officials worried that if the United States continued to limit the Emirates’ access to American technology, the Persian Gulf nation would try Chinese alternatives.

The hawks are concerned, because the hawks largely do not think that the key question is who will get to sell chips, but rather who gets to buy them and use them. This is especially true given that both America and China are producing as many top AI chips as they can, us far more successfully, and there is more than enough demand for both of them. One must think on the margin.

Given that so many China hawks are indeed on record doubting this deal, if you are perplexed by this I suggest reading their explanations. Here is one example.

Tripp Mickle and Ana Swanson (NYT): Mr. Goodrich said the United States still had the best A.I. engineers, companies and chips and should look for ways to speed up permitting and improve its energy grid to hold on to that expertise. Setting up some of the world’s largest data centers in the Middle East risks turning the Gulf States, or even China, into A.I. rivals, he said.

“We’ve seen this movie before and we should not repeat it,” Mr. Goodrich said.

Sam Winter-Levy, a fellow at the Carnegie Endowment for International Peace, said the huge chip sales did “not feel consistent with an America First approach to A.I. policy or industrial policy.”

“Why would we want to offshore the infrastructure that will underpin the key industrial technology of the coming years?” he asked.

This does not seem like a difficult position to understand? There are of course also other reasons to oppose such deals.

Here is Jordan Schneider of China Talk’s response, in which he is having absolutely none of it, explicitly rejecting that either America or China has chips to spare for this. rejecting that UAE and KSA are actual allies, not expecting us to follow through with reasonable security precautions, and saying if we wanted to do this anyway we could have held out for a better deal with more control than this, I don’t know why you would be confused how someone could have this reaction based on the publicly available information:

Jordan Schneider: It’s going to cannibalize US build-out and leave the world with three independent power-centers of AI hardware where we could’ve stuck to our guns, done more power generation at home, and only had China to deal with not these wild-card countries that are not actual allies. If this really is as important as we believe, why are we letting these countries and companies we deeply distrust get access to it?

The Gulf’s BATNA wasn’t Huawei chips, it was no chips. Whatever we’re trying to negotiate for, we can play harder to get. BIS can just say they can’t buy Ascends and it’s not like there’s enough capacity domestically in China to service global demand absent the TSMC loophole they charged through. Plus, we’re offering to sell them 10× the chips that Huawei could conceivably sell them anytime soon even if they use the TSMC-fabbed wafers.

Where’s the art-of-the-deal energy here? Right now I only see AMD and NVDA shareholders as well as Sama benefiting from all of this. I thought we wanted to raise revenue from tariffs? Why not charge 3× the market rate and put the premium into the US Treasury, some “Make America Great Again” industrial-development fund, use it to triple BIS’ budget so they can actually enforce the security side, put them on the hook for Gaza…I don't know literally anything you care about. How about a commitment not to invest in Chinese tech firms? Do we still care about advanced logic made in America? How about we only let them buy chips fabbed in the US, fixing the demand-side problem and forcing NVDA to teach Intel how to not suck.

Speaking of charging through loopholes, all of the security issues Dylan raises in his article I have, generously, 15 % confidence in USG being able to resolve/resist industry and politicians when they push back. If it’s so simple to just count the servers, why hasn’t BIS already done it / been able to fight upstream industry lobbying to update the chips-and-SME regs to stop Chinese build-outs and chip acquisition? What happens when the Trump gets a call from the King when some bureaucrat is trying to stop shipments because they see diversion if they ever catch it in the first place?

Why are we doing anything with G42 again? Fine, if you really decide you want to sell chips to the UAE, at the very least give American hyperscalers the off-switch. It’s not like they would’ve walked away from that offer! America has a ton to lose in the medium term from creating another cloud provider that can service at scale, saying nothing of one that has some deeply-discomforting China ties pretty obvious even to me sitting here having never gotten classified briefings on the topic.

Do the deal’s details and various private or unvoiced considerations make this deal better than it looks and answer many of these concerns? Could this be sufficient that, if looked at purely through the lens of American strategic interests, this deal was a win versus the salient alternatives? Again: That is all certainly possible!

Our negotiating position could have been worse than Jordan believes. We could have gotten important things for America we aren’t mentioning yet. The administration could have limited room to maneuver including by being divided against itself or against Congress on this. On the flip side, there are some potentially uncharitable explanations for all of this, that would be reasonable to consider.

Rhetoric Unbecoming

Instead of understanding and engaging with such concerns and working to allay them, Sacks has repeatedly decided to make this a mask off moment, and engage in a response that I would expect on something like the All-In Podcast or in a Twitter beef, but which is unbecoming of his office and responsibilities, with multiple baseless vibe and ad hominem attacks at once that reflect that he either is willfully ignorant of the views, goals and beliefs of those he is attacking and even who they actually are, or he is lying and does not care, or both, and a failure to take seriously the concerns and objections being raised. Here is another illustration of this:

David Sacks (May 17): After the Sam Bankrun-Fraud fiasco, it was necessary for the Effective Altruists to rebrand. So they are trying to position themselves as “China Hawks.” But their tech deceleration agenda is the same, and it would cost America the AI race with China.

There are multiple other people I often disagree with on important questions but whom I greatly respect who are working on in administration on AI policy. There are good arguments you can make in defense of this deal. Instead of making those arguments in public, we repeatedly get this.

Everything Sacks says seems to be about vibes and implications first and actual factual claims a distant second at best. He doesn’t logically say ‘all so-called China hawks who don’t agree with me are secret effective altruists in trench coats and also decels who hate all technology and all of humanity and also America,’ but you better believe that’s the impression he’s going for here.

Could China Have ‘Done This Deal’?

Would China have preferred to ‘do this deal’ instead? That at best assumes facts, and arguments, not in evidence. It depends what they would get out of such a deal, and what we’re getting out of ours, and also the security arrangements and whether we’ve formed a long lasting relationship in which we hold the cards.

I’m also not even sure what it would mean for China to have ‘done this deal,’ it does not have what we are offering. Semianalysis says they don’t have similar quantities of chips to sell, and might not have any, nor are their chips of similar quality.

I do agree China would have liked to ‘do a deal’ in some general sense, where they bring UAE/KSA into their orbit, on AI and otherwise, although they don’t need access to electrical power. More capital and friends are always helpful. It’s not clear what that deal would have looked like.

[Jordan: the fact that the administration is citing some old news of Huawei promising to gift just $75m in chips to the Malaysian government (and not even today! Over two years!) is policy malpractice. Plus, this administration just made clear that using Huawei Ascend chips violates US export controls, giving America an enormously powerful lever to dissuade fence-sitting countries not to buy into what is already an inferior tech offering.]

Tyler Cowen Asks Good Questions

Here’s Tyler Cowen being clear eyed about some of what we are selling so cheap. The most powerful AI training facility could be in the UAE, and you’re laughing?

Tyler Cowen: Of course Saudi and the UAE have plenty of energy, including oil, solar, and the ability to put up nuclear quickly. We can all agree that it might be better to put these data centers on US territory, but of course the NIMBYs will not let us build at the required speeds. Not doing these deals could mean ceding superintelligence capabilities to China first. Or letting other parties move in and take advantage of the abilities of the Gulf states to build out energy supplies quickly.

Energy and ability to overcome NIMBYs is only that which is scarce because America is refusing to rise to this challenge and actually enable more power generation. Seriously, is there nowhere in America we can make this happen at scale? If we wanted to, we could do this ourselves easily. We have the natural gas, even if nuclear would be too slow to come online. It is a policy choice not to clear the way. And no, I see zero evidence that we are pulling out the stops here and coming up short.

I think this frame is exactly correct - that this deal makes sense if and only if all of:

1. The security deal is robust and we retain functional control over where the compute goes.

2. We trust our friends here to remain our friends at a reasonable price.

3. We counterfactually would not have been able to buy these chips and build data centers to power these chips.

As far as I can tell China already has all the power it needs to power any AI chips it can produce, it is using them all, and its chip efforts are not funding constrained.

So for want of electrical power, and for a few dollars, we are handing over a large amount of influence over the future to authoritarian powers with very different priorities and values?

Tyler Cowen: In any case, imagine that soon the world’s smartest and wisest philosopher will soon again be in Arabic lands.

We seem to be moving to a world where there will be four major AI powers — adding Saudi and UAE — rather than just two, namely the US and China. But if energy is what is scarce here, perhaps we were headed for additional AI powers anyway, and best for the US to be in on the deal?

Who really will have de facto final rights of control in these deals? Plug pulling abilities? What will the actual balance of power and influence look like? Exactly what role will the US private sector play? Will Saudi and the UAE then have to procure nuclear weapons to guard the highly valuable data centers? Will Saudi and the UAE simply become the most powerful and influential nations in the Middle East and perhaps somewhat beyond?

Yes. Those are indeed many of the right questions, once you think security is solid. Who is in charge of these data centers in the ways that matter? Won’t they at minimum have the ability to cut the power at any time? Who gets to decide where the compute goes? What are they going to do with all this leverage we are handing them?

Is this what it means to have the future be based on American or Democratic values? Do you like ‘the values’ of the UAE and Saudi Arabian authorities?

Tyler Cowen: I don’t have the answers to those questions. If I were president I suppose I would be doing these deals, but it is very difficult to analyze all of the relevant factors. The variance of outcomes is large, and I have very little confidence in anyone’s judgments here, my own included.

Few people are shrieking about this, either positively or negatively, but it could be the series of decisions that settles our final opinion of the second Trump presidency.

The administration thinks that the compute in question will remain under the indefinitely control of American tech companies, to be directed as we wish.

Sriram Krishnan: Reflecting on what has been an amazing week and a key step in global American AI dominance under President Trump.

These Middle East AI partnerships are historic and this “AI diplomacy” will help lock in the American tech stack in the region, help American companies expand there while also building infrastructure back in the U.S to continue expanding our compute capacity.

This happens on top of rigorous security guarantees to stop diversion or unauthorized access of our technology.

More broadly this helps pull the region closer to the U.S and aligns our technological interests in a very key moment for AI.

It’s a very exciting moment and a key milestone.

I hope that they are right about this, but I notice that I share Tyler’s worry that they are wrong.

Saudi Arabia Also Made a Deal

Similarly, Saudi Arabia’s Humain is going to get ‘several hundred thousand’ of Nvidia’s most advanced processors, starting with 18k GB300 Grace Blackwells.

The justification given for rescinding the Biden diffusion rules is primarily that failure to do this would have ‘weakened diplomatic relations with dozens of countries by downgrading them to second-tier status.’

But, well, not to reiterate everything I said last week, but on that note I have news.

One, we’re weakening diplomatic relations with essentially all countries in a series of unforced errors elsewhere, and we could stop.

[Jordan: that feeling when you treat Dubai better than Canada]

Two, most of the listed tier two countries have always had second-tier status. There’s a reason Saudi Arabia isn’t in Five Eyes or NATO. We can talk price about which countries should have which status, but no our relations are not all created equal, not when it comes to strategically vital national interests and to deep trust. I don’t share Sacks’s stated view that these are some of our closest and most trustworthy allies. Why does this administration seem to always want to make its deals mostly with authoritarian regimes, usually in places where Trump has financial ties?

Tripp Mickle and Ana Swanson (NY Times): The announcements of the two deals follow reports that $2 billion has flowed to Trump companies over the last month from the Middle East, including a Saudi-backed investment in Trump’s cryptocurrency and plans for a new presidential airplane from Qatar.

There’s always Trust But Verify. The best solution, if you can’t trust, is often to set up things so that you don’t have to. This can largely be done. Will we do it? And what will we get in return? What is announced mostly seems to be investments and purchases, that what we are getting are dollars, and Bloomberg is skeptical of the stated dollar amounts.

At Best A Second Best Solution

This deal is very much not a first best solution. It is, at best, a move that we are forced into on the margin due to our massive unforced errors in a variety of other realms. Even if it makes sense to do this, it makes even more sense to be addressing and fixing those other critical mistakes.

Electrical power is the most glaring in the context of this particular. There needs to be national emergency level focus on America’s inability to build electrical power capacity. Where are the special compute zones? Where are the categorical exemptions? Where is DOGE with regard to the NRC? Where is the push for real reform on any of these fronts? Instead, we see story after story of Congress actively moving to withdraw even the supports that are already there, including plans to outright abrogate contracts on existing projects.

The other very glaring issue is trade policy. If we think it is this vital to maintain trade alliances and open up markets, and maintaining market share, why are we otherwise going in the opposite direction? Why are we alienating most of our allies? And so on.

The argument for this deal is, essentially, that it must be considered in isolation. That other stuff is someone else’s department, and we can only work with what we have. But this is a very bitter pill to be asked to swallow, especially as Sacks himself has spoken out quite loudly in favor of many of those same anti-helpful policies, and the others he seems to be sitting out. You can argue that he needs to maintain his political position, but if that also rules out advocating for electrical power generation and permitting reform, what are we even doing?

If we swallow the entire pill, and consider these deals only on the margin, without any ability to impact any of our other decisions, and only with respect to ‘beating China’ and ability to ‘win the AI race,’ and assume fully good faith and set aside all the poor arguments and consider only the steelman case, we can ask: Do these deals help us?

I believe that such a deal is justifiable, again on the margin and regarding our position with respect to China, if and only if ALL of the following are true:

1. Security arrangements are robust, the chips actually do remain under our physical control and we actually do determine what happens with the compute. And things are set up such that America retains the leverage, and we can count on UAE/KSA to remain our friends going forward.

2. This was essentially the best deal we could have gotten.

3. This represents a major shift in our or China’s ability to stand up advanced AI chips, because for the bulk of these chips either Big Tech would have run out of money, or we would have been unable to source the necessary electrical power, or China has surplus advanced AI chips I was not previously aware of and no way to deploy them.

4. Entering into these partnerships is more diplomatically impactful, and these friendships are more valuable, than they appear to me based on public info.

Check out some ChinaTalk coverage from least year about China’s Middle East AI ambitions.

Mood Music

What does the future of industrial policy in America look like, and what state capacity investments are needed to get there? How does China factor into the future of the U.S. semiconductor industry? And what do government affairs offices actually do? To explore these questions, we’re concluding our CSIS Chip Chat series with Bruce Andrews. Bruce has had a long career on Capitol Hill, led government affairs for Ford, served as Deputy Secretary of Commerce under President Obama, and most recently headed up government affairs at Intel. He’s now a fellow at CSIS.

We discuss…

• How to bring expertise and legislating back to Capitol Hill,

• The case for a new “Department of Competitiveness”

• Industry’s role in policymaking and what it took to get semiconductor manufacturers on board with the CHIPS Act,

• Why Silicon Valley suddenly became interested in politics,

• How to optimize industrial policy in a stick-focused political environment.

Listen on Spotify, iTunes, or your favorite podcast app.

Jordan Schneider: Bruce, you were intimately involved in the birth and execution of the CHIPS and Science Act. Throughout the discussion around the legislation and grant implementation, politicians and officials kept saying, “We haven’t done this in decades. We’re building this muscle again. We’re learning on the fly.” This is something America had done in the past, but the energy atrophied over time. Reflecting back, maybe let’s start with the legislation writing process. How did the fact that these policy tools hadn’t been worked out in so long impact the discussion and development of the legislation?

Bruce Andrews: The United States government is actually not set up well and is challenged in two ways when implementing big industrial policies. First, the executive branch is not organized properly — we really have a 20th-century cabinet structure for 21st-century problems. Second, we don’t have the muscle memory of how to effectively implement industrial policy.

We’ve been fortunate in the United States to have a robust private sector that, with government investment and R&D spending, has been successful. But the world has changed significantly. For Congress in particular, there’s not necessarily the expertise, and they often rely on dealing directly with the executive branch to identify issues and the lack of authority that the US Government has. In many cases, we don’t have statutory authorities for executive branch agencies to carry out these policies.

When writing legislation, there’s always a balance between being specific enough while also giving enough flexibility to the administration to execute those policies. The CHIPS and Science Act was actually a good example, and it was done in two parts. The first part was drafted in the second half of the Trump administration and passed at the beginning of the Biden administration. Then they had to provide the funding because you had the authorization, but needed to appropriate the dollars that would actually be spent.

This was a huge legislative lift, and Congress added many new requirements during the funding legislation phase.

Jordan Schneider: You started your career in the early 90s and ended up as general counsel to Senate Commerce in the early 2010s. Thinking back on the arc of legislative ambition, principal quality, and staff quality — what trends did you observe over that time?

Bruce Andrews: Things have changed significantly since the early and mid-90s. The quality of staff — and this is not to say there aren’t still tremendous staff on Capitol Hill — but there used to be really deep expertise. The growth of lobbying is partly to blame for this change, as it has deprived Capitol Hill of many good staff members.

I’ve always believed we need to pay congressional staff more. Frankly, we should pay executive branch career civil servants more as well. When you’re a young Hill staffer or even a very experienced one, and you see your colleagues in the private sector making two or three times what you’re making, your incentive to stay on Capitol Hill and develop deep expertise is diminished.

Jordan Schneider: Last year, I did a show with Philip Wallach, who wrote the book Why Congress?, which examined the post-war arc of congressional capacity and willingness to put itself center stage. We’re recording this on May 6, and Congress is currently in a situation where money it appropriated isn’t being spent because an executive feels like withholding it, and they’re not doing much about it.

We’ve seen this before. After Nixon, Congress regained a real sense of institutional pride and reasserted itself in a way it hadn’t for decades. Who knows if that’s the future Donald Trump will leave us with in 2028? But if you’re imagining a more ambitious and energetic legislative branch, where does that leave the institution? Where does that leave America? What would need to happen to bring about a new flourishing?

The CHIPS and Science Act was a perfect storm of COVID, bipartisanship, and an eager executive that allowed for a significant bill. But for a broader shift in congressional ambition and execution capacity, what ingredients would we need to see in the body politic or what incentives would enable more legislation like the CHIPS and Science Act?

Bruce Andrews: I think success does beget success, but I also think it’s both sides coming together and identifying what the issues are. There are really three things. One is agreement and bipartisan agreement on what the issue is that you’re trying to solve. In the CHIPS and Science Act, there was real clarity on trying to get particularly advanced logic, but also trying to move semiconductor production back to the United States.

Second is an agreement on what tools the government has, or should have, in order to do that. In the CHIPS and Science case, people looked and they said, “Taiwan, Korea, and China have spent the last 30 years providing incentives to help their industries be successful. The US has sat on the sidelines and let the private sector do it. And we now have watched the majority of Atlantic Advanced Logic production move overseas.” Second is seeing the problem, which is the need to close the cost gap between the US and Asia, and putting in a solution that was able to do that.

Third is bipartisan support. There were a number of heroes of the CHIPS Act on both sides of the aisle. On the Democratic side you had Ro Khanna and Chuck Schumer and Doris Matsui and people who said there’s a problem. And then on the Republican side, you had great senators like Todd Young and John Cornyn, Mike McCall in the House who said, “We agree with you, this is a big problem and let’s work together and work even over the politics."

As you’ll recall, at the end, the CHIPS Act sort of got caught up in a little bit of last-minute politics because of some other things that were going on unrelated to the CHIPS Act. To their credit, the champions on the Republican side, including Mike McCall, John Cornyns, and Todd Young, really helped push it through and helped to get it done. You need all three. You need clear identification of an issue, you need a clear government solution in a narrow and tailored way to provide the tools to do it, and you need bipartisan support to be successful.

Jordan Schneider: That’s nice to hear, but it seems like we have never been further away from that. When you have a president who’s blowing up the NIH because he feels like it’s too big, we’re going to need to have some big thermostatic response to what we have now, because this is not the glide path. This is the plane that crash-lands, and only then we become motivated to do something.

Bruce Andrews: That’s right. You’ve got to have an agreement that there are solutions we can put in place. When you get down to the member level, you actually get a lot of good thinking and a lot of good bipartisan agreement. But then translating that — unfortunately, if I were going to have one critique of Congress right now, it would be that often we’re not legislating big issues. We’re more legislating when there’s a crisis.

There are lots of really important issues out there every single day that we need to be addressing. Unfortunately, it seems that we get to the level of a crisis, and then it’s easier to pass legislation. We need to get back to the old days — real bipartisan agreement and the ability to work in a bipartisan way, and not to be pressed by your colleagues in either party to say, “Why are you working with those people of the other party? Let’s just win the next election, and then we can do it our way.”

That’s been going on for a long time, where both sides have thought, “If we just wait until the next election, we’ll win — we’ll be in charge of the House, Senate, and the White House, and then we can get our package.” The reality is that it doesn’t work. With the filibuster in the Senate, with the need for getting things done in the close margins of the House, and with the nature of the White House, it’s a mistake to put off our problems because, “Oh, if we just win the next election, we’ll solve them.” It’s never that easy.

Beefing Up The Executive

Jordan Schneider: Let’s turn to the executive branch. The CHIPS and Science Act basically got to hire the pick of the litter because they had special hiring authorities and were able to really bring together this crack team from some government vets, but a lot of folks from Wall Street and a little bit from the semiconductor industry, too. Recently, Dan Kim said that in 2015, there were only two people who were thinking about the commercial semiconductor industry in government. What does good look like here? What are the institutions you would have loved to have had that were building that commercial understanding and institutional muscle for whenever we have to do this the next time, so it’s not from such a dead start?

Bruce Andrews: First of all, I don’t think the answer is just taking a chainsaw and eliminating agencies, but I do think having some flexibility to reorganize government for the 21st century and for the digital world would help. We still have a very 20th-century analog model that was built mostly post-World War II and over the second half of the 20th century for a very different world.

If I were king for the day, I would start by reorganizing the government to be much more focused and move some things around so that they made more sense. President Obama tried to do that or talked about doing that with government reorganization.

The Commerce Department is a perfect example. I would meet with counterparts from METI in Japan, or Ministry of Commerce in China, or the European Union. Counterparts would have a much clearer set of mandates and authorities and a clearer set of responsibilities, and would have basically all of the business-related functions within the country within one agency. Organizing government in the right way is important, but I don’t know if that’s going to happen anytime soon.

Jordan Schneider: No, let’s stay here. Let’s do a little more fan fiction. What are you pulling in and putting out?

Bruce Andrews: If it were up to me, I would pull all business-related economic development and competitive stuff into the department. I actually would rename it the Department of Competitiveness. I would focus on the economic tools, the commerce tools, the trade tools, the industry tools and give the authority for real industrial policy to work with public-private partnerships with the business community to make sure that we’re making these smart investments in R&D and building the future.

Second, I would set up a set of programs that help the Commerce Department, or hopefully the new Department of Competitiveness, have a set of tools that it can use to support industry. China, Japan, Korea, and Taiwan have all for the last 30 years had very focused policies, and they’ve been successful at encouraging certain industries.

We can criticize some of the competitive practices the Chinese used, but they put in a Made in China 2025 plan and set out key industries that they wanted to successfully dominate. If you look at those industries that they identified, they’ve actually done pretty well.

You need the magic of the private sector. A huge advantage that the United States has is our robust private sector. But you also need public-private partnerships in order for that leadership to take place. The world is too complicated right now for companies to just go at it alone.

When I was at Intel, I would always say, “We’re not just competing with companies, we’re competing with countries.” That’s the case — many US companies are going at it alone, competing not just against their foreign competitors, but the tools and resources of their foreign competitors’ governments as well. We don’t want to recreate that. We don’t want to copy, but we need to establish our own model that works.

Secretary Lutnick has proposed what I would call “SelectUSA on steroids.” During the Obama administration, we created the SelectUSA office to attract foreign investment. To Secretary Lutnick’s credit, I think he’s identified that there are investments and then there are what I would call the mega investments.

Having an office within the Department of Competitiveness that would help facilitate both US companies and foreign companies making those investments in the United States would be fantastic. Giving the US Government some more flexible tools to help incentivize those investments and to help those investments be successful is important for us to be competitive.

For 30 years, there were reasons why globalization was a good economic theory, and there was a lot that was good for consumers and a lot of other things. But we also live in a very different world now. We need to have a structure and tools that allow the US Government to be competitive and the US to be competitive in this very different world.

Jordan Schneider: Not to be your assignment editor, but I think this should be your first CSIS piece — what authorities you want to be here. It’s a fun future-casting exercise where you don’t really have to engage with the present messiness. But can we do one more pass on this? What do you give up? The Commerce is already a giant mess. What do you slough off it? I guess there are three buckets — what do you slough off, what do you take from other agencies, and aside from SelectUSA, what’s on your wish list for new authorities and power?

Bruce Andrews: First of all, I would point out that the Commerce Department is perfect. I would probably change nothing. But if I had to change something — Obama made a proposal, for example, to move NOAA (National Oceanic and Atmospheric Administration). I could argue that that’s important to business and that things like the National Weather Service and others actually are key facilitators of public goods that help business.

There was talk of moving the Small Business Administration into the Department of Commerce. It doesn’t necessarily make sense that we have SBA, we have SBIR programs around the government, we have Economic Development Administration in Commerce, and that we have the Minority Business Development Administration.

1. I would consolidate all of these different economic and small business minority business programs into one part of the Department of Competitiveness for it to be successful.

2. I would also take all the technology agencies around the government. We have NTIA, we have NIST, we have various technology programs, rural broadband at the Department of Agriculture. We have all these different programs around different agencies. I would consolidate them together and say the Department of Competitiveness has an economic development piece, it has a technology piece, and it has an international trade piece.

One of the things that was very controversial, and I could argue this both ways, is moving USTR within the Department of Commerce and having the International Trade Administration and the US Trade Representative’s Office be one super trade organization.

Government statistics — why are the Bureau of Labor Statistics, the Bureau of Economic Analysis, and the Census all separate organizations throughout the government? Take all government statistics and put them in one place.

I think there are a bunch of things that you could do to rationalize all this and to help the US really be supercharged and competitive. Frankly, one of the reasons that Obama ran into trying to get this consolidation authority was Congress didn’t want to give up the various committee jurisdictions. When I was at the Commerce Department, we had, I think, eight full committees and 79 subcommittees that had jurisdiction over the Department of Commerce. That doesn’t really make a whole lot of sense.

Jordan Schneider: Within this context, let’s do a little history lesson on how CHIPS and Science came to be.

Bruce Andrews: The interesting thing — you mentioned something that’s really important, in 2015, there were only two people within the US government who were experts on semiconductors. I learned that in November of 2015, when I spoke as the Deputy Secretary of the Commerce Department to the Semiconductor Industry Association.

Interestingly, at that time, the SIA (Semiconductor Industry Association) CEOs, who previously had all been very confident of “government stay out of our way, we don’t need your help, we’ve got this, we’re 10 years ahead of China, we don’t really need government assistance” — when I spoke to them in November of 2015, I saw what I would almost describe as a sense of fear in their eyes about the Made in China 2025 program and what was called “The Big Fund,” the $250 billion fund that the Chinese government had set up to help facilitate their semiconductor industry.

I took that back and sat down with my then-boss, Secretary Penny Pritzker. Penny and I said, “Let’s put a plan together to help the US Semiconductor industry. What would that look like to help the US semiconductor industry compete with Made in China 2025?”

Subscribe now

What’s funny is I called the Assistant Secretary of the International Trade Administration for Industry and Analysis. I said, “Okay, we need to put together a semiconductor plan.” He said, “Okay, I’ll bring my one guy up and let’s talk about it.” We only had one guy doing this. Now, the good news is there’s a guy named Travis Mosier who is super talented, super smart, and knew these issues quite well.

We put together a plan, a playbook, and we said, “Okay, what can the US Government do?” Interestingly, we initially got some pushback because people said to us, “Why are semiconductors special?” Secondly, we got some pushback because people said, “Well, if we help the semiconductor industry, won’t other industries be unhappy and feel like we should help them too?” My response was, “Well, yes, but that doesn’t make it any less important.”

We spent several months working with the White House team and the interagency team. Interestingly, Paul Selva, who was the Vice Chair of the Joint Chiefs of Staff, was also really interested in microelectronics. Paul and I got together and pushed through the NSC process a plan to start having much more focus and attention on semiconductors.

We had none of the tools. Those came later. Eventually, Penny Pritzker gave a big speech that really helped define the challenge. We had a PCAST report, which is the President’s Committee of Advisors on Science and Technology, which for the first time made a series of recommendations. That was right at the end of Obama, but it then carried into the Trump administration, and all of that attention and focus eventually became the CHIPS and Science Act.

It shows two things. One is you need to identify a problem and start talking about the solutions to those problems. Two, you need focused energy and intensity because nothing happens fast, either in the federal government or in Congress. Even though it took several years for all of this stuff to come together, we were successful at both putting focus on the attention, setting a set of tools, and then enacting a program that represents by far and away the biggest piece of industrial policy we’ve seen, at least in the last 50 years.

The Policymaking Ecosystem

Jordan Schneider: Let’s continue the story from the perspective of an Intel executive. What is Government Affairs? And maybe using this story as a case study, what can and can’t — or what is easy and what is hard — for industry to do to make things happen in the legislative and executive branch?

Bruce Andrews: Government affairs is really about education. It is bringing information to government policymakers to help understand the challenges that the private sector is facing and then also help to start having a discussion about what the solutions look like.

The irony of the CHIPS Act is that it was actually not started by Intel, although Intel will be the biggest beneficiary. Where it started was actually Mike Pompeo trying to recruit TSMC to come to Arizona. What TSMC said to Pompeo was, “We have a huge cost gap. We get incentives in Taiwan to build. We need something in the United States."

Ironically, when I was Deputy Secretary of the Commerce Department, I met with Terry Gou, the chairman and CEO of Foxconn, who basically said to me, “I want to build in the United States, but you need to give me big federal incentives.” I said, “Well, here’s the problem. We don’t have federal incentive programs in the United States. It doesn’t work the same way it does in China and Taiwan.” Terry eventually did start that project, but it ended up not working.

People were starting to realize that if we wanted to bring manufacturing back — and I give Secretary Pompeo a lot of credit because I think he recognized that was the way to get TSMC back. Interestingly, when Pat Gelsinger came in as the Intel CEO, he recognized that Intel needed to become a contract manufacturer for other companies. Foundry would provide more competition to TSMC and Samsung. The CHIPS and Science Act allowed Intel to do the construction in the United States. It allowed Intel to really make a fundamental change to its business model, moving manufacturing or building manufacturing in the United States.

It’s often forgotten that the CHIPS Act started in the Trump administration. It didn’t get passed, but when Biden came in, it got passed quite quickly. That set off the debate for funding the CHIPS and Science Act, which was eventually passed in July of 2022.

Jordan Schneider: Coming back again to — what does education mean? How does one educate? Is it easier or harder today versus back in the day where you have more or less sophisticated staffers and all the stuff we were talking about?

Bruce Andrews: In the same way that you did not have expertise within the Commerce Department, I’m not sure there was a single person who had any level of expertise on semiconductors in the Congress. Someone’s probably going to call me and say, “Hey, I was that person. You’re wrong.” But let’s just say for the purposes of argument, there really was not a deep level of expertise.

The good thing about a company like Intel is it’s got a huge number of very smart experts who are thinking about understanding the global industry. Being able to present a picture through data — and I will say the Semiconductor Industry Association is actually really good at providing data and information that gives a very full and complete picture to help educate lawmakers.

Generally you’ve got to assume lawmakers start at a relatively low level of understanding just because they’re not experts. You do get some members who come in as experts or you get a guy like Todd Young who really learns about the semiconductor industry and has a state like Indiana and a university like Purdue that want to focus on how to be effective and successful in the semiconductor space. You generally start with not a huge amount of information, but you start with an interest.

I give Doris Matsui and Mike McCall credit in the House. As I mentioned, you had Young, Cornyn, and Schumer in the Senate who said, “This is really important.” They then turned to industry and said, “Help educate us on two things. First, what is the status of the global industry? What’s the problem we want to solve? Second, what tools do you need to help us?”

In the case of the CHIPS Act, it was identifying that there was a 30-55% cost gap between manufacturing in the United States and manufacturing in Asia. The CHIPS Act — including the grants, but also the 25% investment tax credit — was all about filling that gap and helping US manufacturing to be competitive. That’s not necessarily something that members of Congress or the people in the administration would totally understand well on their own. That’s where private sector actors can come in and help educate them as to what those tools would look like to be the most successful.

Jordan Schneider: That is a nice story. Perhaps your average American — or maybe your average independent tech policy podcaster — would see this and see something wrong here. The fact that the US government had one guy, the fact that Congress, the executive branch had one guy, the fact that Congress had no one, and that there were not think tanks — we can debate how independent they are from corporate money — academics, we can also debate how they are from corporate money, particularly when you’re talking about these high technology industries where there’s a lot of university to industry connection.

As a big fan of the CHIPS Act, it was kind of concerning to me seeing that every voice in this debate was funded or connected to industry. If we end up doing more and more industrial policy, it would be nice to have voices whose paychecks are not necessarily connected to these outcomes to also be able to inform legislators.

Bruce Andrews: I would say two things to that. The American people had a real crash course in the importance of the semiconductor industry during COVID when they couldn’t buy products because there was not an availability of semiconductors because they were all being sourced out of Asia.

I hear you, and obviously this is why I think bipartisanship is so important. You bring members from both sides who have different constituencies, but they come to the table and they say, “Let’s get the best policy we can and reflect a lot of views.”

The semiconductor industry has never been a particularly big or politically active industry. They don’t have big PACs. It’s not like a lot of industries where there’s a huge amount of money flowing through. It’s actually been relatively unsophisticated in Washington.

That’s where the identification of the problem was so important. I think there were two voices that were critical. One was actual consumers. I’ll never forget when we went to meet with a very senior senator, and he told us that he had autos sitting on lots in his state waiting for semiconductors to be put in so that they could go out to dealers. One senator told us he actually bought a Ford F150, but it didn’t have the heated seat yet because the chip that would heat his seat was not available. This is a guy who lived in a really northern state that was super cold. I felt bad for him.

The second piece of this is the voice of the national security community. What also makes semiconductors kind of unique is the national security implications. As I mentioned, Paul Selva, the Vice Chair of the Joint Chiefs, was very focused on microelectronics and having a trustworthy supply of microelectronics in the United States.

You saw a lot of leading members of the national security community and the committees in Congress. Mark Warner is a great example — both a tech guy and a national security leader, as the chair of the Senate Intelligence Committee — is one of the most vocal voices for the CHIPS Act. Not because Virginia is a big semiconductor manufacturing state, although I think Mark would like that. It was more out of a recognition that this is a huge national security imperative that he was very supportive of.

There’s no doubt that there’s lots of stuff where there needs to be public scrutiny, there needs to be public debate. You need to hear all kinds of voices to make sure that special interests don’t just take advantage of a process. But I think in this case there were two pieces. One was the passage of the congressional legislation, and then the second was the implementation by the CHIPS Program Office to try to make sure that this was done in the way that was most focused on the national interest.

Jordan Schneider: Do you have any thoughts about think tanks as institutions and what they can and can’t do and what their purpose is?

Bruce Andrews: I think think tanks are actually very important. I should have mentioned them earlier. My experience with think tanks is, yes, they need money to run and some do take corporate and other types of money. It’s not just corporate — there’s all kinds of money that help fund think tanks. But in my experience, people tend to be very independent because their reputation is on the line. They’re very focused on doing the highest quality work.

Think tanks are really important because they do provide an independent perspective and sometimes an independent validation. Sometimes they can do studies that a company can’t do. I’ll give you an example. I just saw this morning that the Rhodium Group came out with a study about the China Made in 2025 plan’s effectiveness. That’s not something that a single company could do.

There are a lot of good think tanks with a lot of really smart, thoughtful people who are adding to the debate.

When I was a policymaker, both in Congress at the Senate Commerce Committee as a very senior staffer and general counsel, but also as the Deputy Secretary of the Commerce Department, everybody who came in had their self-interest. Part of my job was to get enough voices to make sure that someone’s self-interest was validated, to try to incorporate other perspectives.

My rule when I was in government was always I would meet with anybody and listen to you. I might tell you no and I might tell you that I think you’re completely wrong. But that’s okay. That’s actually how the system works. It should be a dialogue and it should be a back-and-forth. It should be a really robust public debate and we should have hearings in Congress and we should have lots of different perspectives aired. And then you’ve got people who in the end are going to have to make decisions, but at least you’ve had a broad set of opinions and views and data put on the table.

Policymakers have to be able to sort through all that. That’s one of the reasons both policymakers having expertise, but also having staff with expertise, helps to sort through this massive amount of information that comes in. I’ve many times seen think tank reports that really help to shape legislation or help to shape either the identification of problems or identification of solutions and frankly sometimes bring bipartisanship — my three components for what you need to be successful. Think tanks play a very important part in that.

Jordan Schneider: You mentioned America or senators getting lessons in the semiconductor industry from cold butts. America’s also about to get a whole lot of lessons in trade policy thanks to what’s been happening over the past few months. What do you want to say about this stuff? We can do chip specific economy in general.

Bruce Andrews: No, I mean, look, this is obviously a very complicated topic. You know, I get that. What President Trump has said, I agree with him on two things. One is that we need to identify strategic sectors where we need to be investing as a country in manufacturing. I wouldn’t say that it’s everything because there are a lot of things that we just can’t do. But one level should be what can we do strategically, both out of our own economic interest, national security interest, and where we can be successful.

Second is how do we work with allies? When I was at the Commerce Department, there was a lot of discussion about NAFTA and Mexico because a good that was made in Mexico had 46% US content. A good that was made in China had 4% US content. Everybody realized — this light bulb went off to people with that data — basically saying, “Hey, for things that we can’t economically or cost effectively do in the United States, it’s actually in our interest to have them be done in Mexico or Canada.” You’ve seen a lot of work with various allies to try to have things manufactured in ways that are good for the United States.

Third is the availability for consumers. On one level, you’ve got to protect the workers. The story I told you earlier about when Penny Pritzker and I decided to make semiconductors a priority for the Department of Commerce, one of the reasons was I had met with the steel industry executives and the National Steel Workers union several days before. I literally said to Penny, “I don’t want to be in a position in 10 years,” and in meeting with the steel workers, I basically said, “Hey, look, we don’t have the tools to be able to be helpful to you. We want to help you, but the statutory tools and authorities we have will not allow us to be helpful to you here.”

Penny and I said, “Okay, let’s make sure that’s not the semiconductor industry in 10 years — the experience we’re having with the steel unions and the steel industry today.” I do think that’s important — identifying what is really in our core national interest, and what can we as a country be successful at.

Tariffing everything in sort of a broad way isn’t necessarily going to be effective. We’re already starting to see some of the economic effects of that. I’m hoping that the administration looks and says, “Okay, what are we trying to solve for here?” What are the authorities we have and how do we do it? Thus far it doesn’t feel like we have clarity on what problem we’re trying to solve or exactly what solutions we’re trying to enact to do that. I’m hoping that the administration, as they do these negotiations with other countries, will get more clarity on that. Otherwise, I do worry that this is going to be a trade war that isn’t successful for the United States, for American consumers, but also American companies.

I’ll give you an example. One of the things that really concerns me is if you are a US manufacturer, there are a lot of things that you have to import from overseas as inputs into machinery and all kinds of things. A great example is Intel and an Intel fab. There is a $450 million ASML EUV lithography machine. What happens if you put a 25% tariff on an EUV lithography machine coming from Europe? Intel has to buy with those tariffs, but their foreign competitors in Taiwan, Korea, China don’t. What happens fairly quickly is it makes US companies completely non-competitive as manufacturers.

If our goal is to be successful in manufacturing, which it should be — I think that’s actually really important — helping US manufacturers, then we need to be doing everything we can to help them be successful, not tariffing intermediate goods in a way that actually harms the competitiveness of US manufacturers.

Jordan Schneider: Anything else you want to say about the future of Intel? Your well wishes for the new squad over there?

Bruce Andrews: Intel is a very important company to the United States, and it was one of the reasons I went there when Pat Gelsinger laid out his vision for a third US competitor. TSMC is a remarkably impressive company. Samsung’s a great company as well, but having a not over-consolidated industry and having manufacturing in the United States is really important. Intel is actually a very important company to this country. I spent a lot of time there because I believe in the mission and American manufacturing, but also the importance of having a leading advanced logic manufacturing company who does its R&D and whose primary operations are in the United States.

Jordan Schneider: Ben Thompson of Stratechery’s diagnosis of the challenge of the CHIPS Act was that it focused too far on the supply side and not enough on the demand side. His vision of what Intel is going to need to succeed on the foundry side at least is designated customers. On one hand, Qualcomm, Nvidia, Apple — they’re not excited to live under a monopoly decade going forward where the only people who can supply them are TSMC. But it is hard to be the first mover to go in with Intel and work with them to get all the tooling right and be that sort of guinea pig when on the other hand, you have someone who can charge you a huge markup, but you know they’re going to deliver.

His vision is that you solve the collective action problem with the government stepping in and banging heads and forcing everyone to go in on Intel. That seems like a high degree of difficulty to put on Howard Lutnick. How would you work on the demand side from a policy perspective to give Intel Foundry a boost?

Subscribe now

Bruce Andrews: I tend to be a carrot guy, not a stick guy. The stick can work in certain cases, but you’re better off having the carrots. One example is that a number of companies on the fabless side have talked about having a design tax credit for American fabless companies. During the CHIPS Act, they lobbied, but Congress did not add an investment design tax credit for the fabless industry.

I could see, and I know there are policymakers who would say, let’s give them a design tax credit, but let’s do it if they manufacture in the United States, because what we don’t want to do is give them a design tax credit to then go and manufacture in Korea or Taiwan or China. That’s one possibility.

There’s going to be a set of discussions because we have identified this as a national priority. Trying to find some tailored solutions is important — Intel’s got to do its piece to be competitive. On the other side, having some carrots to encourage companies to embrace manufacturing here in the United States would actually be a really good thing.

Jordan Schneider: Earlier you mentioned as Deputy Secretary of Commerce and in the Senate as well, you meet with everyone. I’m curious, thinking back to that, before you spent the past few years in the semiconductor industry, were there other industries where you’re just like, “Oh, man, these guys really get it. They’re super sharp?”

Bruce Andrews: It’s a really good question. The short answer is yes. There are a lot of really smart and talented companies out there. I was constantly surprised by the quality, and I think the tech space is one. There are a whole bunch of areas where you have really sharp, really talented leaders in those companies.

Tech was definitely one because tech, particularly since the late 90s, has attracted a lot of great talent. The people running these companies are actually very smart, talented people. When I meet them I go, “Wow, they are really smart.”

Every once in a while I would meet with people and I’d think, “God, your arguments suck. I could help you write your talking points better than you have written them.” But for the most part, I found most companies, particularly in areas that tended to be both technical but very competitive industries — I actually found people to be quite smart and impressive most of the time.

Jordan Schneider: There’s a story that Nvidia had like two government affairs officials before October 7, 2022. You were talking earlier about how the semiconductor industry was quite hands-off for a while. But even Robert Noyce in the 80s was like, “We’re going to do Sematech.” It’s this weird ebb and flow of caring about Washington, not caring about Washington, really realizing you have to care about Washington again. Is it a California thing? At what point, maybe for semiconductors and industries broadly, does the light bulb turn on that this place is really important to us and we’ve got to get serious about it?

Bruce Andrews: I think it’s been an evolution over the course, but particularly it’s accelerated in recent years. For manufacturing companies, they tend to get it more because they actually had physical locations. For a company like Intel or Global Foundries or Texas Instruments, you understand the importance of interrelationship with the government.

Nvidia is a good example where they didn’t have a PAC, they didn’t have any lobby, they didn’t have anybody registered to lobby probably until 2022. When I first met them when I was at SoftBank, they said, “Oh yeah, we have nobody registered to lobby in Washington. We don’t do lobbying.”

A couple things have happened. One is the world has become much more complicated. The geopolitics are driving a lot of the technology policies. Second is things like export controls which directly impact companies.

There has always been, particularly in Silicon Valley, but in a lot of other industries, a deep libertarian streak where it’s like, “Hey, we don’t need help from government. Leave us alone, we’ll compete, we’ll be successful, we can do this, we don’t need you.”

Other industries— pharmaceuticals is a good example where government health policies have always been very directly related because you have Medicare and Medicaid funding, or the auto industry where government policies on CAFE standards matter.

The semiconductor industry had not been particularly heavily regulated, particularly true on the fabless side — the companies that just did design but didn’t do their own manufacturing. The world has changed, and as the world changed, people realized, “Wow, we really have to engage here and we have to be in the game.”

That is not unusual of many other industries. The libertarian ‘leave us alone’ streak in Silicon Valley really tended to be much deeper than a lot of other places.

Jordan Schneider: Yeah, I think Trump trade policy is going to put that to bed for literally every industry probably.

Bruce Andrews: It has been a wake up call to a lot of companies, for sure.

Jordan Schneider: I’m curious for your reflections from a sort of GR corporate messaging perspective on how you bend to the prevailing winds in Washington. There have been a lot of examples, I think, over the first few months of the Trump administration, which almost show the extreme of this, with corporations really bending over backwards to kind of use the language and rhetoric of the Trump administration. But this is something that happens all the time. I’m curious, to what extent you see this as a disjuncture versus just kind of par for the course of trying to ingratiate yourself with the party in power.

Bruce Andrews: Look, I would say it’s a little of each. Historically, all players — it’s not just companies and corporate CEOs and lobbyists or whoever — have always tailored their message to how they think it will be best received by policymakers in power. I do think starting in 2017, but accelerating, China has become a much more used argument for almost every regulatory policy or whatever. When you see Meta saying ‘Oh, we need to not be regulated as much because we need to compete with China’, you realize that’s something probably because it actually appeals to policymakers. The speakers are always going to tailor their message to what they think is going to be most effective with the audiences. That’s one piece.

President Trump is very assertive. Companies have approached the administration carefully, wanting to get on their good side and understanding what those things are that President Trump sees as part of his priorities and his legacy. You have seen companies tailor their messaging as they have for pretty much every other administration. But they have definitely tailored that messaging as part of dealing with the administration.

Jordan Schneider: Can we reflect back a little bit on the PAT era? Because it was sort of remarkable, right, where he wouldn’t say China, he would say American national competitiveness or what have you. But then I remember he still had to go to China every once in a while because it was such an important market for Intel to sell into. Reflecting back as you guys were sort of managing that and then managing annoying podcasters like me whining that Intel is investing in Chinese chip design companies or whatever— how do you think through that calculus as you’re trying to maintain market access to a country that the US is not super happy you get so much revenue from?

Bruce Andrews: Well, I mean, I guess I would say two things. One is you’ve got to look at facts. For everybody in the semiconductor industry, and Intel is probably in the pack sort of average, but there are definitely companies that have a much higher amount of their revenue come from China. For American semiconductor companies to remain not just competitive, but to remain as leaders, the Chinese market is quite important because about anywhere from 30 to 40% of the market is in China.

That’s changing a little bit now that Apple’s moving phones out and Dell and HP are moving laptop production out. You’re seeing changes in the footprint, but you’re still going to see China be a very significant market for all these companies. The companies have a balance, right? Because on one hand, in order to continue investing in R&D and investing in factories and other things in the United States, you can’t drop your revenue 30%. If you lose 30% of your revenue, you’re dead. If you lose 30% of your revenue, you stop being competitive.

The question is how to find a balance which is to remain competitive and be successful while also respecting US national security concerns, which is a lot of what has driven the government to make policies that tend to reduce companies’ access to the Chinese market.

What’s interesting about this is each company has its own profile. They have to make a decision — how important is access to the Chinese market? What percentage of the revenues and then how are those revenues spent? What does it mean to their long-term and future competitiveness? It’s a very challenging calculation for any company to make.

For companies that want or need to be successful in the Chinese market to remain competitive, it is definitely a delicate process. How do you maintain that? I do think you see a lot of companies trying to do that because they recognize the importance of the Chinese market.

I don’t know if you’ve read Eric Schmidt’s piece in the New York Times, but there are a lot of very successful Chinese companies. Sometimes, if you go too far on export controls, what you actually end up doing is accelerating indigenous innovation in China. There are plenty of Chinese companies that would much rather buy, for example, Nvidia chips. But then you’ve got people in the United States government who are saying there are national security concerns.

It’s finding that balance and recognizing that there are really two pieces to this. One — how do we help American companies run faster to remain competitive and successful? Two, how do we balance the national security concerns?

Jake Sullivan called it the small yard and high fence. People in the industry always said, “Sure, but just keep the yard small and the fence low.” That is a constant balance. Industry is always going to prioritize self-interest, and they’ve got a fiduciary duty to their shareholders to try to maximize revenue. US government officials — there are plenty of people who’d say, “Oh, let’s just not sell anything to China,” which I don’t think is realistic because it will both undermine the long-term competitiveness of US companies, but also undermine the short-term competitiveness because of massive revenue loss where there’s not really a clear national security interest.

Jordan Schneider: I think that might be a nice place to end it. Bruce, thank you so much for being a part of ChinaTalk and looking forward to having you back once you’ve fleshed out your Department of Competitiveness.

Bruce Andrews: I’m excited for that. The Department of Competitiveness is actually very important — something that I used to spend a lot of time thinking about, I have thought less about. But it’s one of the things that has really struck me in reading about Ezra Klein’s new book, Abundance, is the need for having a government that works well. We are not organized for how the world exists today. We’re organized for how it used to exist.

Mood Music

From an anon

On April 25, observers of China’s AI scene got an important new statement of Xi Jinping’s views on AI in the form of remarks concluding a Politburo “study session” on AI led by Xi’an Jiaotong University professor Zheng Nanning. Couched in the turgid language of Partyspeak, the readout nevertheless merits close attention as one of precious few utterances direct from the General Secretary himself on AI. To read this new tea leaf, we need to understand some background on study sessions in general, and this one in particular.

What are study sessions?

Politburo study sessions, or 集体学习 (literally, “collective study”), are regular two-hour meetings of top CCP leadership devoted to learning about some topic deemed a priority by the General Secretary. Typically, most of the session is taken up by a lecture from an academic expert in the matter in question, but occasionally Politburo members themselves make presentations. The structure was originally established by Hu Jintao shortly after he was elevated to General Secretary of the CCP in 2002, and used to consolidate his power and promote his policy priorities.

These study sessions are a far cry from your undergrad TA office hours. The topics reflect key focus areas of the paramount leader, ranging from foreign policy to technology to stuff like “Opening New Frontiers in the Sinicization and Era-ification of Marxism”. The process for putting them together is extremely involved. Party functionaries choose an expert and work with them to ensure the lecture is pitch perfect for the leader’s priorities. A professor brought in to recant the gospel of historical materialism for the group in 2013 said it took over three months to prepare for his session. There can be as many as three dress rehearsals. Study sessions typically serve to solidify and broadcast the leader’s views on some developing policy topic, not to workshop or introduce new policy. However, they do sometimes signal further action coming down the pike — occasionally to dramatic effect. A 2023 July study session on “governance of military affairs,” for example, preceded a wave of PLA purges including the ouster of 9 generals in December 2023, and more in 2024. Leader’s comments in readouts are sometimes referenced in later policies, as with “guiding opinions” on blockchain from the Ministry of Industry and Information Technology in 2021 that quoted a 2019 study session.

Why study AI now?

With plenty of other signs of attention to AI coming out of Beijing these days, it’s not surprising that Xi would want to get everyone on exactly the same page on this notoriously complex topic. The last time the Politburo had a study session on AI was in 2018, led by Peking University Professor and Chinese Academy of Engineering Academician Gao Wen. This study session followed the publication of China’s landmark New Generation AI Development Plan in 2017 and presumably served to clarify how that plan should be interpreted and implemented. It could be that this April’s study session was intended partly to inform the several new funding programs announced recently.

The obvious answer, of course, is Deepseek. DeepSeek’s impressive releases of late 2024 and early 2025 catapulted the previously “low-key” company to direct attention from the very top echelons of the CCP. If the process for organizing a study session was initiated in January 2025, then a few months of preparation time would land us exactly in April. (Perhaps coincidentally, last week’s study session also comes exactly 2 and a half years after the launch of ChatGPT in November 2022 — almost precisely the same amount of time as between AlphaGo’s victory over Lee Sedol in November 2016 and the last Politburo study session on AI in October 2018. This gives some indication of the metabolic speed of the CCP system.)

What did Xi have to say about it?

Alas, the Politburo neither livestreams their study sessions on Zoom, nor even shares the slides after class. But we do have a summarized version of Xi’s closing comments for the session. Given the role of study sessions in communicating the leader’s views, this is at least as important as the content of the lecture itself.

China’s approach to AI has demonstrated a lot of consistency since 2018, so Xi’s comments on this study session hit many familiar notes. The topline summary, as given in the first paragraph, is that China will use the advantages of its “whole of nation” system to persist in “self-strengthening,” with an orientation towards applications, promoting “healthy and orderly” development in a “beneficial, safe and fair” direction. Self-strengthening should especially target “core, high and foundational” technologies. Essentially, China will maintain aggressive industrial policy in an attempt to indigenize important tech supply chains, while also developing valuable applications of AI, but all subject to certain guardrails and, of course, the salubrious guidance of the Party.

Similar themes through the two texts include concerns about China’s weakness in basic research and foundational tech but confidence about strengths in scale of data and market, a goal of properly integrating research and industry, and cultivating a thriving talent ecosystem. On many counts, however, the exact language is tweaked, often expanded or made more specific.

But some things are new! This year’s mention of “application-orientation” is similar in nature to mentions of “needs-orientation” or “problem-orientation” in 2018’s readout, but it’s now far more prominently placed, alongside “self-strengthening” in the title and in the one-sentence summary in the first paragraph. It also seems to imply more of a focus on diffusing technologies that already exist rather than creating new ones, even if “mere” engineering rather than fundamental breakthroughs, to serve national goals. This year’s readout also calls specifically for developing strategic emerging and future industries (战略性新兴产业和未来产业) and seems to give a nod to “AI for science” by referring to “a revolution in the paradigm of scientific research led by AI” (以人工智能引领科研范式变革).

As analysts have pointed out, Xi’s discussion of safety issues here is more forward-leaning than in 2018, or possibly any statement coming directly from the leader’s mouth. He describes risks from AI as “unprecedented,” and suggests implementing systems for “technology monitoring, risk early warning, and emergency response.” This is much more specific than previous policy statements calling to establish an “AI safety supervision and regulation system” or to strengthen “forward-looking (risk) prevention.” The study session readout’s language almost more closely echoes that of documents passed around at the recent Paris AI Action Summit by China’s new AISI-equivalent body, the China AI Safety and Development Association. Among a litany of priorities for the new organization, one of the more ambitious referred to setting “early warning thresholds for AI systems that may pose catastrophic or existential risks to humans.” Clearly, the thinking of safety-concerned AI experts such as famous computer scientist Andrew Yao and Tsinghua professor Xue Lan, who himself has lectured at study sessions thrice, most recently on emergency management in 2019, is finding resonance at the very top of the Party hierarchy.

A wholly novel component of the 2025 readout is the discussion of international engagement. While vague enough to be consistent with an intention to engage productively with the US and its partners’ efforts on AI governance, Xi’s focus on capacity building in the Global South and closing the “global intelligence gap” (全球智能鸿沟), as well as calling elsewhere to “grasp the initiative” in governance, suggests that this is also viewed at least partly as a dimension of international AI competition. We can imagine that Xue Lan, in particular, may have taken note of Xi’s description of AI as a potential “international public good” (国际公共产品) benefiting humanity. Xue was the lead author on a report that framed AI safety as a “global public good.” The same language later appeared in a statement from a group including Xue at the International Dialogue on AI Safety in Venice, and in a paper published by Oxford University including coauthors who work closely with Xue. Although it seems plausible that Xi’s language was influenced by this meme, there are two key differences here.

1. Here, Xi is referring to the benefits of AI development being a boon to the world. Though the paragraph does mention strengthening governance and creating a global governance framework, it is in the context of AI capacity-building. The most natural interpretation is that the idea here is to ensure a harmonized global regulatory regime to facilitate global diffusion and adoption of, and preempt backlash against, Chinese AI. This is distinct from the idea that AI safety is a public good, in that mitigating the downsides of AI development is beneficial to all by reducing transnational risks.

2. The second distinction is more subtle: Xi’s vision of AI as a public good is “international” (国际) rather than “global” (全球), notwithstanding Xinhua’s mistranslation in its English reporting on the comments. “Global” implies the whole world, taken as one single entity (especially in Chinese where the word is literally 全球 quánqíu “whole globe”). “International,” on the other hand, merely implies something that occurs between nations — and if we know one thing about nations, it’s that they don’t always treat all their fellow nations equally. In the context of a paragraph about diffusing AI technology in the Global South, a cynical read is that what this is saying is essentially: “work with China on AI, you’ll get a good deal.” In the past, the Chinese government has rhetorically promoted “open source” as a pillar of international cooperation and capacity building; here, Xi does not go so far as to use those exact words, but the fact that AI tinkerers in Indonesia can download and build on DeepSeek’s R1 may in fact be what he’s gesturing towards. The upshot here may be not so much that China hopes to save the world with, or much less from, AI, as that they are willing to cut deals with positive externalities to bind a sphere of influence together into a “community of common destiny.”

Whether in the intervening years or during their cram session, the Party has clearly learned some things about AI development.

What does this all mean for AI in China?

Although an important statement of policy, this study session may or may not indicate any significant change in policy. As discussed above, study sessions are typically less inflection points than moments of crystallization. Thus, it is most important as evidence for things we have the least evidence for otherwise.

Most striking here may be the surprising introduction of new language related to risk management. It is notable that both of the two experts to ever brief the Politburo on AI have been vocal on potential risks, with Gao Wen coauthoring a paper in 2021 on “Technical Countermeasures for Security Risks of Artificial General Intelligence” (though he is better known as the Director of the military-associated Peng Cheng Laboratory in Shenzhen). Let’s not forget either Xi Jinping’s personal letter of high praise to Andrew Yao, a key leader in the development of China’s STEM talent but also one of the key pillars on the Chinese side of the International Dialogues on AI Safety. Xi’s hand-chosen experts on AI seem more like the Yoshua Bengios and Geoffrey Hintons of the Chinese AI world than the Yann LeCuns. This would seem to bode well for the prospects of China making reasonable efforts to mitigate risks in AI development domestically, as well as via international coordination. However, besides the establishment of the China AI Safety and Development Association, a body seemingly mostly positioned currently as a talk shop to engage with the barbarians, the mention of “constructing an AI safety supervision and regulation system” in July 2024’s Third Plenum decision has yet to bear any substantive fruit. Whether they’re serious about risk mitigation will ultimately be decided by actions, not words.

For our recent debate on whether China is “racing towards AGI,” this study session is a continuation of ambiguity. Xi definitively did not signal any particular emphasis on AGI, and the focus on “application-orientation” and overall ecosystem development falsifies any kind of specific, singular technological goal for China’s AI policy. At the same time, it shows that AI is indeed a top priority for Xi. The intent to “comprehensively plan” compute resources and promote data sharing point towards the kind of national-scale mobilization of resources that would likely accompany a push for breakthroughs in general intelligence. And the Party clearly has very concrete ideas about how to compete in AI. Things are serious now.

What this study session should impress on us more than anything, however, is that top CCP leadership is not only thinking about AI, but may even be having relatively in-the-weeds shower thoughts about things like data resources, talent pipelines, and risk indicators. This attention may facilitate prudent action to avert catastrophe, but it is guaranteed to stimulate action to advance China’s long-standing goals of achieving supply chain independence and strategic technological breakthroughs. Their success will be decided by the cooperation or contention of millions of ordinary Chinese with their own hopes, fears and uncertainties for their lives in a world with AI, as well as the sharp strikes and careless fumbles of China’s geopolitical competitors, and the assistance of what partners they can muster.

But who is Zheng Nanning in any case?

Experts selected to deliver study session lectures are not so much chosen to bring something new of their own to the table as much as to bring exactly what Xi expects. So, what does the ideal avatar of Xi’s preferences for AI look like?

Zheng Nanning is an accomplished academic who takes his broader role in society seriously. His expertise lies particularly in computer vision and pattern recognition, but he has increasingly also been involved in robotics and industrial automation, promoting an idea of “intuitive intelligence” focused on enabling AIs to interact with an uncertain physical world based on semantic understanding. While we don't have the slides of the Politburo's study session, we do for a talk on “Machine Behavior and Embodied Intelligence” Zheng gave in 2023, with wide ranging discussion of machine perception, robustness, human-AI coordination and more. An academician in the Chinese Academy of Engineering, he has served as Chief Scientist in the Information Science field with the 863 Program, one of China’s longstanding key strategic technology development programs, as a member of the first Expert Advisory Committee for State Informatization, and as President of his alma mater Xi’an Jiaotong University. He has also been active in talent development efforts, leading the establishment of one of China’s first experimental undergraduate AI majors.

A profile on Zheng from last year paints him as empathetic and well-rounded: during a graduation speech in which he takes personal responsibility for the failings of super seniors and dropouts who didn’t graduate on time, he quoted from Lu Yao’s thousand-page novel Ordinary World, famous for depicting the struggles and aspirations of ordinary rural Chinese during the 1970s and 1980s. He is something of a fitness influencer, who claims to be able to do 50 pushups in one go and encourages students to set their hearts on “50 years of healthy work for the motherland.” Although it may seem cringe for your professor to try to get you to pick up a jogging habit, it instead reads as poignant in the context of a recent spate of early deaths of high-profile AI leaders in China which many attributed to the intense pressures of the field. Zheng even describes his educational philosophy as importing the “physical education spirit” into academics, saying that studies and research require the same tolerance for loneliness and constant pursuit of breakthroughs as athletics. Nor does he ignore imparting a moral formation on his pupils, stressing the importance of giving students “full and correct faith” in the CCP’s socialist core values.

Give him credit he practices what he preaches!

This personal history is a close match for much of China’s focus in AI currently: computer vision is key to the extensive surveillance apparatus of the Chinese state and a field where China leads the world technologically; “embodied intelligence,” and especially industrial robotics, is a core part of China’s bet on AI to produce “industrial upgrading” and escape the middle-income trap, called out as one of four priority “future industries” at the 2025 Two Sessions; talent development is a fundamental pillar underlying China’s goals in both basic research and diffusing AI into commercially valuable applications; and while Xi seeks to re-engineer China’s economy towards a high-end, high-tech structure, he is also attempting to re-engineer Chinese society towards cultural self-confidence, greater athleticism, and — as always — ubiquitous, intimate alignment with the Party.

There is one subtle surprise in Zheng’s profile, however. Though he attended Xi’an Jiaotong University for his bachelor’s and master’s, he did his PhD at Japan’s Keio University in the early 1980s. In recent years, foreign educational experience has flipped from a coveted asset in the Chinese professional landscape to arguably more of a liability, especially for roles with the government. In Xi’s increasingly nationalist China, a history of immersion in Western cultures suggests possible spiritual corruption. (In an overview of the study session, one Chinese blogger jokingly referred to recent comments from the chairwoman of appliance manufacturer Gree Electric that she refuses to hire such “returnees,” as there are “spies among them.”) We should not read too much into this, as education in Japan was very common for Zheng’s cohort and is generally viewed as less potentially problematic than education in the US. However, if Party planners had at all wanted to choose someone who was only educated domestically and is at least as impressive as Zheng, they probably easily could have. Under Xi, study session lecturers have increasingly come not even from academia, but from within the party-state system itself. It’s possible that, like the inclusion of an apparent foreigner in the audience at Xi’s recent visit to an AI hub in Shanghai, this weak signal of endorsement of international exchange was, though not actively sought, at least passively welcomed to communicate a marginally more open attitude to foreign and foreign-educated talent in AI. With the Trump administration giving anyone on a visa in the United States a new gray hair daily, China must certainly recognize an opportunity to win back diaspora STEM talent, and potentially even some from third countries.

Zheng also has a long history of giving comments on AI and society which appear to attune closely with Xi’s approach to AI. Already back in 2016 he was talking about the need to pursue both basic research and application-driven development as “grabbing with both hands.” He has discussed in detail the dilemma inherent in this two-pronged approach, where heavy pressure to produce scientific work can lead to researchers feeling “like ants on a hot wok” and choosing projects less ambitiously as a result. In line with Xi’s focus on attending to both safety and development, he has cautioned that AI is a double-edged sword, noting that technology has been used in the past to “stymie societal development and even cause disasters for humanity.” As a well-rounded scholar, he has pointed to the need for interdisciplinary contributions from the humanities, philosophy and law to address the risks of AI. He carves a middle path unfortunately rare also in the Western AI ecosystem in calling for scientists to neither buy into hype or spread unrealistic projects regarding AI, nor to engage in goalpost-shifting or downplaying advancements.

What does Zheng see for the future of AI? Since at least 2016 and as recently as 2021, he has described the field as facing three major challenges. The first is to make machines “learn without teachers.” Arguably, this has already been achieved by LLMs, pretrained on massive datasets via unsupervised learning and capable of impressively broad in-context learning with mere prompting. The second is to make machines perceive and understand the world like humans do. It’s unclear what exactly this would entail but it's hard to feel like we aren’t on the way there with, for instance, AI systems that can distinguish any arbitrary object in any scene. The final challenge in Zheng’s telling is to endow machines with “self-awareness, emotions and the ability to reflect on their situations and behaviors.” Hopefully, thoughtful AI researchers such as Zheng will reflect on the complex risks and philosophical questions which situational awareness and AI sentience may pose before the wrong edge of the sword cuts.

Jordan is in Tel Aviv for the next two weeks. It would be fun to do a meet up if there’s a critical mass! Respond to this email to connect.

Best Podcasts

Inside the Soviet Cold War Machine

Sergey Radchenko’s To Run the World explores the Cold War not as a clash of ideologies, but as a tragic and often absurd contest for prestige, legitimacy, and recognition among insecure leaders struggling to validate their power, both externally and at home. In this interview, Radchenko argues that authoritarian regimes, especially the USSR and China, pursued global influence to compensate for internal weakness.

Part two came out in April, and it’s even better than part one! In this deep-dive, Radchenko unravels how personal egos and the battle for international prestige shaped Soviet decision-making — from Khrushchev’s downfall to Brezhnev’s Vietnam gamble, the paranoid Sino-Soviet split, Nixon’s unlikely détente, and the disastrous invasion of Afghanistan. This episode asks the question, what if boredom, not grand strategy, is what starts wars?

Allied Scale and Net Assessment with Rush Doshi

This interview with Rush Doshi explores how the U.S. should strategically compete with China by leveraging partnerships with allies. While China faces real challenges like demographics and debt, Doshi argues that China’s scale, manufacturing dominance, and industrial capacity pose enduring strategic threats. He critiques both the Biden and Trump approaches to alliances: Biden’s overemphasis on persuasion and Trump’s heavy-handed use of coercion. Instead, Doshi emphasizes the need for capacity-centric statecraft, where allies help each other build economic, technological, and military strength.

EMERGENCY EDITION: Trump's Pivot to Putin + AGI and the Future of War

Defense analyst and Economist columnist Shashank Joshi alongside former Pentagon official Michael Horowitz explore the future of war. So much talk online around AI and national security flattens out to “AGI is a nuclear bomb, the first to get there wins” that it was a real treat to get to explore a richer vision of the future with two true experts. I want to point you in particular toward the second half of the episode, where we explore the Pentagon’s bureaucratic inertia, the potential for AI to reshape warfare, and the possibility that an adversary launches a first strike on the eve of AGI.

I’m hoping to expand on this show with an interview series exploring AI’s impact on national security. Too often today, debates center on “superweapons” lazily pattern-matched to the nuclear era or go in circles on cyber offense vs defense. The goal instead is to repeat the exercise Dario did for biotech in Machines of Loving Grace: deeply explore the bottlenecks and potential futures across domains like autonomy, decision-support, stealth, electronic warfare, robotics, and missile defense. Guests will be engineers and technologists who can also explore second-order operational and strategic impacts.

But this needs a sponsor in order to happen! If you work at an AI firm, defense tech, VC, university or think tank and want to help facilitate the best conversations about the future of warfare, please reach out to jordan@chinatalk.media.

China AI

Is China Racing to AGI?

This article explores whether China is truly racing toward artificial general intelligence (AGI) by staging a debate between two perspectives: the Believer, who argues that China is committed to beating the US to AGI, and the Skeptic, who contends that China’s focus remains on practical, application-driven AI development rather than AGI moonshots. While China has prioritized AI in general, the article argues that China's fragmented AI ecosystem, bureaucratic caution, and investor risk-aversion could disincentivize transformative superintelligence-focused research.

DeepSeek and Destiny: A National Vibe Shift

This guest post by Afra is a must-read. The rise of Chinese AI startup DeepSeek has ignited a nationwide wave of techno-cultural euphoria, reawakening the traditional concept of Guóyùn (國運) — the belief in a nation’s destiny. Viewed as both a technological milestone and a symbol of China’s long-awaited ascension, DeepSeek has become a focal point for patriotic pride after the trauma of COVID-19. As state rhetoric, popular culture, and grassroots sentiment converge around national destiny, DeepSeek's story reveals how technology in China is never just utilitarian — it is deeply entwined with myth-making, nationalism, and a collective need to prove that China’s time has finally come.

Manus: A DeepSeek Moment?

In this podcast discussion, Rohit Krishnan, Shawn Wang, and (now White House official!) Dean Ball analyze the potential of AI agents and the degree to which Manus matters overall. Unlike the research-heavy ethos of Western labs, Manus reflects a pragmatic, product-focused approach, prioritizing functionality over grand AGI visions. The conversation explores why Western AI giants have lagged in building compelling agents, citing safety concerns and regulatory risk aversion.

We also took a look in an article at Manus’ founder. Founded by pragmatic serial entrepreneur Xiao Hong and prodigy Ji Yichao, Monica evolved from a browser plugin into an AI agent business focused on product-market fit and international expansion, eschewing grand AGI ambitions in favor of practical use cases and aggressive data collection. With a flashy invite-only launch, Manus positions itself as a user-friendly, multilingual AI tool targeting overseas markets — a strategic move amid geopolitical tensions and growing scrutiny of Chinese tech.

Tariffs

How the Drive for Autarchy Caused WWII

In this sweeping conversation from the ChinaTalk archives, historian Nick Mulder explains how the obsession with national self-sufficiency fueled the economic insecurities that led to World War II. From the League of Nations’ failed sanctions on Italy to the Nazi quest for “raw materials freedom” and Japan’s desperate turn to war after facing an ABCD (America, Britain, China, Dutch East Indies) embargo, the episode shows how economic pressure, when mishandled, can backfire and accelerate conflict.

MAGA: A Guide for the Perplexed with Tanner Greer

In this podcast episode, Tanner Greer analyzes the chaotic dynamics of Trump’s second administration, particularly its approach to China and global trade. Greer explains Trump’s unpredictable decision-making style, his use of internal factional conflict as a management tool, and the administration's disjointed tariff policies. The conversation explores four quadrants of Trump World ideology and how adherents of each quadrant approach trade, industrial policy, and Taiwan.

Dylan Breaks Huawei and Tariffs Right

In this podcast, Dylan Patel of SemiAnalysis outlines a smarter semiconductor tariff policy aimed at boosting US manufacturing, moving supply chains out of China, and increasing America’s capital-intensive domestic production. This podcast also includes discussion from Dylan and Doug O’Laughlin about how Huawei is successfully circumventing US export controls by leveraging its vertical integration, supply chain workarounds, and large-scale system engineering.

Are We Cooked?

In this podcast, Peter Harrell, Kevin Xu, and Matt Klein discuss the chaotic implementation of Trump's new tariffs, the damage they’ll cause to international alliances, and the broader risks of US governance failures. The conversation explores the interaction between structural American strengths — innovation, entrepreneurship, and private R&D — and damage from erratic policymaking, attacks on the rule of law, and capricious foreign policy. The guests cautiously conclude that America is not yet “cooked,” though the heat is rising.

Trump's Semis Trade Policy

This CSIS Chip Chat episode explores the conflicting goals of Trump’s semiconductor tariff strategy. Bill Reinsch and Jay Goldberg highlight how poorly coordinated tariffs risk undermining enforcement of semiconductor manufacturing equipment controls, inadvertently benefit Chinese chip firms, and alienate allies needed for collective action.

Export Controls and Chinese Compute Infrastructure

Why China's Cloud Lags

This article by JS Tan analyzes why China’s cloud computing sector lags behind the US, despite massive investments from tech giants like Alibaba. While Chinese cloud providers have built extensive low-margin Infrastructure-as-a-Service (IaaS) capacity, they struggle with the high-value Platform-as-a-Service (PaaS) offerings that drive profitability in the US. Factors such as low enterprise IT spending, cheap labor, a weak enterprise software ecosystem, and the absence of professional IT consulting services hinder China’s cloud adoption beyond basic functions. Additionally, state-owned enterprises dominate China’s cloud landscape, prioritizing strategic control over compute capacity rather than developing profitable software services.

Mapping China's HBM Advances

This article by Ray Wang examines China's accelerating progress in high-bandwidth memory development, focusing on CXMT's narrowing gap with global leaders like SK Hynix, Samsung, and Micron. Despite US export controls aiming to restrict China's AI capabilities by targeting HBM supply chains, CXMT is now only 3-4 years behind in HBM technology. China's cloud and AI sectors, especially firms like Huawei, drive demand for advanced HBM, pushing CXMT to leapfrog into higher-value technologies. While CXMT’s still faces major obstacles, its rapid progress could reshape market dynamics, especially within China, and pressure global memory players on pricing, even if technological parity remains elusive.

Chinese AI Will Match America's, But Will That Matter?

While China may achieve AI model parity with the US in 2025, this article argues that America’s real advantage lies in its vastly superior compute capacity, which enables broad economic integration, innovation, and global AI leadership. Despite setbacks like TSMC's illicit chip production for Huawei, US export controls have successfully slowed China's AI progress by raising costs and limiting scale. The author contends that US policymakers should focus less on temporary model comparisons and more on leveraging compute dominance for sustained technological and economic leadership, warning against complacency and policy missteps that could squander this critical edge.

Ban the H20: Competing in the Inference Age

This article argued that China's AI ecosystem, long hampered by fragmented infrastructure, is increasingly well-positioned to dominate in an inference-heavy era, thanks to access to inference-optimized chips like NVIDIA’s H20, stockpiles of older GPUs, domestic alternatives like Huawei’s Ascend 910C, and major advances in inference efficiency. Evidently, the call to ban the H20 reached the right person, and the Trump administration has since cracked down on China-bound H20 exports.

Chinese Industry

Unitree CEO on China's Robot Revolution

In this translated interview, Unitree Robotics CEO Wang Xingxing lays out a bold, techno-optimist vision for the future of humanoid robots, predicting they will transform every industry within our lifetime — from manufacturing and construction to healthcare and even environmental restructuring at the microscopic level. While dismissing large language models as insufficient for robotics, Wang anticipates a breakthrough in general-purpose AI models tailored for robots by the end of 2025. He discusses Unitree’s competitive edge and the strengths of China’s domestic supply chains for manufacturing cost-effective robots. Despite skepticism toward Silicon Valley hype, Wang remains confident in China’s potential to lead the humanoid revolution, citing a rapid pace of development, youthful entrepreneurial energy, and abundant capital inflows.

Humanoid Robots: The Long Road Ahead

In this Q&A with Angela Shen, robotics PhD “KL Divergence” outlines the challenges and future trajectory of humanoid robotics, emphasizing that true viability hinges on achieving a “data flywheel” where robots deployed in real-world settings collect diverse, high-quality data to train better models. Industrial applications like logistics and manufacturing will likely see robots first, while home use remains at least a decade away. This article discusses what it will take to succeed in the humanoid robot race — as the field rapidly evolves, companies must navigate tough trade-offs between making robots in-house versus partnering, and focus on amassing the data, talent, and infrastructure that will ultimately determine long-term advantage.

Pharma Access with Chinese Characteristics

Angela Shen investigates China’s pharmaceutical landscape as it undergoes major government-led reform. Through programs like the National Reimbursement Drug List (NRDL) and centralized volume-based procurement (CVBP), authorities have dramatically slashed drug prices, yet concerns about the quality and effectiveness of low-cost generics persist. Corruption and inefficiencies remain endemic, while access to cutting-edge imported drugs is still limited, partly due to Beijing’s protectionist tilt toward local firms. As China’s disease burden shifts toward chronic and complex conditions, its ambitious vision of “Healthy China 2030” hinges on whether it can deliver innovative treatments through a system strained by economic and political pressures.

WWIII

Nuclearization

In this interview, Vipin Narang, Pranay Vaddi, and Junichi Fukuda explores how the Trump administration’s approach to alliances is shaking the foundations of America’s nuclear umbrella. They discuss America’s historical role in preventing ally proliferation, and analyze the “hardware” (military capability) and “software” (political will) components of deterrence. While hardware investments continue, allies like Japan, South Korea, and Poland are increasingly hedging against perceived US retrenchment. China's rapid nuclear expansion compounds these fears. The panel warns that allied proliferation would weaken U.S. security, destabilize the global order, and risk entangling the US in unwanted conflicts.

Industrial Policy and Grand Strategy

Rickover’s Lessons

This article by Charles Yang highlights the enduring relevance of Admiral Hyman Rickover’s approach to industrial leadership. Rickover, known as the “Father of the Nuclear Navy,” spearheaded the development of nuclear propulsion through a hands-on, deeply technical, and unorthodox management style that emphasized personal responsibility, rigorous training, and relentless oversight. His success lay not just in technological breakthroughs but in building a resilient industrial ecosystem.

Manufacturing’s Missing Revolution

Despite early hype, the US has largely failed to realize the promise of the Industrial Internet of Things (IIoT). In this article, Gary Wang argues that the IIoT “revolution” faltered not due to a lack of technology, but because of deep coordination failures. IIoT is a patchwork of technologies requiring complex integration across connectivity, cloud computing, cybersecurity, and machine learning — a challenge the free market struggles to solve alone. While US manufacturers remain stuck in pilot projects, China has leapfrogged ahead by using top-down industrial policy to address technical bottlenecks and platform fragmentation, enabling broad adoption of automation and AI in “dark factories.” Wang warns that similar challenges threaten US leadership in AI and quantum computing, and calls for industrial policy that supports entire ecosystems, not just headline-grabbing tech — before the US falls even further behind.

How to Build Compute in America

As US demand for AI computing infrastructure surges, a looming bottleneck in power generation threatens to stall progress. This podcast explores the challenges in energy, permitting, and financing for scaling America’s AI infrastructure. Arnab Datta, Tim Fist, and Ben Della Rocca also discuss the promise of next-generation geothermal energy as a power solution alongside gas and solar, as well as fusion and small modular reactors in the long term.

How to Compete

In this podcast episode, “Tony Stark,” author of Breaking Beijing and Ex Supra, discusses what the US must prioritize to compete with China throughout each decade of the 21st century. Stark critiques Washington's lack of a coherent China strategy and warns that without decisive action, the US risks strategic drift while China expands its global influence and military capabilities. The conversation also explores how AI will change warfare, how foreign aid supports U.S. security interests, why literacy and education are critical to military effectiveness, and how thoughtful, independent writing can shape policy discourse.

Weapons of Cold War 2.0 + 'People's War' Invasion Fleet

In this installment of Friday Bites, Kyle Chan argues that the US-China conflict is a full-scale cold war, not just a trade war, encompassing economic, technological, military, and ideological competition. Both sides are wielding a range of tools — from tariffs to cyberattacks and supply chain disruptions — without clear escalation dominance, as each action inevitably harms both parties due to deep economic interdependence. In the second half, Joseph Webster argues that Taiwan is unprepared to confront an invasion assisted by civilian ships armed with drone fleets, and suggests ways to get serious about national defense.

Media mentions

Jordan and Angela Shen coauthored an opinion piece in the Washington Post, entitled “Trump’s crackdown on foreign students is a gift to China.”

Jordan had a ton of fun on TBPN discussing business in China, Deepseek, and the search for truth:

That was way too fun so I went in for a second time, not too happy about Trump’s S&T policy.

I also went on CNBC to discuss export controls and chip smuggling.

Finally, ChinaTalk got a shout-out in FTSG’s report on tech trends in 2025 as a “pioneer and power player.”

ChinaTalk is so lucky to have such a supportive audience. Thank you for being a part of this project!

Could America pursue an abundance agenda without the threat of the PRC? And can podcasters change the world?

To discuss, ChinaTalk interviewed Ezra Klein and Derek Thompson, who need no introduction, as well as Dan Wang, who has written all those beautiful annual letters and is back in the US as a research fellow at Kotkin’s Hoover History Lab. He has an excellent book called Breakneck coming out this August, but we’re saving that show for a little later this year.

Today, our conversation covers…

• The use of China as a rhetorical device in US domestic discourse,

• Oversimplified aspects of Chinese development, and why the bipartisan consensus surrounding Beijing might fail to produce a coherent strategy,

• The abundance agenda and technocratic vs prophetic strategies for policy change,

• How to conceptualize political actors complexly, including unions, corporations, and environmental groups,

• The value of podcasting and strategies for positively impacting the modern media environment.

Listen now on iTunes, Spotify, or your favorite podcast app.

So last week we published our 400th episode! I’ll be doing a Q&A show in the next few weeks. Leave your questions in the comments.

The Thucydides Stimulus

Jordan Schneider: Let me start with a line from Ezra’s show with Tom Friedman. Friedman said, “Whether I’m writing about China from Washington, or whether I’m writing about China from China, I’m always just writing about America. My goal is to use China as my permanent Sputnik.”

Ezra, you mentioned over email that initially there was more China in “Abundance,” but in the final text, I think it primarily just plays that Sputnik role too. What do we lose when policymakers and policy advocates treat China primarily as a rhetorical device rather than a complex object of study?

Ezra Klein: Those feel like two separate questions. One reason “Abundance” is not heavily about China is that I’m perfectly aware I don’t understand China. What I do understand is the gravity that China exerts on American politics.

We have a very complex relationship of antagonism, competitiveness, and envy with China.

Going back to at least the 2010s, probably before, I’ve begun to really notice this feeling in American politics that they can build and we can’t. This became a pathway through which different kinds of bipartisan legislation that would not otherwise have been bipartisan began to emerge.

The re-emergence of industrial policy in America is 100% about China. Take China out of the equation, and there is no re-emergence of American industrial policy. It’s reasonable from the American perspective, when you’re trying to understand American politics, to understand China as an American political object, because that’s what it actually is in our discourse.

American policymakers don’t understand China at all. Most of what they think about it has a high chance of proving to be dangerously misguided. Dan will be much more expert here than I will, but I’m very skeptical of the bipartisan consensus that has emerged. Nevertheless, it’s completely trackable that China exerts a force on American politics. It has reshaped the American political consensus, often in ways that operate in the shadows because they don’t become part of the major partisan fights of modern American politics.

Derek Thompson: The only thing I would add, because I also don’t think I understand China, is that this was primarily a book about trying to deeply understand not all of America, but some very specific questions — Why can’t America build houses? Why can’t America build clean energy? What’s wrong with America’s invention agenda? These are very narrow questions about this country. The book’s scope did not include a deep dive into Chinese industrial policy or the nature of Chinese politics.

My interest in China is one of relatively blue-sky curiosity. We have this term “the Thucydides trap,” which explains how throughout history, when you have a dominant geopolitical power and a new rival emerges, that new rival discombobulates the status quo power. Modern American history has demonstrated something more like not necessarily the Thucydides trap, but the “Thucydides stimulus" — the idea that when a new rival emerges, it can inspire the existing superpower to think differently.

Sometimes that Thucydides stimulus comes through a sense of geopolitical threat. I’m very interested in the history of Sputnik and how the Sputnik moment inspired all sorts of changes in American policy in the 1950s and 1960s. But there’s also an approach that isn’t based on fear of geopolitical threat, but rather on open curiosity. How does China build its trains so quickly? How does China build so many bike paths that are beautifully integrated into the environment? What are the different ways in which Chinese versus American politicians think about designing their society such that each could learn from the other?

Without venturing too far into territory we’re saving for Dan’s wonderful book, there’s a way to engage with this idea of the Thucydides stimulus that isn’t exclusively motivated by fear, but rather by open curiosity.

Jordan Schneider: I’m going to throw it to Dan, but I love how your response to my first question shows that you guys are the masters and I’m still the student. I was debating for a whole week what my first question should be, and then I just gave it to OpenAI and asked them to squish these two ideas together. Your seeing through that is very impressive. Dan?

Dan Wang: I would always be the first person to put my hand up to say I know nothing about what’s going on in China. That is always true.

Ezra Klein: Welcome to ChinaTalk, where nobody knows anything about China.

Dan Wang: Well, Jordan knows something about China.

Jordan Schneider: Less than Dan.

Dan Wang: China is very messy. That is always my first proposition about China — it is very big, and many things are true about China all at the same time. They are a country that claims to be pursuing “socialism with Chinese characteristics,” which is still one of the most wonderful political science terms ever.

What sort of socialism is this? In my view, this is one of the most right-wing regimes in the world. A country that would make any American conservative salivate in terms of its immigration restrictions, its incredible amount of manufacturing prowess, and its enforcement of very traditional gender roles in which men have to be very macho and women have to bear children.

China is all of these things. It is also a place where there are really wonderful bike paths, specifically in Shanghai. This year, Shanghai has completed around 500 parks. By 2030, they want to create 500 more parks. It is a country that is getting better and getting worse all at the same time.

Ezra Klein: This goes back to this idea of envy — the degree to which the right envies China is fascinating. It doesn’t just want to compete with it or beat it. It’s not just afraid of it. What it wants is to be more like it.

In wanting to be more like China, the right is now overlooking virtually every advantage and competitive strength America itself has.

America’s politicians are so obsessed with trying to take manufacturing back from China, which I don’t think they have a well-thought-through approach to doing, that they look quite ready to give up America’s financial power. They seem to have reconceived of dollar dominance, which used to be called the “exorbitant privilege” because we got so many advantages from it, as some sort of terrible weakness that has hollowed out our industrial base and that we need to shatter.

Throughout history, being the power that controls the money flows has proven to be an extraordinary lever of control. But it has been recast in current New Right thinking as a sort of feminized decadence — something that “not real” countries and “not real” powers do, a distraction from the “real economy” and the “real work” of making things.

I’m not against bringing back manufacturing. I support the CHIPS Act. There are many aspects of manufacturing that I would like to bring back. But we can become so envious that it becomes hard to see our own advantages and strengths, and then make serious policy built on what we are doing well. That strikes me as one of the profound weaknesses of Washington’s approach to policymaking. It is so obsessed with what we are not doing well that it seems ready to set fire to what we are doing well.

Dan Wang: Edward Luttwak has this term “great state autism,” which he created regarding the US thinking about the Soviet Union. There is certainly an aspect, once you are a “superpower,” of becoming obsessed with the other party. You have to choose your enemies very carefully because you will end up looking quite a lot like them.

I wonder in which way the US is actually quite mimetic in thinking about how to be like the other superpower. In my sense, China — after the 2008 financial crisis, or perhaps after 2012 when Xi came into power — Beijing decided it does not really want to look too much like the US, which has been driven by Wall Street on one coast and Silicon Valley on the other in terms of economic growth.

Rather, Beijing has this purely mercantilist view, which would be recognizable to anyone in the 18th century, which is, “Let’s just make a ton of products. That is our source of power, that is our source of advantage.”

Jordan Schneider: Maybe now’s the time to bring up Gerstle and his book, which Ezra has been referencing frequently lately. His argument is that the Soviet Union as a memetic object facilitated a long 20th century of liberal governance. People don’t date it to the 1940s but to 1917 with Lenin and all the progressivism that unlocked. We saw Eisenhower buying into Social Security and domestic politics, plus the role of the USSR in the US Civil Rights movement. It’s interesting because there’s a bit of that with Todd Young and Mike Gallagher trying to implement a different version of self-strengthening than the one dominant in the White House today. I’m curious about lessons from how the Soviet Union shaped American politics. What different futures could that suggest for the US over the next few decades?

Derek Thompson: Gary Gerstle, the Cambridge historian, has this beautiful theory of political orders, which says American political eras are essentially defined by both an internal conflict and an external threat. Together, these create a consensus between parties that lasts for decades, even as headline disputes make it seem like parties are at each other’s throats.

The two big political orders are the New Deal order from the 1930s through the 1960s-70s. The initial internal crisis was the Great Depression, while the external threat was the rise of communism and socialism around the world, particularly in Europe. The New Deal order essentially synthesized these ideas by responding to the Great Depression and softening the introduction of socialism to American politics. It created a political order that was much more expansive with aggressive and muscular domestic policy, spending to reduce unemployment and poverty through the Social Security Act, employing millions of people.

As you mentioned, Dwight Eisenhower in many ways acquiesced to that political order in the 1950s. He was a strong advocate of social welfare policies and a proponent of continuing to build in America. He built the highway system.

In the 1970s and 1980s, that order broke down. Instead of the Great Depression, you had economic stagnation in the 1970s. Instead of the spectral threat of socialism in Europe, you had the direct threat of the Soviet Union and its capacities. A new political order emerged, defined by individualism rather than the collectivism of the New Deal era. Gerstle calls this the neoliberal order, which reigned over American politics from the 1980s to roughly the 2010s.

Gerstle’s theory, which I think Ezra and I subscribe to, is that we’ve seen a decline in that neoliberal order. Today’s problems cannot just be solved by cutting taxes or embracing Reaganite conservatism. Housing scarcity, building sufficient clean energy, and building the technology we invent in the US — these aren’t problems solved merely by cutting taxes and deregulating at the national level. We need more specific solutions.

Abundance liberalism is our answer to these new problems. It tries to synthesize the best of the New Deal order and the neoliberal order. The New Deal order taught us about the power of government to intervene and see what markets themselves cannot see. The neoliberal order recognized that government bureaucracy can sometimes get in its own way. We’re trying to advance a theory of progressive governance that sees how government creates rules for itself that make it harder to achieve outcomes.

Abundance liberalism finds a way to advance a muscular theory of government that says we can build houses, clean energy, and do extraordinary things with technology. But this requires identifying how we’ve written rules that get in our own way.

Dan Wang: Derek, why fold neoliberalism into this big package? Why not just embrace the New Deal agenda in its classical flavor? When I think about the New Deal, I think about the construction of power plants, homes, and broader infrastructure. They’d pack the courts if necessary because that was part of the agenda. What are we adding with the neoliberal flavor here?

Derek Thompson: When I say we’re adding a neoliberal flavor, I mean there were insights in the 1970s and 1980s about failures of the New Deal order that were accurate diagnoses. The New Deal order built extensively, often without consideration for either the voices of the marginalized or the poor, or for the environment. As a result, one legacy of the New Deal order is that the amount of construction was partially responsible for creating a groundswell against the state and against the growth machine. This empowered the legalistic revolution of the 1960s and 1970s that we now see across the country — in NEPA, in CEQA, in the proceduralism fetish, in adversarial legalism at many different levels of government. We are, in many ways, a society defined by a very activist lawyer class.

It’s important to recognize that there were legitimate legal responses to the growth machine that emerged in the 1960s and 1970s. Another part is environmentalism. The world of the 1940s and 1950s bequeathed to us by the New Deal age was truly disgusting — the rivers and air were disgusting, and tailpipe emissions from our cars were choking the planet. It was important to pass environmental rules to allow cleaner growth. We succeeded in many ways — the air is cleaner, the water is cleaner, and tailpipe emissions are a lot lower.

Now we need a new synthesis because climate change presents different environmental problems than those of the 1940s and 1950s. We need ways to build clean energy that allow people to live modern lives without choking the biosphere. That’s why we need a new synthesis rather than returning to the 1930s.

The last point: when people ask why we can’t just roll back the clock to 1932, remember that the government then, responding to the Great Depression, hired millions of people through the Works Progress Administration. Today’s US government, with its deficit and current interest rates, not to mention the political climate, simply doesn’t have the ability to hire 5 million people to build something like the Tennessee Valley Authority. It’s not going to happen.

To build houses, energy infrastructure, or transit today, we need the private sector to do much of that building. This means we need to build a legal architecture that allows the private sector to build without despoiling the planet or running roughshod over consumer or environmental interests.

Jordan Schneider: I understand your Iraq War hangover perspective — if there’s this much consensus, maybe something is fishy. But I’m frustrated, and perhaps this shows a broader frustration with how the show has evolved into “Jordan, Media Critic.” It seems to be less about learning about a field and more about booking someone to let Ezra make a point. Over the past five years, you’ve had Tom Friedman, Jessica Chenweis, and Dan, but the show with Dan wasn’t really about Chinese foreign policy. You’re leaning heavily on one side of the professional community when it’s now split roughly 90-10 in the other direction. You have a big platform, Ezra, and I’m curious how you feel about weighing in versus interrogating policy questions that aren’t directly in your area of expertise.

Ezra Klein: One pattern I notice in media criticism is that people feel you aren’t using your platform in a curious enough way when they disagree with you, but when they agree, they find it perfectly curious. I get this from many different perspectives.

When I think the consensus is wrong, my show goes counter-consensus. When I thought Joe Biden was too old to run again while most people believed the primary had settled it, I was willing to take heat for saying Democrats should have an open convention. In this particular case, I think the ambient coverage of China on my show — except for a few specific China episodes, of which there have only been three or four — reflects the broadly speaking Washington consensus that we need industrial policy to recapture major industrial sectors from China.

What worries me most, having covered Washington for a long time, is when things become unsayable without good evidence for why they shouldn’t be said. I don’t mean within the professional China debate community, but within the political community of people who need to win elections and advance their careers. I’m not a dove who thinks there are no problems with the Chinese Communist Party, but I’m not convinced the relentless buildup of antagonism and increasingly harsh policies is working — or that we’re even evaluating if they’re working.

My show isn’t unwilling to give air to mainstream perspectives. I had Ben Buchanan on AI policy and Jake Sullivan, who defended efforts to deny China certain technological exports that would allow them to attain leadership. At the same time, Tom Friedman’s point — which you might call dovishness — is that Washington tends to compete with China as it was 15 years ago, failing to recognize how much manufacturing innovation they’re currently capable of. The belief that we can simply wall ourselves off or engage in a trade war and rebuild the manufacturing sector they’ve painstakingly developed over time isn’t clearly supported. Maybe it’s clear to you — I’d actually like to hear you defend whether tariff policy as currently structured will achieve that goal.

More broadly, is it so bad for the world if China makes many solar panels and EVs? If climate change is as significant a problem as I believe, that rapidly accelerates the dispersion of renewable energy and electric vehicles. I understand why the Biden administration made its decisions. I’m not even sure they’re wrong, but I would like to hear the other side argued more publicly.

Similarly, in AI governance, is the race dynamic between our countries beneficial? If you have concerns about AI safety — and I still do, even as it’s become somewhat gauche to express them — the fact that both countries are willing to disregard other concerns to beat each other to usable AGI should raise concerns about the structure of development emerging on both sides.

I don’t see the preconditions for wise policymaking here. The absence of those preconditions is one reason the Trump administration, in its careening effort to construct some kind of sensible tariff policy, retreated from an all-out trade war with the entire world to focusing on China. They thought that would be more defensible, but did they plan for it? Did they think through the outcomes?

I’d turn this question back to you, Jordan: Do you think the current direction of US-China conversation is leading us toward good policy?

Jordan Schneider: Obviously not, but that feels like a straw man. You’re asking me to defend Trump’s tariffs.

Ezra Klein: Hold on. I want to push this because you brought it up. The Friedman conversation is about Trump’s tariffs. It’s a view that we are getting these policies because of the views that have begun to take hold, at least in the Republican Party, about China.

Jordan Schneider: The issue is that Jake Sullivan and Ben Buchanan are not people who have spent much time reading party documents. Fundamental to all of this is understanding the system and its intentions because China is relatively equal to the US from a national power perspective and will remain so for the coming decades, regardless of what we do. Understanding Xi and the broader system is a very important intellectual foundation that deserves rigorous interrogation. Tom Friedman spending a week at the Huawei Campus won’t necessarily give you that.

Ezra Klein: But this is a trade policy conversation. What would change in your trade policy from understanding their objectives? Let’s have the concrete conversation here. I agree my show hasn’t dug into party documents — there are reasons for that, but regardless, we haven’t done it. In terms of what we’re trying to examine, which is the output of US policy and whether it will achieve our goals, what do you think is being missed? What premise needs to be inserted?

Jordan Schneider: What are we missing? The world has the potential to go in different directions. It’s more difficult in the context of Trump because I’m not sure that China being a boogeyman is what’s empowering Trump. Yes, at some level it helped him get elected, but he could say whatever he wanted because he has this hypnotic control of the Republican Party. I don’t know why I was scared of this show, Dan, because I don’t have podcast hosts who just throw this stuff back at me.

Ezra Klein: I’m not —

Derek Thompson: Can I ask a version of the question?

Ezra Klein: Yeah.

Derek Thompson: Let’s say that I am a dedicated listener of the Ezra Klein show. I just mainline the podcast, and his ideas become my ideology. What is my ideology missing right now regarding China?

Jordan Schneider: That China is an ideologically driven system. The world in which China is able to more dramatically reshape the global balance of power over a 15-year horizon is one we should be really concerned about. I have a deep discomfort with the timeline where America and its friends are waning relative to China on a multi-decadal horizon.

It is not possible to know with a high degree of confidence what China really wants, but there’s at least a 25% chance that the Chinese government we get in the coming decades is a deeply Leninist and expansionary one. That is a scary timeline that the world needs to price in and prepare for.

That’s for the modal Ezra listener. For the modal DC Hawk, I’d just say that Chinese people are people too.

Share

Dan Wang: I definitely want to defend the dulcet tones of both Ezra and Derek, but as an amateur member of the community of China watchers, there are debates that aren’t easily resolved. For example, a question I would pose to US policymakers would be: Do you judge it is in America’s interest that China is richer, or is America better off if China is poorer? Having that answer would help structure many subsequent policy choices.

There is debate within the China community about how expansionist China is. They certainly want Taiwan — no question there. But is the next step that they want to take Vietnam, Philippines, as well as Japan? People are extensively debating this. When we can answer these more technocratic questions and reach some agreement, many things become easier.

This isn’t about Ezra’s show, but in the US there aren’t many experts really trying to debate and resolve these questions. In my field studying Chinese technology development and manufacturing, policymakers frequently use the laziest trope that China got where it is totally through stealing. This is easily disprovable, yet we hear it all the time. As long as we can’t move beyond these tropes, it becomes much more difficult to resolve even the harder questions.

Ezra Klein: It’s funny because I definitely didn’t expect to come on the show and articulate my own quite weak philosophy of China. But this goes back to what Tom was saying and what I’ve said in the book — this is a way of talking about America. Jordan, if your disagreement with me is that you think I’m excited for a world in 15 years where America has broken substantially and China has reshaped the global order, let me set your mind at ease. As dulcet as my tones may be, I don’t want that world. What I want is a world where we try to achieve our goals through sensible policy aims.

My views are actually quite weak on many of these things. There are areas where I have very strong views about how America should build more and faster. A big portion of the book Derek and I wrote is fundamentally motivated, as we say at the end, by competition with China. We believe we won’t continue thriving as a nation in terms of our own strength if we don’t get better at manufacturing, construction, deployment, innovation, and cyclical experimental policy. There’s something for us to learn and compete with there.

On the narrower level, there’s a view that has taken hold in Washington that some version of decoupling is the way forward. One place where I’m uncertain — not certain I disagree, but the conventional view is so dominant that I’m more interested in the counter-argument — is Tom’s argument from the Huawei campus and his other experiences. He suggests we should do with China in the 2020s what we did with Japan in the 1980s and 1990s when they were outcompeting us on cars: create joint ventures in America where we develop their technological and manufacturing processes and embed them in our own companies. China did this with us too.

In Washington, this is considered virtually unsayable. I’d like to hear a better argument against it than I’ve heard because it’s not obvious that our current approach will accelerate the sophistication of our manufacturing chains.

My view is similar to Dan’s — I’d like us to have more precise conversations about means and ends. But that’s difficult in the current political atmosphere where you have to out-compete others to be symbolically tough or hawkish.

The Taiwan problem and the Indonesia question are both very severe issues somewhat beyond my ability to address confidently. America has a very unclear internal stance on whether we would actually go to war to defend Taiwan — people don’t want to answer that, and I don’t even know what I think the answer should be.

Regarding what we need to do to accelerate our manufacturing and innovative ecosystems, the question of whether we should be decoupling or trying to couple and do tech transfer, engaging in more direct competition with products like Chinese EVs while heavily subsidizing our own industries with clear goals — that doesn’t seem completely crazy to me.

Abundance Media

Jordan Schneider: Maybe the distinction is between the technocratic and the prophetic. Once discussions turn into US-China World War III prophecy, that’s where all of us get frustrated. The way you both approached abundance — you started on the technocratic side of writing. There’s now more Frederick Douglass and MLK elements — creating a vision with moral force behind it to transform policy ideas into something rhetorical that resonates. I’m curious how you think about broader theories of policy change, technocratic versus prophetic voices, and how you grappled with that in framing the book and discussing it on your media tour.

Derek Thompson: The book is both poetry and prose. We have an introduction and conclusion with sentences that I described to Ezra as aphoristic and epigrammatic — sentences that wanted to be underlined, that wanted to have that glow on Kindle when 10,000 people highlight them. It’s strange to write for that piece of coding to become illuminated in the text, but that’s writing in a memetic way — a very 2025 approach. We want to write sentences people will quote and remember, and use terms that will infect the software of people’s minds. People talk about an abundance agenda now.

Neither Ezra nor I are trying to be shy about wanting aspects of this book to be highly memetic while other aspects are admittedly and painstakingly technical. The analysis of Tahanan (the affordable housing complex in San Francisco), the analysis of funding opportunities for chips, the history of the NIH in America, and the development of solar technology — these are highly technical sections. We’re not just trying to be pie-in-the-sky poets. We really want to understand how the world works, how government works, because there’s no way to understand how to make it work better without understanding the thing in the first place.

During the podcast and speaking tour, I’ve repeatedly been asked how we expect this book to make contact with the 2026 midterms or the 2028 presidential election cycle. My feeling, which I deeply believe, is that nobody votes for books. Even in American history, where you could weakly argue a president was influenced by a book, people are still voting for that person. People vote for people. FDR voters didn’t elect John Maynard Keynes in 1932 — they elected someone inspired by Keynes. No one elected Milton Friedman in 1980, but Ronald Reagan was inspired by Friedman.

The job of writers, podcasters, and public intellectuals is this — in a crisis, and crises will come perhaps twice as fast in this decade as previous ones, people in power reach for ideas that are on the shelf around them. They throw out an arm and say, “What’s the idea nearest to me that can help explain this moment of chaos?"

What we’re trying to do with a book like this is stock the shelves with better ideas, allowing Democrats to respond to housing crises in San Francisco or Los Angeles, clean energy construction challenges in Massachusetts, or the general problem that America invents many things but can’t seem to build them the way the Chinese can. How do we resolve these observed crises? You reach for ideas on the shelf, and abundance is trying to be on that shelf.

We are trying to influence people, but we understand no one’s ultimately going to vote for a book. The world changes because of people filling out institutions — ultimately a result of personalities we can’t directly influence.

Jordan Schneider: Ezra, do you want to address the “Ezra as America’s only real rabbi” angle? I felt like I got more substantive Israel content after October 7th from you than anywhere else. You started doing very wonky Obamacare analysis, and now there’s a values discussion that you inject into more people’s lives than perhaps anyone else in the country.

Ezra Klein: I appreciate that. The reputation I developed in my wonk blog days, in my Obamacare days, as somebody who only cared about appendix tables and CBO documents was never true. I do care about appendix tables and CBO documents and the granular sections Derek described. I believe you need the narrow, granular texture of things to understand them.

I also believe that moments and eras have zeitgeists and values. I’m probably more of a mystic than people realize — although a coming podcast with Ross Douthat, probably out by the time this airs, will reveal some of that. I try to be honest about my own reactions and struggles with the moments we’re in. Like anyone else, my reaction in many of these moments is emotional.

Something I believe strongly about my work is that if you aren’t making space for the emotional layer of reaction, you aren’t making space for the reaction itself. If you can’t speak to where the audience is emotionally — which may or may not be where you are — it will be very hard to get them to listen. If people don’t feel understood, they won’t listen to you, and maybe they shouldn’t.

I don’t think I’m alone in doing this, but it has needed doing. I’ve always felt as a writer, and maybe as a person, that the emotional layer of moments isn’t unusually visible to me — it’s something that feels unusually invisible to others. I don’t know what it’s like to walk through the world and not be incredibly affected by the emotional currents of every room you’re in. I meet people who don’t seem as overwhelmed by this, and their lives seem blissful to me. It’s just how I experience the world, and I think it comes out in the show.

When you’re writing a piece or doing a podcast — and I’d be curious for Derek’s reflection on this because he’s very good at it too — you have the audience’s beliefs in your head. What am I arguing with? What is the structure of prevailing sentiment? How is the audience feeling at moments of high emotion, like October 7th or this period with Donald Trump? That’s part of the structure you’re engaging with. To ignore it makes no sense.

I was never a fan of the Ben Shapiro line that “facts don’t care about your feelings.” Maybe facts don’t care, but people sure care. Feelings sometimes should sensitize us to the idea that there are facts we don’t yet know or experiences we haven’t absorbed. It would be hard to find many people in journalism more interested in the wonky details of policy than I am, but perhaps because of that interest, I understand that wonkiness only goes so far and misses a lot. We never have full understanding of anything. We were saying this at the beginning with China — the idea that any of us, anybody in this debate, even anybody in China itself, can understand a country that big? I don’t understand America. You have to recognize that there are many ways of knowing.

Subscribe now

Derek Thompson: I love all of that. I agree with just about all of it. What we do — me, Ezra, Jordan, Dan — we make media. I don’t know if the full value of media is feelings, but certainly an enormous amount of the value of media is feelings, even when you read a very detailed analysis of wonky things.

Just the other day, I had a conversation with an astrophysicist about transmission spectroscopy, which is how we measure faraway starlight passing through planets 100 light years away. As that light passes through those planets’ atmospheres, we can analyze the wavelengths to determine which chemicals have blocked aspects of that starlight. By determining that, we can guess what chemicals are present on that distant planet. We can say, “That blur on a telescope — that’s ammonia, methane, dimethyl sulfide. That means that planet 100 million light years away has aquatic life, has algae."

The details are technical, but the value is wonder. The value is a feeling. Ultimately what I want from people who listen to my podcast or read this book — it’s always nice to hear someone learned something, that the takeaway was informational. That’s beautiful. I love learning and information. But information itself feels like something — it feels like a dopamine hit, the relief of anxiety that I didn’t know something and now I do, or that my worldview, my understanding of what’s possible in human life, has expanded by one unit of new fact. That’s all feeling.

The greatest compliment I can receive is “your book made me feel hope” or “your podcast made me feel wonder.” When people came up to Ezra and me for book signings, many said, “You make me feel sane.” They had five seconds to speak to someone they’d listened to for hundreds or thousands of hours, and the words that came out were “you make me feel sane.” That’s the outcome of the work.

I agree with Ezra. Not only are my emotions front of mind when I’m writing, but also the audience’s emotions. That’s true both within podcasts and across podcasts. Within podcasts, I love the tension between going deep into something technical and then zooming out to the emotional impact. Across podcasts, after doing many shows about how problematic tariffs are, I might say, “We need to talk about aliens. We have to discuss transmission spectroscopy so people can get excited about detecting alien chemicals on planets 100 light years away.” We need to balance.

This play of feelings, this idea that our work exists as a kind of symphony of emotions within and across podcasts, is extremely present for me. It’s great to hear it’s present for Ezra as well.

Dan Wang: It’s really wonderful that these dulcet tones are getting more fiery. Maybe no one votes for a book, Derek, but I would vote for podcasts. I would cast a vote for Plain English, and I would cast a vote for the Ezra Klein Show.

Jordan Schneider: Now we have a choose-your-own-adventure here. Making people feel sane and heard is lovely, but there’s also a dark version of abundance. If you weren’t, as you self-describe, “pathologically agreeable,” where you have more enemies and villains — what would that book look like? How would that be embodied?

Alternatively, you partially answered this in the past 10 minutes, but are there other failure states you worry about for yourselves over the next five or ten years that you’re trying to orient away from in your work?

Ezra Klein: Let me take the first question about dark abundance. People kept telling Derek particularly that he’s too agreeable, but they tell me that too. Matt Yglesias had this funny reaction to the book where he said, “I read this book and thought it was good, but too agreeable. Nobody would argue over it because it was too soft" — and then everybody argued over it. So maybe he was wrong.

One thing I’ve learned on the tour is that I have a different sense of how to treat coalitional politics than many people I agree with, both to my right and left. Others seem more certain that you can create a group — call it corporations, unions, or whatever — and politics is really about deciding if that’s an in-group or out-group. If it’s an out-group, you should attack it as your villain. If it’s an in-group, you should ally with them as your partner.

I view this much more situationally. Environmentalists, unions, corporations — these are very big, diverse categories with a lot of internal fractiousness, as anyone who has reported on them knows, and I have reported on all of them. Different issues have different coalition structures.

It’s not that our book doesn’t have “villains,” although we don’t think many of the people we discuss are ill-intentioned. The book has plenty of instances where you can see who is standing in the way or governing poorly, and that’s led many people to get angry. When I talk to centrists, they often ask, “Aren’t public interest unions your real enemy?” From the left, the question is more, “When will you admit that corporations and corporate power are the real enemy?"

Jordan Schneider: And that’s just not interesting.

Ezra Klein: I wouldn’t say it’s not interesting. I’ve been preparing for a conversation with Zephyr Teachout and someone else that will reflect some of this debate. I’ve been reading her work and thinking about how she views the endpoint as power itself. Her critique of many she argues with across domains is that they focus on specific issues but not on power. There’s a sense that if you can move power from one place to another, you’ve solved the problem.

But what if that relocated power is used poorly? Her confidence seems to be that if you moved power from corporations to government, it would be well used. Maybe sometimes it would, but often it wouldn’t. Government gets captured by many different groups, not just corporations.

Similarly, the new left moved power from government to individuals, planning meetings, and people who can bring lawsuits, assuming individuals would be safer repositories of power than government and should tie government up in process. Sometimes that’s true; sometimes it’s not.

Many people seem to think there’s some secret “Straussian” version of Abundance in the back of our minds where we’re extremely clear about who the villains are, but due to our agreeable nature, we edited that out. Much of Abundance is an argument about being outcome-oriented, asking what we need more of and how to get it, then being rigorous about the answer.

Sometimes that leads to identifying corporate interventions that have blocked progress. In other places, it’s local homeowners or environmental groups. There are places where unions use environmental laws for reasons unrelated to the environment, making these laws potent tools of delay. Then we can’t reform environmental law in California because these groups are so committed to leveraging it that they resist any changes, even though these laws block things we need for other reasons.

There isn’t a “dark Abundance” book sitting somewhere. We don’t have the luxury of believing in some stable equilibrium of interests that gets us what we want. It’s about being deeply committed to achieving what we promise. If certain groups stand in the way of those achievements, then on that issue, they’re not our allies, and we should try to overcome that problem.

Derek Thompson: I want to ground Ezra’s principles in this current news moment because they’re so appropriate. When I talk with people on the left — I was just on Mehdi Hasan’s show the other day — they ask versions of this question: “Why isn’t your book more anti-corporate? Why isn’t it more anti-billionaire or anti-oligarchical?"

I wish I had this framing then: I don’t see business as the permanent out-group of progressivism, and I don’t see government as necessarily the singular, appropriate wielder of democratic interests. Look at what’s happening right now. The government is trying to take away the independence of the Federal Reserve. Trump has bragged about using tariff policy to try to crash the stock market. They’re using government power to entirely reshape the economy in ways that have been terrible for manufacturing.

Meanwhile, Wall Street is begging Scott Bessent to keep the Fed independent, to not crash the economy, and to remove these tariffs to protect the manufacturing sector. If your first principle is “if Wall Street asks for it, then it’s bad,” you find yourself questioning the independence of the Federal Reserve or defending the idea of trying to disrupt the stock market through random tariffs on various countries.

It’s much more helpful to recognize that in-groups and out-groups aren’t universal — they’re situational. It’s better to reason from first principles of politics rather than assigning entire groups as allies versus enemies. I hope that came across in the book, for better or worse. I’m really against designating entire groups as permanent enemies of progress, because economics is complicated and life is complicated, and sometimes people are on both sides of issues.

Jordan Schneider: Let’s speed it up. I’m going to ask you five different questions about hosting podcasts. You can pick whichever ones you want to answer: Do you have magic words you give as a pre-recording pep talk to guests? What goes through your heads during interviews? What tricks do you use to turn around guests who are performing at a B-minus level for the first 10 minutes? What are you working on as interviewers? Are there any content ideas or initiatives you’d pursue if you weren’t affiliated with large outlets?

Derek Thompson: Let me answer the meta question about shepherding guests who aren’t giving me what I want in the moment, because it’s a really tough challenge. Sometimes you can feel the conversation slipping out of your hands.

I’m never afraid to simply step up and say, “Hey, can we go back to the very first question? This is on me. I think I understand how to work our way through this conversation, and I’m going to ask a different set of questions. If you want to play with me here, let’s do A, B, C."

I try to be relatively explicit about what I want if the conversation isn’t going as planned. Much of this is to prevent the guest from feeling it’s their fault. If they become self-conscious, they develop this metacognitive layer of thought constantly judging their thinking, which prevents them from being expressive, interesting, and fluent. I try to blame it on myself and say, “Let’s try essentially a new game. We played game number one; let’s play game number two and see how it goes.” That’s how I massage the conversation.

Regarding the guidance I give guests, I don’t want to be overly prescriptive. Ezra and I were actually talking earlier about another podcaster who is unusually prescriptive at the beginning of interviews. I want to hear guests’ natural personalities come through. However, I remind them, especially those in technical fields: “This is a podcast for a generalist audience. I want you to use jargon — because jargon actually sounds cool — but please slow down and explain everything that isn’t basic 101 material in a class you might teach."

This encourages them to recognize when they’re using specific terminology. For example, as they begin talking about transmission spectroscopy, they’ll think, “Wow, that was super multisyllabic. I should definitely slow down here.” I want the conversation to be slow, conversational, and feel like a discussion after half a beer. But I typically don’t give them that level of specificity in my instructions.

Ezra Klein: When something isn’t working, you have to try something else. The more it’s not working, the more radically you need to pivot. I don’t always succeed at this. It’s very hard to abandon the mental map you had for the conversation.

Sometimes you’re just hoping things will improve later, and you’ll edit out the beginning. Sometimes I decide, “We’re throwing away the first 12 minutes of that one.” But you have to start trying riskier strategies to knock people off their current track.

The truly challenging situation is when a guest clams up — when they’re unwilling to be as open on mic as they were in previous reporting, or they’re simply nervous. To some degree, you sometimes have to rescue them. Usually, there’s a conversational vein that will activate them. Being very sensitive to people’s energy in the moment is important.

Subscribe now

Jordan Schneider: Aside from having a better one-line takeaway for what people should think about China, do you have other feedback for China Talk? What should we be doing more of here?

Derek Thompson: I learn so much from your Substack and your show. Sometimes I feel there’s a historical layer I always appreciate — understanding how things got to where they are. People who are experts in a particular domain often forget that others might want to walk the path they walked to arrive at that moment of relative expertise.

I’m always a fan — and maybe this is just my general preference for explainer media — of stepping back and providing a little bit of the “how we got here” context. That’s not so much a criticism as an observation, but it’s a type of media I always appreciate.

Ezra Klein: You can never have too much Dan Wang. He’s been the one on the island.

Jordan Schneider: Derek’s show is called Plain English, but I’m more in the Dwarkesh mindset of just doing podcasts at my level. The audience will get 80% of it, and if needed, they can Google things. ChatGPT can now fill in the gaps for you. But I hear your point.

Let me shift topics. Ezra, I went back to your Tim Ferriss interview where you spoke about how Matt Yglesias giving you an encouraging comment when you were 19 was a big psychological boost to continue down this unusual path of writing on the internet.

I’d like all three of us to do something similar now — highlight people in the sub-10,000 Substack follower range who we think are doing excellent work.

I’ll start — , who has a weird and offbeat perspective on China and broader Asia. She went to Taiwan recently and is taking an interesting approach to technology writing.

, who writes a Substack called Sinocities, exploring regions in China. There’s so much focus on Beijing, Shanghai, and Shenzhen that examining what’s happening in other provinces is really valuable.

is great. He actually dives deep into what Huawei and BYD are doing in the broader Chinese industrial ecosystem. He can be a better guide than someone like Tom Friedman.

Lastly , who publishes less frequently but offers deep dives into Chinese economic policy and communist ideology, covering both the Soviet Union and China, which I think are excellent.

Let’s go around. Who would you like to highlight?

Dan Wang: I have only one nomination — Jordan Schneider of ChinaTalk. He brings tremendous enthusiasm, making a potentially dry topic broadly accessible. Today, we heard not only from the mystical side of Ezra but also about starlight from Derek. You don’t expect that when you tune into ChinaTalk, but that’s what you get.

Ezra Klein: To be honest, having written a book with Derek, you can’t get this guy to stop talking about starlight. There was a beautiful starseed passage that I don’t think made it into the final book — I believe because we removed some of the nuclear material. Give Derek 30 minutes, and you’re guaranteed to hear about starlight.

I’ll recommend , whose “Programmable Matter” Substack is absolutely essential. He’s one of the great minds of our time.

— her Substack is called “Eating Policy” — it goes more deeply into how policy actually works in practice than almost anyone else.

I’ll also recommend , a professor and expert in administrative law and administrative complexity. His Substack is called “Can We Still Govern?”

Dan Wang: I second Henry Farrell. I think of Henry as a giant intellectual teddy bear. Whenever I see Henry, all I want to do is grab him in my arms and squeeze.

Derek Thompson: My two recommendations come from the world of science — and , both exquisite writers and wonderful thinkers on genetics, innovation, culture, and how science actually works. I’m incredibly interested in various scientific frontiers right now and always hungry for people who can explain complex concepts clearly and memorably. They both excel at what they do.

Mood Music:

Arrian Ebrahimi is a J.D. candidate at Georgetown Law and a former Yenching Scholar at Peking University. Today, he’ll attempt to quantify China’s state equity investments in semiconductor manufacturing equipment. You can read more of Arrian’s writing on the excellent Chip Capitols Substack. Special thanks to Lily Ottinger for assisting with the charts.

The recent tariff chaos — first exempting only the most advanced semiconductors, then exempting a broad base of electronics important, followed by the current waiting period for more calculated semiconductor tariffs — should not come as a surprise.

Not only was the President sounding a clarion call for tariffs through the 2024 campaign, but rumors were circulating in Washington and foreign capitals that the then-candidate Trump’s tariff wishes would manifest as a tariff on the Chinese semiconductor content of downstream electronics imported into the U.S. A New York Times article recently confirmed those rumors by unnamed administration sources.

Targeting the Chinese chip content of electronics reflects a concern over the PRC’s semiconductor industrial policy that is not unique to President Trump, nor even to U.S. policymakers. European policymakers have also sounded the alarm over China’s allegedly subsidy-induced mature node overcapacity, and your author spoke at a European Commission event in Brussels last November to address just those concerns.

Washington and Brussels’ concerns, however, sound as they grasp in the dark for answers to one as-yet unanswered question: How much public money is the Chinese government spending on semiconductors… total?

Many studies over the past half-decade have tried to figure out how public funds flow from the various organs of the Chinese government to the semiconductor sector. However, the use of conservative methodologies has prevented scholars from uncovering numbers for the entire ecosystem. The two standard approaches are:

1. Policy Announcement Hunting: China-watching platforms have tried compiling announcements of new semiconductor incentive schemes from China’s central and local governments (see Chip Capitols here on local government programs and here on central government tax subsidies). These program compilations help explain what sorts of policy tools the Chinese government deploys, but they cannot provide even a ballpark number for the total amount of RMB invested, because the Chinese government does not have transparency standards for public expenditures in the way the U.S. does.

2. Public Company Calculations: The OECD’s seminal 2019 report on market distortions in the semiconductor industry examined the subsidies that governments around the world, including China, gave to their champions. However, the study limited itself to 21 publicly listed firms, only 2 of which were Chinese, because private companies do not have annual financial filings from which they could pull statistics on state investments and subsidies. This approach offers greater accuracy, but only captures a small slice of the Chinese public investment pie.

I set out to compile data as comprehensively as possible on Chinese equity investments, subsidy grants, and tax credits for the country's key semiconductor manufacturing equipment (SME) companies — regardless of whether they are public or private. This challenge required estimation based on the limited public statistics available for private companies, but has allowed me to amass a treasure trove of insights about the Chinese SME sector.

Estimation is critical for reaching conclusions about the macro-state of upstream Chinese chipmaking equipment firms. The SME sector is small — relatively few firms are publicly listed, and some of the most important firms, like Shanghai Micro Electronics Equipment (上海微电子) (China’s only lithography firm), are notably absent from public markets. At the other end of the spectrum, Huawei has increasingly sought to integrate itself vertically by investing in Shenzhen’s SiCarrier (深圳市新凯来技术有限公司), but public numbers are not available about that nascent company which is yet to release most of its products to the open market. Notably, Huawei does not count among the top investors of any of the public SME firms surveyed in this article, suggesting its SME investments are focused nearly exclusively on firms like SiCarrier that haven’t attracted attention from the state-backed Big Fund or institutional private investors.

Ass more Chinese SME firms go public and their financial details become available, I will invariably need to revise these findings. Nonetheless, the world deserves a first (if fuzzy) glance at the totality of China’s industrial policy for chipmaking equipment. In this first of two articles, we find that:

• Government and private-sector investments into SME firms dropped precipitously in 2022, the year of China’s COVID lockdowns, and only recovered slightly in 2023 in the wake of the U.S.’s October 2022 export controls.

• Beijing’s investment decisions have no correlation, positively or negatively, with SOE investment decisions. Their choices of which SME firms to invest in and when to invest are entirely disjointed.

• The amount of liquidity created for companies via subsidies is much smaller than the liquidity the PRC government creates via equity investments. In 2021, subsidies stood at 27% of investments, in 2022 at 31%, and in 2023 at 28%.

• Within the subsidy bucket, tax credits have fallen sharply as a tool of industrial policy, and politically maleable grants have come to occupy the majority of China’s subsidy tools.

Billions of Pandemic-Sensitive Dollars

China’s investment in SME firms peaked in 2021 at $6.27 billion (a figure that includes investments by the central government, state-owned enterprises (SOEs), and private entities). Investment then fell to a trough of $1.57 billion in 2022 during the height of China’s COVID-19 pandemic. By 2023, investment rebounded to $2.86 billion — less than half of the 2021 figure.

Although COVID-19 first spread in China in late 2019, stringent lockdown policies kept the country functioning mostly as normal until stronger strains forced policymakers to adopt a “Dynamic Zero-COVID” policy in 2022 that wreaked havoc on the country’s economy. (Your author first landed in China at the height of the Zero-COVID era in fall 2022 and remembers getting his nose swabbed every day.)

Around this time, local governments poured inordinate amounts of money into COVID testing programs and quarantine hotels, leaving the localities strapped for cash more broadly. The sharp dip in semiconductor investments in 2022 likely reflects across-the-board belt-tightening during that difficult year. This chart only categorizes investors into SME firms as those of (1) the central government (namely the Big Funds run as independent corporations with the Finance Ministry as lead investor), (2) SOEs (including state-owned banks), and (3) private (including all foreign) investors. As a result, the chart cannot isolate investments from local governments to see if the decline was also due to non-COVID-related trends in 2022. However, given the interlocking ownership by local governments and the central government of the largest SOEs, it is likely that the decline in SOE investments from $2.61 billion in 2021 to $0.73 billion in 2022 reflects local governments’ COVID-induced financial constraints.

Subscribe now

It is less likely that the decline in central government investments from $0.65 billion in 2021 to $0.16 billion in 2022 was due to COVID. The only mechanism of central government investment this research identified was the (at the time) two iterations of China’s Big Fund (大基金). As a reminder for newer readers:

• The first phase of the "Big Fund" raised 139 billion yuan ($20 billion) and invested in 23 companies across the chip industry from September 2014 to May 2018. There were 16 shareholders in the first phase of the Big Fund, among which the Ministry of Finance accounted for the largest share at 36.47%. Among projects receiving investment, chip manufacturing accounted for 67%, chip design 17%, packaging and testing 10%, and SMEs/materials 6%.

• The Big Fund’s second phase was established in October 2019, aiming to raise 201 billion yuan ($29 billion). Besides government and SOE contributors, some private companies also joined the round, but the Ministry of Finance still accounted for the largest share at 11.02%. As of March 2022, the second phase fund had announced 79 billion yuan ($11 billion) in investments in 38 companies, with 10% for design, 2.6% for packaging and testing, and 10% for equipment/materials.

Because the Big Fund investments were arranged in advance, it is more likely that the dip in investments in 2022 was coincidental, rather than due to COVID-induced restraint.

Beijing and SOEs Not in Sync

Whenever a communist country releases its 5-year or 10-year plan, Washington takes a collective gasp. These fears assume that countries like China function like centrally controlled monoliths, where policymakers in Beijing can command and control every yuan spent across its vast territory —or at least, every government yuan.

However, the idiom 山高皇帝远 (“mountains are high, and the emperor is far”) tells a different story — it is not easy to centrally manage a country as vast and diverse as China. Government actors like state-owned enterprises (SOEs) and local governments operate with their own parochial interests in mind, and Beijing cannot always afford the political capital required to bring SOEs and local governments in line with national-level industrial policy goals.

If China’s central policymakers in Zhongnanhai (中南海) and the cadres leading SOEs were politically in sync, I hypothesized that investments by the central government into SME firms would be followed by a commensurate bump in investments by SOEs into the same firms.

For each SME company receiving investments from China’s Big Funds (大基金) in 2021, 2022, and 2023, I examined the three-month periods following investments by the Big Funds to search for such a bump in SOE investment interest:

• First, I defined a “Pre-Central Stock Purchase” number as the share of total investments in each calendar year from SOE investors. This number gave me a baseline of how interested SOEs were in each particular SME firm in a given year.

• Second, I defined a “Post-Central Stock Purchase” number as the share of total investments from SOEs in the three months following each company’s receipt of Big Fund investments. This number served to show what the short-term reaction by SOE investors was to demonstrated interest in an SME firm by the central government.

• Lastly, I averaged out the “Pre-” and “Post-” numbers across all SME firms getting Big Fund investments to get each year’s SOE investment baseline and average post-Big Fund SOE investment bump.

The results show that investments by the central government’s Big Fund have no consistent correlation with SOE investment decisions. In 2022, there was a 25% decline in the share of total investments made by SOEs in the three months following Big Fund investments. In 2022, there was an 8% increase in SOE investments. And in 2023, the correlation was again negative at an 11% decline.

This inconsistency bears out on an individual company level, too: Naura (北方华创), for example, saw its SOE investments drop every year following central government investments. Tianshui Huatian (天水华天) saw increases in 2021 and 2022 and a slight decrease in 2023. Meanwhile, Piotech (拓荆科技) saw virtually no change to its SOE investments in any of the three years.

Beyond showing that central government investments do not affirmatively signal to SOEs that they should invest more or less in particular SME firms, these statistics show that central government investments do not signal anything to SOE investors.

An alternative explanation to the inconsistent investments received by companies above could be that the central government does in fact direct SOE investments behind closed doors, but just gives different investment instructions each year. Perhaps the central government is pursuing a deliberate substitution strategy, directing SOEs to invest in firms that haven’t already received central government funds, except in years of extreme financial hardship, like 2022.

If this theory were true, we would see SOEs making investment decisions in lockstep, which is not what the data suggests. To demonstrate investment disunity among SOEs, we look to the example of AMEC (中微公司), which is among the most important SME companies in the Big Fund’s investment portfolio. Out of 11 three-month periods following shifts in the central government’s investment stake in AMEC, SOE investors only responded uniformly (either buying or selling AMEC stock) in four cases — that is, they were aligned only 36% of the time.

(In the graph above, 1 represents all SOEs buying stock in the three month period following a central government investment; 0 represents SOEs selling stock for that time period; and 0.5 represents half of SOEs buying and half selling.)

For SME companies other than AMEC, there is a similar lack of cohesion. It is therefore unlikely that central policymakers were successfully orchestrating any unified strategy for SOE investment.

There appears to be no consistent correlation, positive or negative, between investment decisions by the central government in Beijing and those by the quasi-governmental SOEs spread throughout the country. The mountains are indeed tall, and the emperor is far.

Subsidies: Smaller Than Expected and Falling

Discussions around China’s industrial policy regularly talk about “subsidies” without really knowing what that means. Absolutely, the PRC government has been offering immense support to its domestic chip sector, but how has it offered this support? Through subsidies? Through state equity investments? More importantly, what does the answer to that question mean politically?

I define subsidies as comprising tax credits and direct financial grants that the Chinese central and local governments provide to semiconductor manufacturing equipment (SME) and chip manufacturing companies, while equity investments are purchases of firms’ newly issued stock to help them generate liquidity. Both are forms of industrial policy support for SME and chipmaking companies, but China’s choice between these policy tools suggests different levels of central government coordination about which companies receive help.

At their height in 2021, upstream SME firms received $0.87 billion from PRC government actors in subsidies (tax credits and grants), while they received $3.26 billion from government actors in equity investments. SME subsidies were also less than equity investments throughout the COVID lockdowns, at $0.28 billion and $0.89 billion respectively in 2022. Then, after the pandemic, both policy tools rebounded to $0.50 billion and $1.78 billion in 2023, though they remained well below their 2021 highs.

Another look at the graph above tells a story about how upstream SME firms benefit differently from subsidies than downstream chipmakers. The smaller scale of subsidies to SME firms (in orange) is not surprising, since SME firms are smaller compared to the firms in China’s much more mature chipmaking sector. The PRC SME sector is largely comprised of small (often private) companies, with the two largest among them, AMEC (中微公司) and SMEE (上海微电子), posting operating profits of only under $0.20 billion at their peak in 2022. In contrast, the downstream chipmaking sector boasts giants like SMIC (中芯国际), which posted over $2 billion in profit in that same year.

Subscribe now

Given that tax credits are calculated proportionately to companies’ operating profits, it is no surprise that upstream SMEs would receive a smaller amount of overall subsidies than their downstream counterparts. What is interesting, however, is that upstream subsidies fell precipitously from 2021 to 2022 during the COVID lockdowns, while downstream subsidies fell only slightly. Then, after the pandemic, upstream SME subsidies recovered about half their lost value, while subsidies to downstream chipmakers continued to fall gradually.

At first glance, these trends suggest that SMEs are becoming a larger target of China’s subsidy efforts compared to downstream chipmakers. To confirm this political observation, however, we will need to dive deeper into the relative share of tax credits and grants within the figures for subsidies overall. Tax credits are a relatively apolitical tool once passed because they apply mechanically to any company that satisfies certain statutory requirements, while grants are allocated on a case-by-case basis and thus react quickly to political trends. These questions are explored in the next section.

The Fall of Tax Credits and Revival of Grants

Tax credits and grants are fundamentally different subsidy tools. Governments can exercise maximum discretion with their grant allocations because they are awarded on a case-by-case basis. In addition to the overall grant numbers in the financial disclosures from which I draw my data, many companies also list the individual sources of their grant awards. For example, Naura received a total of 121.7 million RMB in subsidies in 2022, of which 30.2 million RMB came from the Beijing Municipal Party Committee Office Project (北京市委办局项目). When a grant is a relatively small proportion of a company’s total, government entities can withhold it without fearing that the recipient will be utterly destroyed.

In contrast, tax credits are given out mechanically to companies that fit the credits’ qualifications. The central government’s largest tax credit is a 15% income tax deduction for companies designated under the Management Measures for the Recognition of High Tech Enterprises program (高新技术企业认定管理办法). Certainly, it is a political decision by the PRC Ministry of Finance whether to qualify companies for a tax credit (see here for an article I wrote in The Diplomat describing China’s tax credits). But tax credit qualification is a stickier (and thus more financially consequential) decision than individual grant awards, so government actors are more hesitant to use the blunt cudgel of tax credits in reaction to moderate changes in political priorities. Subsidies, rather, are the scalpel best suited for reacting to modest political shifts.

Some interesting trends emerge in the three-year period covered by my analysis. Even though the PRC released ever larger R&D tax credits over the past few years, the amount of tax credits China has provided to the chip sector has fallen since 2021. In part, this could imply that firms’ profits declined over this period, thereby decreasing the size of their tax obligations and thus making tax credits appear less valuable.

However, the graph above demonstrates that operating profits did not decline in 2022 for the companies I studied, contrary to the decline in tax credits. This suggests that a policy shift in 2022 reduced total tax credits, even as the income tax credit stayed stable and the R&D tax credit increased.1

The other trend that becomes apparent is the oscillation of grant numbers (see the first graph in this section). 2022 saw a sharp fall in the value of subsidies apportioned via grants because of COVID-induced financial strain. Since grants can be given or withheld relatively flexibly, there was a sharp decline as soon as local governments reprioritized their resources to pandemic-prevention activities.

In 2023, the value of grants recovered while tax credits continued to fall. Knowing that the PRC chip industry’s operating profits also fell in 2023, the rise of grants that year suggests that government actors are not simply doling out support to profitable companies in their jurisdictions, but rather to companies that are a political priority. This does not mean that local governments’ choice of which chip companies to subsidize overlaps with central government priorities (the second chart in this article about lack of central government-SOE coordination in investment suggests otherwise), but the political prioritization of the chip industry as a whole does seem to withstand the sector’s economic struggles.

Methodology — Equity Investments

I analyzed the following companies for this article on equity investments. The scope of firms selected comprises (1) all of those known to have received investments from either of the first two Big Funds, as well as (2) those with the most advanced domestic technology in China in the following equipment areas: lithography, etching, deposition, implantation, epitaxy, and metrology.

Despite your author’s greatest efforts, I could not collect enough data about the not-publicly listed Shanghai Micro Electronics Equipment (SMEE) to speak confidently about its equity investments; however, subsequent charts about China’s subsidy grants and tax credits will include estimations of how SMEE benefited from those policy tools.

Firms analyzed:

• 中微公司｜Advanced Micro-Fabrication Equipment Inc. China

• 北方华创｜NAURA Technology Group Co., Ltd.

• 拓荆科技｜Piotech Inc.

• 天水华天｜Tianshui Huatian Technology Co.,Ltd.

• 长川科技｜Hangzhou Changchuan Technology Co., Ltd.

• 芯源微｜KINGSEMI Co., Ltd.

• 盛美上海｜ACM Research (Shanghai) , Inc.

• 中科飞测｜Skyverse Technology Co., Ltd.

• 华峰测控｜Beijing Huafeng Test and Control Technology Co., Ltd.

• 华海清科｜Hwatsing Technology Co., Ltd.

• 新益昌｜Shenzhen Xinyichang Technology Co., Ltd.

• 屹唐半导体｜Beijing E-Town Semiconductor Technology Co., Ltd.

• 北京烁科中科信｜Beijing Zhongkexin Electronics Equipment Co., Ltd.

• 凯世通｜Shanghai Kingstone Semiconductor Corp

• 浙江镨芯（万业企业）｜Zhejiang Praseodymium Core Electronic Technology Co., Ltd.

• 至微半导体｜Not certain about English name: Zhiwei Semiconductor

• 沛顿存储｜Payton Technology(Shenzhen) Co., Ltd.

• 东科半导体｜Dongke Semiconductor Wuxi Co., Ltd.

• 精测半导体｜Shanghai Precision Measurement Semiconductor Technology,Inc.

• 睿励科学仪器（上海）有限公司｜Ruili Scientific Instruments (Shanghai) Co., Ltd.

• 东方晶源｜Dongfang Jingyuan Electron Limited

• 合顿科技｜Hefei Payton Storage Technology Co., Ltd.

For public companies on the list, I identified their top ten equity holders (central government, SOEs, and private) per quarter from Wind (万得), a site similar to Bloomberg that aggregates financial data on public companies. I then calculated each investor’s quarterly change in position to determine how many of each company’s stocks changed hands per quarter. Then, I calculated how many new stocks each firm issued per year to determine how much new liquidity these investors provided each firm through their stock purchases. That “new liquidity” is the measure of support via equity investments.

For private companies on the list, I found public reporting on their investing rounds and categorized investors into the same three buckets (central government, SOEs, and private) as for public firms.

Methodology — Subsidies

Getting subsidy data for SME companies posed similar challenges as equity investments in that many of these firms are small and not publicly listed. To that end, I relied on liberal estimation methods.

For upstream public companies, I sourced all my data from publicly available financial reports. For upstream private companies, I tried to find at least one publicly reported statistic in Chinese media, like revenue or operating profit for each company in each year. Then, I estimated all of each company’s other stats by assuming they were proportional with the average ratios from all public companies of the same year. (For example, Shanghai Microelectronics Equipment 上海微电子was not public in 2022, but I found a report of its operating profit, which was 1.2 billion yuan. Therefore, I estimated its "statutory tax obligation" as 1.2 billion/[the average operating profits of public companies in 2022]*[the average statutory tax obligations of public companies in 2022].) This method is not accurate at the individual company level (some estimates even resulted in negative tax credits); however, it results in a reliable estimate in aggregate. More importantly, it provides macro-level insights about China’s SME subsidies that, though imperfect, can help Western government policymakers get a grasp on how much China is spending to catch up in SMEs.

Additionally, it was not enough to look only at SME companies’ financials to get a grasp of China’s countrywide support for these firms because China also subsidizes demand for semiconductor tools when it gives subsidies to the purchasers of these tools, i.e. downstream chipmakers. To that end, I examined downstream companies to estimate the “subsidized demand” for SMEs—i.e., the portion of subsidies received by downstream chipmakers that is used to purchase SMEs. I estimated the subsidized demand for each downstream company as [sum of subsidies]x[capex]/[total expenditures]. I got the underlying numbers for this section similarly as for upstream SME companies, but since most downstream chipmakers are public, I only needed to use media statistics–based estimations for two firms.

Upstream SME companies surveyed:

• Advanced Micro-Fabrication Equipment Inc. China（中微公司）

• Naura（北方华创）

• Yitang Semiconductor（屹唐半导体）

• Piotech Inc.（沈阳拓荆）

• Skyverse Technology Co., Ltd.（中科飞测）

• Shanghai Precision Measurement Semiconductor Technology, Inc.（上海精测半导体技术有限公司）

• Shanghai Microelectronics Equipment（上海微电子）

• Cetc Electronics Equipment Group Co., Ltd.（中电科电子装备集团有限公司）

• Beijing Semicore Electronics Equipment Co., Ltd.（北京烁科中科信电子装备）

• Shanghai Kingstone Semiconductor Corp（上海凯世通半导体股份有限公司）

• RSIC Scientific Instrument (Shanghai) Co., Ltd.（睿励科学仪器（上海）有限公司）

Downstream chipmakers surveyed:

• SMIC（中芯国际）

• Guoxin Micro (紫光国芯)

• AllwinnerTechnology (全志科技)

• Changsha Jingjia Microelectronics (景嘉微)

• Nations Technologies (国民技术)

• Orbit (欧比特) (航宇微)

• Shenzhen Goodix Technology (汇顶科技)

• Datang Telecom Technology (大唐电信)

• Ingenic Semiconductor (北京君正)

• Hangzhou Silan Microelectronics (士兰微)

• Sino Wealth Electronic (中颖电子)

• Qingdao Eastsoft Communication Technology (东软载波)

• GigaDevice Semiconductor (兆易创新)

• Beijing Philisense Technology (飞利信)

• Ninestar (纳思达)

• Shenzhen Kaifa Technology (深科技)

• Hua Hong Semiconductor (华虹半导体)

Can China use military force to achieve its political goals, without triggering nuclear war? To find out, ChinaTalk interviewed Fiona Cunningham, a professor of Political Science at the University of Pennsylvania and author of the new book, Under the Nuclear Shadow: China’s Information-Age Weapons in International Security.

Co-hosting today is Michael Horowitz, Penn professor who served in Biden’s Department of Defense.

We discuss…

• How to use open source PLA documents to conduct deep research,

• The evolution of Chinese defense strategy, including the impact of the third Taiwan Strait crisis,

• Nuclear modernization and China’s “no first use” policy,

• How the PLA makes decisions, including why they chose to develop cyber capabilities, anti-satellite weapons, and hypersonic missiles over proposed alternatives.

Listen now on iTunes, Spotify, or your favorite podcast app.

PLA Sleuthing

Jordan Schneider: Fiona, why are PLA studies cool and important?

Fiona Cunningham: One of the biggest issues of the day is the prospects for conflict between the United States and China. How would such a conflict unfold if it were to occur? What can both sides do to improve their chances of not only succeeding in a conflict but preventing it from taking place? US allies, countries in the region, and stakeholders worldwide want to understand these dynamics.

The PLA is also an important domestic actor in China. It plays a big role in the party’s history and serves as the ultimate backstop that keeps the Chinese Communist Party in power because it is the armed wing of the Communist Party — not a state military. This arrangement would be similar to if the Republican Party or the Democratic Party in the United States had exclusive control of the US Military.

For these reasons, we need to understand everything possible about how the PLA operates, its weapons procurement, war planning processes, and its assessment of US and other countries’ actions.

Mike Horowitz: From a US national defense perspective, this is the most important part of the most important country in the world.

Fiona Cunningham: Mike says it better than I do.

Mike Horowitz: Not better, just shorter.

Jordan Schneider: There are obvious epistemic challenges when studying an adversary communist system’s military, particularly using open-source materials. Nevertheless, you wrote 300 detailed pages with over 200 interviews and an enormous amount of documents about Chinese domestic debate and doctrinal evolution regarding nuclear war — how to prevent and avoid it. When people say, “How can you ever know anything about the PLA?” what’s your retort besides waving your book in their face?

Fiona Cunningham: Most militaries, including the PLA, must communicate internally. They teach their students and share information within the organization. At least ten years ago, you could walk into certain bookstores in China and find books with ISBNs that revealed what the PLA was researching and teaching its students about almost every aspect of warfare, not just the areas I researched for my book.

Paradoxically, with nuclear deterrence and other capabilities China uses for deterrence, they need to communicate certain aspects of these capabilities. The objective is to avoid using them while using the threat of their use to shape adversaries’ behavior. You must disclose certain information about your capabilities, usage plans, and force organization to make those threats credible. Otherwise, you won’t get the deterrent value from the resources spent to develop them.

I acknowledge there are certain gaps that open-source researchers like me can never fill. What’s in China’s actual war plans? What are the exact specifications of their weapon systems? What are the precise details of certain PLA organizations?

There are also fundamentally unknowable elements: what was in Chinese decision-makers’ minds when they made certain choices? What choices will they make in a crisis over Taiwan? How will the plans I tried to understand through doctrinal debates actually function during wartime? Some things you can know, some things you can learn from open sources, and some things remain unknowable even with the best intelligence.

Jordan Schneider: Mike, would you like to add to that?

Mike Horowitz: This is tremendously challenging research. From a research design perspective, you face enormous uncertainty about China’s military. There’s a premium on gathering as much information as possible, especially given the challenges involved in obtaining that information.

There’s a justified premium placed on those talented scholars who, despite all constraints, can find primary source documents and information. Even if incomplete, even if different from what you would get researching the UK’s or US military, you can still gather enough information to make substantive inferences about how China’s military might behave.

Jordan Schneider: The PLA does have to talk to itself, but it doesn’t necessarily have to talk to Fiona Cunningham.

Mike Horowitz: It certainly doesn’t talk to me. Actually, I was at a conference right after I left the administration and saw somebody from China, which was probably the first time I’d seen somebody from China in years due to COVID and then being in the Defense Department. The person approached me, and I said, “Hi, it’s nice to meet you. My name is Michael Horowitz.” They replied, “Hi, I know who you are. We read all of your things.” I thought, “Well, that’s scary,” and wanted to back slowly away, much like the Homer Simpson meme.

Fiona Cunningham: Michael speaks the truth. I remember picking up a set of journals in one of the university libraries while looking through recent PLA research. There were many articles on lethal autonomous weapons and military AI applications. Many contained English language footnotes citing Michael Horowitz in the Texas National Security Review. I realized I had colleagues influencing how the PLA thinks about important questions at the cutting edge of warfare. They don’t cite me, though.

Jordan Schneider: Let’s stay on that topic. It’s fair to say that Chinese defense analysts constitute a much larger industry studying America than Western defense analysts studying China. Thoughts on that?

Fiona Cunningham: Several factors contribute to this imbalance. First, the US military is recognized by the PLA and most militaries worldwide as the premier fighting force. If you’re trying to learn from the best, you’ll dedicate a lot of resources to studying what makes the US military so effective, including how America leverages its technological and other advantages.

For China, the US is not just a potential opponent in their most important conflict scenario but also the model of military excellence. They have two compelling reasons to study the United States. Michael can probably elaborate, but I believe the United States learns less from China as an example of how to do things differently and is more interested in studying the PLA to understand how one of its two most serious potential opponents might fight in the future.

Mike Horowitz: That’s certainly correct, but the US does spend enormous time and resources studying everything we can possibly find about China’s military, though from a different perspective than learning lessons.

One challenge is that China hasn’t fought a war since the late 1970s. If we’re considering what lessons the US would learn from China in a military sense, these wouldn’t be battlefield lessons since there’s no recent battlefield data. The lessons would instead concern capability development, acquisition approaches, or force structure concepts — if you could access their doctrinal manuals.

This differs from how the world is currently studying Russia and Ukraine, intensely examining everything about battlefield lessons and what they mean for the future of warfare. The US is trying to glean everything possible about China’s military, but it’s learning lessons in a fundamentally different way.

Fiona Cunningham: The language barrier is also significant. In China, most people learn English from elementary school, making it a very common second language. This removes a major barrier for Chinese analysts studying the US military.

In the United States, or my native Australia, studying Chinese isn’t everyone’s first choice for a second language because of its difficulty. There are fewer college or high school students learning Chinese now than 10-15 years ago, as the incentives have shifted from business to national security.

Beyond language, there’s the issue of available materials. The United States publishes extensive open-source information about its military thinking. There are types of materials I would love to see in Chinese about the PLA that simply aren’t available, while equivalent US materials are accessible to Chinese researchers. I’d particularly value joint publications that would give me confidence that the doctrinal debates and materials I’ve studied accurately reflect actual PLA planning.

Jordan Schneider: Let’s start with the period from the 1960s to 1984 very briefly. China experienced the Sino-Soviet split, fought some border conflicts with the Soviet Union, and nearly engaged in nuclear war. That remained the main focus for a long time — defense in depth, learning from the Soviets.

Fast forward to 1984, when Deng realized Gorbachev was serious and wouldn’t invade China anytime soon. They could reprioritize defense relative to other national priorities. Take us from there through the period when global nuclear disarmament was being considered, up to 1995-1996.

Fiona Cunningham: This is a fascinating period for Chinese defense strategy, particularly regarding nuclear weapons’ role in China’s national defense.

Starting in the mid-1980s, China downgraded the possibility of fighting a major war against one of the superpowers. This coincided with China’s economic opening and reform gathering momentum, prompting the government to redirect resources from national defense to economic development.

China also reassessed what conflicts it might face. Prior to 1988, they prepared for general war with a superpower potentially invading China. After 1988, they shifted focus to “local wars” related to territorial disputes on China’s periphery. Taiwan was only one of several potential conflicts, with uncertainty about whether the US would become involved.

Two interesting developments emerged during this period. First, China debated the role of nuclear weapons in local wars. In conventional invasions, the role of nuclear weapons seemed clearer, even though China didn’t plan to use them first to deter conventional attacks. But their utility in local conflicts was less obvious.

Harvard professor Alastair Iain Johnston wrote a landmark article about the concept of “limited nuclear deterrence,” which was being debated within the PLA in the late 1980s through the 1990s. This concept involved acquiring tactical nuclear weapons to create more options on the escalation ladder during conventional conflicts.

Since Johnston’s influential article, more sources have become available, revealing another perspective in China’s nuclear strategy debate — waiting to see what the Soviets and Americans would do, as global nuclear disarmament might make additional nuclear investments unnecessary. By 1992, China’s leaders concluded they still needed nuclear weapons because the US and Soviets wouldn’t eliminate theirs completely, but China didn’t require large numbers.

Another interesting development arose from financial pressures on China’s military, which turned to exports to sustain its defense industrial base. They developed conventional short-range ballistic missiles intended for Middle Eastern markets. When the US pressured China not to export these weapons, China’s previously nuclear-only missile forces saw an opportunity to find a role in local wars by using short-range ballistic missiles to threaten Taiwan. This was the contingent origin of China’s conventional missile force, which presents a big challenge for the US and its allies today.

Jordan Schneider: Qian Xuesen 钱学森, who most listeners will be familiar with, deserves a mention here. He was forced out of America during the McCarthy era and became the father of China’s missile program. He remained skeptical in the 80s. You have a great quote where he warned that reports heralding a “post-nuclear era” were “deceiving people and they are all false.” The man still wasn’t buying into the new world order decades after seeing the West for what he believed it really was.

Another thing worth emphasizing is the level of military downsizing during this period. They shed almost a million people from the PLA — a dramatic reorientation for any military.

Fiona Cunningham: I don’t cover this in tremendous detail in my book because others have examined it thoroughly, but China’s conventional military modernization accelerated after the Gulf War, when China developed a clearer understanding of what they would need to do.

Mike Horowitz: They observed what the US accomplished.

Fiona Cunningham: Exactly — they saw what the US did and realized that to fight future conventional wars, they needed to develop similar capabilities. This became a decades-long project for the PLA. The decision to change conventional military strategy to enable China to fight “local wars under high-technology conditions” — their strategic guideline — was inspired by watching US operations rather than perceiving a direct threat from the United States. Taylor Fravel’s book Active Defense does an excellent job of explaining that decision. It was a very influential book while I was writing my dissertation.

Jordan Schneider: Mike, can you give us the 101 on why the 1991 Gulf War was so mind-blowing to so many people?

Mike Horowitz: The 1991 Gulf War blew everyone’s mind because it revealed the “second offset” on the public stage. All those developments in stealth technology, precision strike, and advanced weapons — things we take for granted today — made their dramatic debut.

I was in middle school when that war happened and remember seeing images of precision strikes on green-screen displays — missiles hitting specific buildings in Iraq. It seemed like magic, the ability of the US military to strike targets so precisely. This capability shocked the rest of the world.

The Soviets, who were becoming Russians at that point, understood the concept but had been unable to execute it. This was fundamentally different from how people thought wars would happen toward the end of the Cold War. It demonstrated sheer technological superiority by the United States and served as a wake-up call to the PLA.

Jordan Schneider: It’s also important to mention that the US even surprised itself with its effectiveness. There were advanced projections about the casualties required to conquer Iraq. The Senate, when Congress was debating whether to authorize war, anticipated 50,000 American casualties. It turned out to be only in the three or four figures.

Mike Horowitz: You’re referring to 2003. I’m talking about 1990-91. However, in both invasions, the projected dangers to US soldiers were dramatically overestimated. In retrospect, much of that was due to operational art (force employment and the effective use of military power) and how much the American military excelled. But equally critical were defense technology breakthroughs and the ability of the American military to integrate them in ways that shocked the rest of the world.

Jordan Schneider: China realized they needed a new playbook, understanding it would be a decades-long effort to approach the capabilities of the world’s superpower in the early 90s. Fiona, take us to the 1995-96 Taiwan Strait Crisis. What triggered it, and how did it affect China’s conception of what it needed to be nationally secure?

Fiona Cunningham: The Taiwan Strait Crisis had two peaks. The first occurred in mid-1995, with another flare-up in early 1996. The initial trigger was the United States issuing a visa to Taiwan’s first democratically elected president, Lee Teng-hui, whom China viewed as agitating for independence. China saw this visa as representing American support creeping toward Taiwanese independence.

The second peak came in March 1996 when President Lee sought re-election. China attempted to influence the outcome of that election. On both occasions, China conducted military exercises across and around Taiwan, including launching some of the short-range ballistic missiles they had fortuitously acquired earlier.

Mike Horowitz: “Fortuitously acquired” sounds like a euphemism for China beginning a military buildup after the Cold War.

Fiona Cunningham: They “fortuitously acquired” these capabilities because China’s defense industry had been instructed to generate revenue. When the US blocked the export of these weapons, they became available for domestic use.

There’s an interesting sequence of events in the late 1980s. The rocket force, which at that time only operated nuclear weapons, was called into meetings with China’s leaders who asked, “What role will you play in a local war?” They responded, “We have a great idea — we’ll be armed with conventional missiles.” They put forward this proposal in the late 80s largely for organizational survival purposes. They needed a way to remain relevant in a changed environment where most of China’s potential conflicts wouldn’t involve nuclear weapons. This demonstrates classic organizational incentives for military branches to seek new roles when the threat environment changes.

Between the exercises in 1995 following Lee’s Cornell visit and those in 1996 aimed at influencing Taiwan’s presidential election, China began its five-year defense plan. In this plan, conventional missiles and what they called “Shāshǒujiǎn” 杀手锏 (Assassin’s Mace, “trump card”) weapons, primarily missile systems, received prominent attention.

A series of leadership meetings occurred from late 1995 onward after the first set of exercises. From these meetings, we can surmise that China’s leaders began to see what the rocket force had recognized in the early 1990s — these missiles could be powerful tools for intimidating adversaries and addressing a new strategic reality. If another Taiwan Strait crisis occurred — as it did in 1996 — China wouldn’t have Gulf War-equivalent conventional military capabilities to counter the United States. This would make it much more difficult for China to use force if they felt their red lines regarding Taiwan were crossed.

These missiles provided coercive leverage — a way to threaten escalation against a powerful nuclear and conventional adversary when China had few other options. Interestingly, China’s leaders determined that threatening nuclear first use, the other obvious option, was unacceptable to them.

Jordan Schneider: Let’s stay on that topic. What is the “no first use” policy? Where does it come from? How does it constrain Chinese doctrinal thinking?

Fiona Cunningham: I’ll make my best case for why I believe it operates as a constraint on the PRC even today, five or six years into a nuclear buildup.

The “no first use” policy originated in a statement accompanying China’s first nuclear weapons test in 1964, where China pledged it would not be the first to use nuclear weapons. They stated they would only use nuclear weapons if first attacked with nuclear weapons by another country. This policy converted Mao’s views about nuclear weapons into a formal stance that later became the strategy given to the Second Artillery (China’s missile forces) when they began formulating how to implement China’s nuclear weapons strategy almost a decade later.

Before 1964, many statements from Mao and other Chinese leaders indicated they needed conventional weapons for conventional conflicts and nuclear weapons to deter nuclear weapons use. Several reasons explain why China adopted this policy during the Cold War. It differentiated China from the nuclear superpowers, perhaps making its nuclear weapons less of a challenge to the Soviet Union and the United States. It also related to China’s geography — they didn’t need nuclear first use against a conventional invasion because China’s size would exhaust any invader. China could survive a conventional conflict, but a nuclear conflict presented a different scenario.

It’s worth noting that China’s nuclear policy isn’t just a military matter — it’s a civilian policy given to the military by top leaders. Military leaders cannot change China’s “no first use” policy. Because it originated with Mao and Deng, it became orthodoxy that’s difficult for even civilian leaders to change.

Looking at available doctrinal materials — though unfortunately we lack probative information for the last two decades — it’s clear the “no first use” policy constrains how China plans to use its nuclear weapons. However, there have been debates about changing it or placing conditions on it, and questions remain within China about whether other countries see it as credible.

Mike Horowitz: All that can be true. You’ve persuaded me that China’s military believes it is constrained by the “no first use” policy and that civilians must make those changes. In a crisis, depending on the stakes — particularly if Xi thought regime survival was at risk — there could be incentives for China to strike first with nuclear weapons. The question is, to what extent would the “no first use” policy constrain China’s military in a conflict where the civilians who can make policy changes are actively engaged?

Fiona Cunningham: Several points are worth noting. First, if you plan to use your capabilities in accordance with a policy like “no first use,” your ability as a rocket force to develop operations involving first-use options becomes constrained. That doesn’t mean it’s impossible, but...

Mike Horowitz: It’s not as simple as just saying “launch the missile."

Fiona Cunningham: Exactly. It would mean launching the missile outside established protocols or training. It’s possible, but would represent a big departure from doctrine.

In my book, I detail an intense debate over adding conditions to China’s “no first use” policy in the early 2000s. These debates are fascinating because we see writing in leaked Chinese missile force teaching materials discussing lowering the nuclear threshold and nuclear deterrence signaling operations. What they don’t include, however, is a nuclear first-use campaign that would follow if an adversary didn’t back down after these signals.

Mike Horowitz: That’s interesting.

Fiona Cunningham: I found a dissertation written by a missile force officer — now I believe a deputy base commander — who essentially asks what a first-use campaign would actually look like. One option he discusses is detonating a nuclear weapon in space if a conflict over Taiwan isn’t going well. Interestingly, China subsequently developed non-nuclear anti-satellite weapons.

However, official materials produced by the Second Artillery don’t describe what warning shots or nuclear first-use actions would follow these threats. We lack visibility into current debates, though other scholars have found evidence of the Rocket Force now exercising for “launch on warning.” In my decade of interviews in China about nuclear strategy issues, I noticed a change in tone between 2014 and 2016, with some experts beginning to question whether launch on warning violates a “no first use” policy.

Jordan Schneider: What is “launch on warning,” Fiona?

Fiona Cunningham: In its simplest form, launch on warning means launching once you receive warning that your adversary has launched nuclear weapons toward you. Rather than waiting until those weapons detonate on your territory, you choose to launch your nuclear weapons during the period between your adversary’s launch and the detonation on your territory.

Mike Horowitz: Can we discuss this in relation to China’s current nuclear modernization? What you’ve described presents a paradox. Launch on warning is generally considered when you’re worried about your nuclear forces being decapitated — when an enemy’s first strike might destroy your ability to retaliate. If you have secure second-strike capabilities, you can absorb a strike and still respond. Since China has been rapidly modernizing its nuclear arsenal, that should make them more secure rather than less. Why would the PLA be thinking more about launch on warning now as their nuclear capabilities become more sophisticated?

Fiona Cunningham: It depends on which capabilities deliver your secure second strike. If you rely on road-mobile land-based missiles or submarine-based ballistic missiles — nuclear missiles mounted on trucks driving around China’s vast territory, perhaps hiding in caves — you’re depending on mobility and concealment to avoid detection by adversary satellites.

If, instead, your path to survivability is through having systems that are easier for adversaries to locate, the mindset about securing a second strike differs. If your plan involves many fixed silo-based missiles that your adversary can see, you’re relying on the fact that they can’t destroy all of them due to their quantity. However, those silos contribute little to second-strike capability unless you plan to launch the missiles before enemy weapons hit them.

The launch-on-warning approach relates partly to which capabilities comprise your secure second strike, but also concerns deterrence immediacy. Consider the effectiveness of deterrence if you wait weeks to pull road-mobile missiles from caves before retaliating against Washington, D.C.

Mike Horowitz: You maintain a hair-trigger posture to increase the credibility of your threat from a signaling perspective and increase the probability of successful deterrence.

Fiona Cunningham: What surprised me in post-book interviews about China’s nuclear modernization is that Chinese experts now acknowledge the US’s improved capabilities to detect road-mobile missiles. Examining the sensors the United States plans to deploy in space, Chinese strategists reasonably worry that the US will better locate capabilities they previously considered secure and survivable.

Jordan Schneider: America had approximately 20,000 nuclear weapons toward the Cold War’s end. Many have been decommissioned, but we still maintain high four-figure numbers. China started with double-digit quantities and is now approaching an arsenal of 400-500. This presents a challenge when you lack nuclear submarines, bombers, and ICBMs to ensure national destruction capability. The concept of playing three-card monte with tens of millions of lives by driving missile trucks around rural China is simultaneously absurd and terrifying.

Mike Horowitz: Road-mobile weapons are extremely difficult to detect. This explains why every country that has faced the United States since the beginning of the Cold War — whether Iraq during the Gulf War, the Soviets, China, or potentially North Korea — has considered mobility as a solution to overwhelming American firepower.

Fiona Cunningham: Such an approach wouldn’t work in the American hinterland, but China presents a different scenario.

Mike Horowitz: The regulatory environment in the US would make that impossible.

Jordan Schneider: Perhaps this could become our trucking industry’s future. Once autonomous vehicles replace truckers, the only humans on the road might be those randomizing their drives around Nebraska.

Fiona, let’s close the loop on Chinese nuclear military modernization.

Fiona Cunningham: My book examines how nations pressure adversaries in conventional wars by threatening escalation. There’s considerable discussion about whether China’s nuclear modernization aims to create this option with its nuclear forces, especially as it develops more theater-range precision options — capabilities approaching what would be needed for credible nuclear first-use threats.

However, my conclusion is that China hasn’t necessarily decided that its nuclear modernization will replace non-nuclear capabilities as its primary source of coercive leverage in conventional conflicts. I see the drivers of China’s nuclear modernization as primarily about achieving a more robust second strike against the United States.

There’s also a political leverage component — China wants enhanced capabilities because they believe this will make the United States behave more prudently. This doesn’t necessarily involve precise calculations about nuclear exchanges or specific force posture changes to enable credible nuclear first-use threats.

My recent interviews in China suggest a psychological and political leverage rationale for modernization that doesn’t necessarily translate to posture changes that make first use more feasible or improving China’s position in a nuclear exchange. Different schools of thought within China point to different modernization rationales, but first use doesn’t clearly emerge as one of them.

Jordan Schneider: The idea that it’s worth investing to ensure nobody in the Pentagon believes they could execute a first strike on China without America suffering consequences seems reasonable. If I were a captain in the Rocket Force, I’d consider that a worthy investment. But how far that extends and what it means from a readiness perspective remains one of those “unknown unknowns” we discussed earlier.

Mike Horowitz: This might transition us to discussing conventional capabilities. What’s remarkable about China’s military development over the last decade is that they’ve rapidly modernized in every area. There was a time when we could say, “China is prioritizing this capability over that one.” Fiona’s book brilliantly discusses this earlier period when China was making some of those choices, particularly regarding conventional missiles. Now, however, the story of the last decade is essentially “everything everywhere all at once” from a Chinese military modernization perspective.

Modernizing Warfare

Jordan Schneider: I love how your book opens in the early 90s, when China essentially said, “We don’t have computers, so we’re resilient to cyber attacks.” Then we witness the informatization of warfare and doctrinal development regarding offensive cyber capabilities to threaten targets when nuclear warfare threats aren’t credible.

Fiona, give us a brief history of cyber weapons in China. How has the PLA’s thinking about their utility evolved?

Fiona Cunningham: The major turning point for offensive cyber capabilities in China — capabilities they considered using for coercion — came after the 1999 US accidental bombing of China’s embassy in Belgrade. From China’s perspective, this incident was anything but accidental. There’s a notable quote from Jiang Zemin where he basically says, “I’m really indignant. This is not a trivial event. This is a big deal, and the Chinese people cannot be bullied."

Jordan Schneider: This presents an irony, right? In 1991, everyone marveled, “These Americans can hit a window from 3,000 miles away.” Then, suddenly, we accidentally bombed the Chinese embassy. Where do you stand on the conspiracy theories about this incident, Fiona?

Fiona Cunningham: My perspective is that even the best organizations can make mistakes, including the US military. I think China’s reasoning was, “This is a highly advanced military — we don’t believe they could have made such a mistake.” Additionally, once a top leader decides something wasn’t a mistake, that becomes the official position, even if contrary evidence emerges later. It becomes difficult to undo those narratives.

Following the Belgrade embassy bombing, a series of meetings took place where Jiang Zemin instructed China’s military leaders to develop capabilities addressing their leverage deficit — something that would make future conflicts, particularly across the Taiwan Strait, too dangerous and risky for US involvement. They concluded their plan to build a massive conventional missile force wouldn’t be sufficient.

Offensive cyber operations emerged as a promising capability, partly because China’s military studied the Kosovo air war and observed Serbian militias defacing NATO web pages — not particularly sophisticated cyber activities, but they recognized how quickly and inexpensively one could create problems for a more advanced military and society.

China’s awareness of cyber operations dates back to the Gulf War, when they began studying what they called “computer virus weapons” and noticed US interest in these capabilities. However, they didn’t actively pursue this option until after the Belgrade embassy bombing. Subsequently, they allowed “a thousand flowers to bloom” (百花齊放), with everyone of significance in the PLA entering the cyber arena. This resulted in capability development on both offensive and espionage fronts, but it lacked coordination and organization.

This disorganization became problematic for China’s leaders around 2010. They observed other countries using cyber operations for military effects that demonstrated the potential they had recognized. Simultaneously, China was becoming increasingly dependent on information networks for military, social, and economic functions. Seeing these cyber domain developments, they instructed their military to adjust their plans and organization for offensive cyber operations, acknowledging that they would be “throwing stones from a glass house.”

Jordan Schneider: What happens next?

Fiona Cunningham: A process of changing China’s cyber doctrine began around 2012. The PLA held many meetings to evaluate its progress and future direction. These discussions produced revealing statements, such as “We’ve made significant progress in capabilities, but everyone is fighting their own war, so we lack coordination.” Another noteworthy comment was: “If we aren’t careful with how we plan and execute these operations, we’ll harm the national interest — yet we still want to use them for leverage purposes."

Over approximately two years, China’s PLA developed approaches to modify their cyber doctrine. Around 2014, Xi Jinping particularly pressured the PLA to reorganize. This coincided with the US indicting several PLA officers for industrial espionage. The following year, China released a white paper acknowledging their cyber defense capabilities for the first time, though they still refused to publicly admit possessing offensive capabilities.

Subscribe now

At the end of 2015, China established a new organization called the Strategic Support Force, which consolidated all the disparate elements of the PLA’s cyber enterprise into a single entity. The success of this consolidation remains unclear — in fact, China disbanded the Strategic Support Force just as I was completing my manuscript.

Jordan Schneider: Toward the end of this first period, the Obama administration attempted to engage Xi and establish norms about appropriate cyber activities. How does this international dimension interact with the Chinese government’s directives to its ambiguously organized cyber force?

Fiona Cunningham: One fascinating aspect of China’s cyber capabilities is that for most capabilities with direct strategic effects on adversary decision-makers, China’s top leaders never delegate authority to lower-level PLA commanders regarding their use. The only area where some de facto delegation occurred was in the cyber arena before the 2014-15 changes I outlined.

The activities the Obama administration complained about — PLA members hacking US corporations to obtain proprietary information for commercial advantage — appeared to occur without China’s top leaders’ authorization or oversight. The international pressure reinforced China’s trend toward consolidating their cyber forces and subjecting them to tighter oversight and command and control from the top. In some ways, the Obama administration was pushing on an open door.

For industrial espionage that supported China’s national priorities in certain sectors, the activities continued but shifted to non-PLA actors, while the PLA focused on military missions. The administration was pushing on an open door in seeing some activity decline. If you examine the timelines, major decreases in PLA cyber activity tracked by organizations like FireEye (a private cybersecurity company) coincided with Central Military Commission meetings deciding on major military reform packages.

It’s difficult to disentangle whether US threats of sanctions over cyber espionage or broader PLA reorganization trends were more influential. However, this suggests US pressure alone didn’t cause China to change its behavior.

Jordan Schneider: We’ve witnessed many high-profile actions by the Chinese government over the past decade-plus, but we haven’t seen something comparable to Stuxnet, Russian cyber attacks on Ukraine’s power grid in 2024, or the Viasat attack in 2022 — incidents where actors not only penetrated critical infrastructure but actively attempted to disrupt operations. Fiona, is this simply because China hasn’t engaged in a major war, hasn’t faced something as pressing as Iranian nuclear capabilities, or is there a broader hesitance to cause damage rather than just steal information, conduct espionage, and prepare groundwork for potential future conflicts?

Fiona Cunningham: The PLA is certainly willing to lay groundwork for disruptive operations. In late 2023, reports emerged about the Volt Typhoon intrusions, where Chinese actors — though public attribution to specific PLA units is limited — penetrated adversary critical infrastructure. This aligns perfectly with their doctrine of infiltrating an adversary’s critical infrastructure to pressure them in peacetime, crisis, or conflict — shaping their behavior through fear of system disruption.

What China hasn’t done, however, is demonstrate that if they wanted to “hit send” on these capabilities, they could effectively disrupt an adversary’s critical infrastructure. I wish I had an interview where someone explicitly explained their reasoning, but these questions are extremely difficult to raise within China. Even if you could ask, individuals might not know the answers.

This remains a puzzle, even in US analysis — why hasn’t China conducted a major public demonstration proving they can cause disruption in ways that would concern adversaries in future conflicts? The reasons likely align with China’s desire to control these capabilities very carefully to avoid blowback. Nevertheless, countries typically want to demonstrate their capabilities, especially with something as uncertain as offensive cyber operations.

Mike Horowitz: Shout out to the names that people give to things. In the official US Government release on Volt Typhoon, they also called that group Vanguard Panda, Bronze Silhouette, and Insidious Taurus, which is unbelievable.

Jordan Schneider: Are these American code names, or do the groups call themselves like Golden Panda?

Fiona Cunningham: I don’t know.

Mike Horowitz: It’s like “AKA Vanguard Panda, Bronze Silhouette.” There are a couple other names that are less funny, like “Dev-0391.”

It’s less puzzling to me why China hasn’t tried to break things. I view what China is doing in this context as building the capacity to break things. The demonstration that they are in US and other countries’ networks is a signal of capabilities, since the actual use of cyber tools to break things is pretty rare. Stuxnet is the overcited one-off in a way, and relied on a lot of very specific factors.

China demonstrated that it was deep in energy infrastructure and water infrastructure at the state level across the United States, which certainly illustrates the capacity to destroy — especially because of the way they gained access through routers, VPNs, and many common electronic devices that lots of critical infrastructure facilities and Americans have.

This points to one of the issues surrounding cyber capabilities in general and relates to what we discussed last time, Jordan, in the context of offensive cyber strikes in a world of advancing AI. There’s sometimes a tendency to think about offensive cyber as this magic thing you can press “go” on, when in reality, because the accesses are so limited and once you use them they can disappear, the incentive structure even for very competent offensive actors like China is often to hold back on breaking things. The more you use these capabilities, the more it becomes all hands on deck to completely knock you out.

I have no doubt that even in the aftermath of the Volt Typhoon revelation, China has other access points that we don’t know about. From a parochial American perspective, that’s incredibly dangerous. The questions become: under what circumstances would China try to activate that, and what would the impact be? There’s uncertainty surrounding both of those things.

Fiona Cunningham: That uncertainty is one of the reasons why it makes the threat to use them credible. It may be a total fizzle and a flop, but it might also be really bad.

Mike Horowitz: I agree completely.

Fiona Cunningham: What’s interesting to me is that China’s behavior with Volt Typhoon runs contrary to a lot of US academic discussions about the utility of cyber operations over the last five to eight years. It does raise these questions about strategy. US Cyber Command has been saying the real strategic value of cyber operations lies in the “death by a thousand cuts” under the threshold of armed conflict. But the PLA looks like it’s preparing for this “cyber Pearl Harbor” scenario.

Do you actually get the leverage that China thinks it will get from preparing for something the US has said it doesn’t really see as being that big of a problem?

Mike Horowitz: It depends on what they can actually do. Consider some of the things that Russia has allegedly done and how they’ve disrupted Americans. Part of this depends on what the goals are.

There’s “death by a thousand cuts,” where cyber is an enabler to other kinds of operations. This might say something about some of the academic literature on cyber, but we don’t need to discuss that now.

There’s also cyber as a disruption and illustration of possibilities for cost imposition, precisely because there’s uncertainty. It might fizzle, but it might be really important. The theory would be that if you could create a little disruption to the lives of average Americans during, say, a Taiwan crisis, it would bring the costs home in a different way.

The question becomes, how much disruption, and what would be the ultimate impact? If what you’re trying to do is influence American behavior — perhaps more so than affecting US military capacity in the Indo-Pacific — and you’re trying to influence public attitudes and perceptions, no one really knows how that would go, including us and including them.

Jordan Schneider: That’s a key factor with a lot of this. I like your framework of brinkmanship — what are we doing with our force posture? Are we engaging in brinkmanship? Are we doing calibrated escalation? Or are we actually preparing to fight a war?

The problem is that the more forward you are with these capabilities, the more likely you are to get America to take it seriously. When China starts taking cyber seriously, as we see with Volt Typhoon, that leads to more awareness, investments, and local water plant owners updating their systems.

The same applies in the Taiwan context. Do you really think that turning off the lights in Texas is going to turn out better for Beijing? This is almost like Japan’s 1941 logic — “They’ll really want to make peace with us after we bomb Pearl Harbor."

It could go both ways, depending on how focused you are, what timelines you’re working with, and how “feminized” you think America is — to use Putin’s term, not mine. These factors impact the way you’re going to think about what you show and what you don’t do on the world stage, which makes this all very tricky.

Mike Horowitz: Just to be clear, countries have made that mistake about the United States forever. Saddam Hussein made similar assumptions. Bin Laden did as well. Foreign powers often forget that the US can be relentless. When someone attacks the US hard, we respond with full force and persistence.

Arguably, those who understand that we’re somewhat unpredictable might find that beneficial for deterrence in the classical sense. I’m curious about Fiona’s thoughts on how China perceives us in this context. I encountered this topic frequently in defense conversations over recent years. Does China assume that if they attack the US, we’ll respond with everything we have? Or do they believe they could cause enough pain that the US would back down?

Fiona Cunningham: Three factors likely matter — timing, stakes, and nuclear weapons.

Regarding timing, if you strike the US hard during an active conflict, the consequences would be severe because the United States would be fully committed. However, if that threat exists before or during a crisis, it might encourage caution and prudence. China’s strategic deterrence approach encompasses wartime planning but is also designed to influence US behavior during peacetime.

From China’s perspective, there’s an imbalance in what’s at stake. Though not explicitly stated in much of the literature I’ve studied, Taiwan represents an immediate, tangible interest deeply connected to the Communist Party’s sense of mission and legitimacy. While China can tolerate the current situation of de facto separation as it has for years, for the United States, Taiwan is an island without a formal treaty alliance, far from the US homeland, and not essential to America’s territorial integrity. The outcome of any Taiwan conflict would matter more to China. The US can have strong interests too, but they’re more diffuse and indirect, whereas China’s interests are specific and direct — Taiwan represents unfinished business from the Chinese Civil War rather than America’s global position and alliance structure.

Finally, nobody wants to fight a nuclear war. Many of China’s information-age weapons — non-nuclear weapons with strategic effects — are designed to push the US to the threshold of nuclear weapons use and then call its bluff. The open question remains whether America’s seemingly unpredictable nature extends to nuclear weapons use. That’s the million-dollar question.

Jordan Schneider: Another point supporting that perspective is that we’ve now seen two consecutive presidents take Putin’s nuclear threats seriously, even though those threats were less credible and not directly aligned with major US strategic interests. This clearly impacted Biden’s calculations. With Trump, throughout his campaign and in his meeting with Zelensky, he repeatedly warned, “You’re flirting with World War III.” It’s one thing to exhibit Jacksonian intensity when fighting overseas wars, but it’s entirely different when you genuinely fear that Los Angeles could be destroyed.

Mike Horowitz: There’s something interesting about how presidents conceptualize nuclear war and risk. Looking at Trump’s public comments, they suggest he might be even more concerned about nuclear risk than some other presidents. While every president since Truman has worried about nuclear war, Trump seems particularly focused on the dangers of nuclear escalation.

Jordan Schneider: It will be interesting when we eventually see a generational change — a president who didn’t grow up during the 60s, 70s, and 80s when nuclear concerns were front and center. But returning to the PLA — Fiona, let’s discuss counter-space capabilities. We’ve witnessed several interesting and explicit demonstrations of these capabilities from China in recent years.

Fiona Cunningham: I should share a personal anecdote — one reason I became interested in writing this book was China’s anti-satellite weapons test in 2007, when they destroyed an aging weather satellite with a conventional missile. I was in college at the time and remember seeing it on the front page of the Sydney Morning Herald, wondering what it meant. The book actually opens with this ASAT weapons test, perhaps for that unwritten reason.

China pursued counter-space weapons capability for coercive purposes — to exert pressure during a potential conventional conflict with the US over Taiwan, particularly after the Belgrade embassy bombing. This decision was difficult to pinpoint precisely, but based on Jiang Zemin’s speeches, it was likely made between late 2000 and late 2002.

China recognized that anti-satellite weapons could disrupt an adversary’s space capabilities and impede the US military’s long-term objective of achieving space control. They developed a range of counter-space capabilities, including lasers that could dazzle optical sensors on US satellites. In fact, they dazzled a US National Reconnaissance Office satellite around 2005, then tested the missile that destroyed a satellite and created substantial debris, generating international criticism.

Their goal was to develop various weapons that could attack US satellites in orbit and disrupt data transmission between those satellites and Earth. These range from non-kinetic reversible effects, such as lasers and electronic warfare jamming, to irreversible effects like completely destroying satellites. More recently, China has developed co-orbital capabilities — satellites that can maneuver to grab onto other satellites, tow them elsewhere, collide with them, or position themselves close enough to jam or dazzle from proximity.

What’s curious is that China didn’t pursue counter-space capabilities earlier. There’s a common perception that China identified attacking US satellites as a valuable coercive tool following the Gulf War. However, when examining the sources, China didn’t begin contemplating counter-space attacks until the late 1990s, after the 1995-96 Taiwan Strait crisis. Before that, in the context of the Gulf War, China primarily recognized that to fight like the United States, it would need its own military satellites to guide weapons and confirm target destruction.

This desire to build military support capabilities in space constrained China’s approach to counter-space weapons. They consistently emphasized that hostilities in space should be limited, unlike their apparent willingness to consider more extensive use of conventional missiles. That’s the doctrinal capability narrative that has emerged.

Jordan Schneider: The pattern is fascinating — China demonstrates a capability, America becomes alarmed and develops countermeasures. This is the challenge with sub-nuclear capabilities — the more you reveal, the more your adversary adapts. By the end of both the cyber and counter-space chapters, you quote Chinese analysts essentially saying, “This didn’t achieve exactly what we wanted. America now has ten times more capabilities than we anticipated a decade ago.” What are your thoughts on this dynamic, Fiona?

Share

Fiona Cunningham: While this dynamic matters, it may not be as important as we might think. Many of these capabilities were designed to give China credible leverage when it was considerably weaker than it is today. Even if these capabilities become less effective over 20-30 years, they still provided China with methods to deter and coerce the United States when China’s conventional capabilities posed minimal threat.

Currently, I don’t believe China’s conventional capabilities threaten decisive victory against the United States in the Indo-Pacific, but they’re substantially improved compared to 20-30 years ago. The US military would face much greater challenges fighting China today than it would have two decades ago.

China’s pursuit of these capabilities has pushed the US to develop countermeasures, which in some ways validates China’s strategic choices — these capabilities delivered on their promise by forcing the US to reconsider how it approaches conflicts and organizes capabilities across different domains. The cyber changes we discussed earlier exemplify this. Additionally, the US withdrew from the Intermediate Nuclear Forces Treaty partly due to China’s conventional missile force growth, creating new questions for China about how US intermediate-range conventional missiles in the region might affect its security.

Similarly, China’s counter-space capabilities have compelled the United States to rethink its satellite architecture. The US formerly relied on “big juicy targets” — large satellites for sensing and signals intelligence that supported conventional conflicts and precision strikes. Now the United States deploys smaller, less vulnerable satellites to complement these larger assets.

Mike Horowitz: Or haven’t yet.

Fiona Cunningham: Right. This adaptation means China faces many more targets if it wants to use counter-space capabilities effectively. It can no longer neutralize US capabilities by destroying just a handful of sensing satellites. Furthermore, the United States now discusses its own counter-space capabilities much more openly.

China has triggered a big reaction in the space domain. Chinese strategists are now concerned about several US initiatives, including distributed satellite constellations, US counter-space capabilities, satellites that monitor activities in orbit, and US partnerships with commercial space companies that enhance American capabilities. They’re also increasingly worried about the potential for nuclear escalation resulting from space-based conflicts — a new development over the past five years.

Mike Horowitz: Another way to think about this more broadly is the reveal-conceal dilemma when developing capabilities. Everything Fiona said is correct. Generally, you reveal capabilities to deter or coerce, while you conceal capabilities for actual conflict. China’s actions in space represent revelation specifically to impose costs, which they have accomplished by driving many changes to US space architecture — both implemented and planned.

However, some areas are difficult to conceal even when that’s the intention. Space is arguably one such domain — concealment is easier in some ways but harder in others. We don’t know exactly what China’s intentions are when they do things like dazzle our assets. Are they purposely revealing capabilities, or would they prefer to conceal these tests but can’t? Their testing options differ from terrestrial missiles, which can sometimes be tested in relative isolation. This fundamentally changes the dynamic.

Fiona Cunningham: To add one more element, PLA space deterrence doctrine often describes a tiered approach. You begin by signaling your capabilities or repositioning assets to make them visible. Mike’s point is particularly relevant because even when examining PRC doctrine, you can see there’s a strategic place for movement and revelation of capabilities intended to send a message. Determining whether they’re following established doctrine or simply couldn’t hide what they were testing remains a key challenge.

Jordan Schneider: Let’s discuss dogfighting. Mike, would you like to address this?

Mike Horowitz: Recently, a US Space Force official published an article discussing China conducting “dogfighting” in space. This terminology demonstrates how we project our understanding of one domain onto another. However, it clearly illustrates what Fiona’s book demonstrates about China’s capabilities and their testing approaches.

Despite changes over recent years, the US Space Force still feels constrained in its ability to conduct similar testing and faces questions about what would be operationally permissible. One interpretation of this discussion about Chinese “dogfighting” in space is that the US Space Force is signaling that what the PLA is doing is legitimate. The implication is that if we want to deter and defeat these tactics — rather than merely exhausting them through proliferated assets that complicate disruption of American capabilities — we need to grant the Space Force more authority to operate in space, including activities that have been considered dangerous for decades.

Fiona Cunningham: If I’m not mistaken, the report on dogfighting described a scenario where several Chinese satellites — more than three — essentially surrounded another satellite. This demonstrates advanced capabilities in rendezvous and proximity operations — the ability to position several objects close to each other in space. This relates to Mike’s point about constraints regarding how close one might be permitted to approach in these operations.

Regarding the specific terminology of “dogfighting,” I’ll note that I don’t know how you would say “dogfighting” in Chinese. This probably indicates that the term doesn’t appear in their doctrine.

Jordan Schneider: Everyone wishes they were still flying fighter jets over Korea. That’s basically the lesson of all this.

Mike Horowitz: The question for Fiona is, how sophisticated do you believe China’s conventional missile arsenal has become? China has, by my count, at least six relatively modern anti-ship missiles, including two different anti-ship ballistic missiles, something like the DF-17 anti-ship hypersonic missile, along with the YJ-class missiles. How important is this development? Coming from the Pentagon in 2024, the importance of these missiles represents one of perhaps the top five questions facing the US military today, particularly the US Navy.

Fiona Cunningham: This is an Air Force issue as well, since these same missiles can target assets on regional airfields. Examining PLA conventional missile doctrine reveals something interesting: as their forces become more precise, the described target set actually narrows. Rather than infrastructure or economic targets, they now focus on missile defenses, electronic warfare capabilities, radars, airfields, and ships.

Some advances are obvious — China’s missiles have become more accurate. Finding reliable estimates of their accuracy is challenging — how many missiles will land within a short range of their target, and how frequently they’ll hit within a certain radius of that point.

The most important change for China comes in sensing capabilities. Precision missile capability requires not just accurate missiles but also sensing systems to locate targets and confirm successful strikes. This presents a particular challenge for anti-ship ballistic missiles. Extensive analysis has examined what space-based and ground-based capabilities China might employ to locate US carriers — the key question being whether they could find and track them sufficiently for missile targeting.

Recently, China deployed an optical satellite in geosynchronous orbit that provides persistent coverage of that region, significantly enhancing their target acquisition capabilities. The black box for me is what happens between the sensor and the shooter — this can be a very difficult process. I have less visibility on whether this presents as big an issue for anti-ship ballistic missiles targeting large vessels as it would for more precise targeting scenarios.

The other question involves US countermeasures and their effectiveness against China’s attempts to hit both moving targets and fixed installations. The US could potentially use its own counter-space capabilities to disrupt Chinese sensing systems important for locating and tracking vessels. However, fixed targets like Kadena Air Force Base remain at known locations. You may not know exactly what assets are present, but you can still strike these fixed targets even with degraded sensing capability.

This leads to considering what else the US can do to disrupt China’s precision strike capability, which ultimately points toward disrupting missiles before launch. This returns us to the challenging problems of tracking mobile missiles that we discussed earlier. You can see how these issues are interconnected.

I’m not directly answering your question, Mike, because uncertainty exists about the steps in the chain that China must complete to successfully hit the more challenging US targets. Substantial uncertainty also surrounds the effectiveness of US countermeasures, both in terms of disrupting PRC capabilities and in hiding from or diverting those missiles once launched and en route to their targets.

Mike Horowitz: That’s a great answer, super helpful. What about Chinese hypersonics in general? China has tested hypersonic systems that have excited missile enthusiasts and generated concern that the US could be falling behind in this technology. Setting aside the “falling behind” narrative — and I’ll save my personal rant about hypersonics for another day —

Fiona Cunningham: I’d love to hear it — come on, we’re here.

Jordan Schneider: We’re two hours in, Mike. The people are waiting for this.

Mike Horowitz: There’s no capability that disappointed me more when I left the Pentagon than hypersonics, particularly regarding the relative value for investment. They have their place but have been somewhat overhyped considering the overall architecture of missile systems. That’s not surprising coming from me, given my advocacy for precise mass and more attritable autonomous systems.

My question for you, Fiona: How fearsome are China’s hypersonics, and how have they managed to deploy so many hypersonic systems so quickly compared to the United States?

Fiona Cunningham: The short answer to why China has been quicker than the United States: I can’t point to a specific line in a PLA manual or teaching text, but if I were to hazard a guess, it’s because this is a priority for China. Conventional missiles, and missiles in general, represent an area where China has invested a lot of effort ahead of other aspects of its conventional military modernization and certainly its nuclear modernization. When you prioritize something, you naturally progress more quickly.

It also relates to China’s specific problem set. The United States has deployed missile defense systems around China’s periphery, most visibly with the THAAD system in South Korea in 2016, which caused considerable political disruption in Northeast Asian security dynamics. China faces the challenge of defeating missile defenses. If that’s your problem at theater range, hypersonics offer a potential solution because of both their speed and maneuverability in the terminal phase, making them more difficult for missile defenses to intercept.

Regarding their effectiveness — they’ve been tested in controlled environments. Their true capabilities won’t be known until they face actual US missile defense systems, and their success will depend on how US capabilities to track hypersonic missiles progress. This represents an ongoing development with the current space sensing architecture.

This creates a cat-and-mouse dynamic where China’s investment may provide a temporary advantage, but the US can potentially close the gap — either through the precise mass approach you advocate or through countermeasures specifically designed to address the problems that hypersonics solve for China.

Mike Horowitz: To me, that’s an “and” not an “or” for the US, but that’s very helpful.

Jordan Schneider: Thinking about that exchange, I was trying to imagine how to persuade Trump to defend spending on basic research and science. This led me to recall the “super duper missile” and “invisible aircraft” comments. We’re recording this on Friday, March 21, when the NGAD, the sixth-generation fighter, is about to be announced.

My mental model suggests Trump will approve spending on things that are big, fast, and shiny — not slow and attritable. There’s an interesting tension here. Many influential tech companies like Palantir and figures connected to Trump’s circle advocate for one theory of acquisitions and victory, whereas Trump himself seems drawn to technologies you can describe with dramatic action-figure adjectives, rather than, say, cute submersible drones.

Mike Horowitz: One notable aspect of the 17 priorities that the Pentagon announced for its review of the FY26 budget is how both sophisticated systems and “one-way attack” precise mass systems were explicitly identified as investment priorities.

My instinct is that you’re probably correct. Additionally, the now-confirmed Deputy Secretary of Defense Feinberg is a strong hypersonics advocate. What we might see, due to both Feinberg’s influence and presidential preferences, is a push toward a high-low capability mix — investing in both the biggest, shiniest assets like NGAD and hypersonics, as well as more distributed mass capabilities.

Since many US capability investments over the last couple of decades have focused on that mid-tier range, this shift raises questions about those programs. As an advocate for a high-low mix for the force, I don’t necessarily find this troubling, even if the pathway there differs from what I might prefer.

Jordan Schneider: Fiona, were there alternative paths the PLA could have taken? Could they have decided in the late 80s to focus on nuclear modernization, making that threat more credible, rather than building a massive conventional force with advanced capabilities in space, cyber, and missiles? Was that a viable option? And would that approach have made China’s rise less concerning to other nations? Or do you see concerns about China as primarily related to economic growth rather than specific military capabilities?

Fiona Cunningham: That’s an excellent question. It depends on which parameters you change. What if those big incidents with the United States in the mid-1990s hadn’t occurred? In the book, I also discuss the EP-3 crisis, when a US reconnaissance aircraft and a Chinese fighter collided over Hainan Island in 2001.

Mike Horowitz: Remember how big that incident was at the time?

Fiona Cunningham: It was enormous.

Mike Horowitz: We’ve somewhat memory-holed it, especially given the current global uncertainty, but when it happened, it was a major international crisis.

Fiona Cunningham: I was in what Americans would call middle school at the time. It was another event I followed closely, similar to the ASAT test, as it occurred very early in the Bush administration. Looking back, it’s clear why my childhood dream of becoming a human rights lawyer at the UN got diverted by media coverage of these events.

Mike Horowitz: By your one true love for nuclear weapons.

Fiona Cunningham: Well, if I truly loved nuclear weapons, I’d probably study a different country. They’re not China’s favorite.

An alternative narrative might have emerged when China encountered these crises with the US that revealed its leverage deficits. China could have decided, “We have nuclear weapons. We’ve observed NATO’s approach during the Cold War. We’re watching contemporary Russian strategies in the 1990s. We’ll simply adapt our nuclear posture and follow those models,” referencing Iain Johnston’s famous “limited nuclear deterrence” article. That’s one potential pathway.

Another scenario: without these crises, China might still have developed its conventional military, but its modernization and strategy would have responded to different variables — how other countries fight wars, party unity (to reference Taylor Fravel, my advisor’s work). You might have seen conventional modernization proceed without these investments in non-nuclear strategic deterrence.

Economic factors could have reshaped China’s conventional modernization trajectory. Jiang Zemin explicitly stated in the late 1990s that China’s military modernization progress toward joining the ranks of advanced military powers — “world-class militaries,” though he didn’t use that specific term — was contingent on the country’s economic circumstances.

A third consideration: how much of the current situation stems from Xi Jinping being a different type of leader with distinct visions for China’s foreign policy and defense strategy compared to his predecessors? I tend to see more continuity than change. However, with a different leader pursuing objectives unlike those of Hu Jintao or Jiang Zemin — perhaps emphasizing greater international engagement or taking a different approach to domestic politics — China might have followed a different path.

Mike Horowitz: It’s tempting to view everything as inevitable in retrospect. My instinct is that if it hadn’t been the Taiwan Strait Crisis or the EP-3 incident, something else would have triggered similar developments. From a structural perspective, China’s rise combined with frequent US demonstrations of conventional military superiority during the Iraq invasion and subsequent conflicts created conditions where accelerated PLA conventional modernization became highly probable. The question is: what additional steps would they take beyond that?

Fiona Cunningham: That’s where politics enters the equation. Stepping back, one of the major conclusions of my book is that China’s decisions about strategic deterrence and capabilities intended to pressure the United States are all connected to political dynamics. This isn’t primarily about US capability dynamics but rather those political crises that create urgent demand for enhanced capabilities to deter the United States from engaging in future confrontations. The US doesn’t always pay sufficient attention to these political dynamics, while from China’s perspective, they’re paramount.

Jordan Schneider: Let’s close with three things, Fiona. First, recommend one PLA book for listeners who’ve made it to the end of this episode. Second, share your favorite Chinese phrase, perhaps something PLA-adjacent. Finally, which meeting in recent or older PLA history would you have liked to witness, and perhaps what organization you might have worked for during which time period.

Fiona Cunningham: I would decline working for any PLA organizations — I don’t think that uniform would have suited me well.

Regarding a meeting I would have liked to witness recently, I’m curious about why China’s military leaders decided to disband the Strategic Support Force around March or April last year. It previously housed space capabilities, cyber capabilities, and the network maintenance organization for all PLA Defense Information Networks. They announced a separate force for information networks, but almost nothing has emerged about what happened to the cyber and space components. I really want to understand why this occurred and why there’s been silence about the other elements.

For historical meetings, in approximately 1978, there was a meeting where Deng Xiaoping commented that China should continue considering tactical nuclear weapons while making decisions about the future of Chinese nuclear forces. Information about these decisions is scarce, but I’d love to know what choices led China’s leaders to still contemplate whether shorter-range, lower-yield nuclear weapons would help defend their borders against the Soviets, and why that idea eventually faded as China maintained its restrained nuclear strategy.

For a PLA book recommendation, I frequently return to an edited volume from the National Defense University titled Crossing the Strait: China's Military Prepares for War with Taiwan with editors Joel Wuthnow, Phillip Saunders, and others. It contains many valuable chapters summarizing China’s doctrine for a Taiwan conflict. Joel Wuthnow’s chapter on PLA command and control systems is particularly helpful for anyone wanting to understand China’s campaign capabilities as of a year or two ago. It exemplifies the best that PLA studies and analysis can offer regarding the dilemma we discussed at the beginning of this podcast — this remains the primary scenario and most difficult problem facing military leaders in the United States and allied countries.

My favorite Chinese phrase would be “惩戒行动” (chéngjiè xíngdòng), which appears in descriptions of China’s counter-space capabilities and roughly translates to “punitive strikes and disciplinary action” — essentially combining “punish” and “warn.” This phrase resonated with me early in developing my dissertation topic because it emphasized how differently China approaches the space domain and space deterrence compared to nuclear weapons. In many ways, it represented the variation in the dependent variable that made this not just an interesting and policy-relevant topic but one with academic merit as a political science dissertation.

Jordan Schneider: Can you use those words in a teaching or parenting context?

Fiona Cunningham: During my interviews in China, I asked what this phrase meant. People explained that “惩戒行动” referred to conflicts with punitive and disciplinary elements, like the Sino-Vietnamese War. When I asked if nuclear retaliation qualified, they said no. There’s a footnote about this in the book. No one suggested it was terminology they would use when teaching students or parenting children, so I would recommend keeping it within a military strategy context.

Jordan Schneider: Tell us more about the general atmosphere of the Chinese defense analyst community. How would you characterize them as people, using broad generalizations? And what do they gain from speaking with you?

Fiona Cunningham: When I conducted interviews for the book, mostly as a graduate student, I believe they were partly motivated by helping a student learn. The desire to assist students transcends cultural contexts. There’s also genuine interest among many Chinese experts to engage with researchers who make the effort to visit China, learn the language, and understand the strategic studies lexicon. They want to clarify misconceptions and ensure that American discourse about these topics shows a more nuanced understanding.

I was in China conducting this research during the 2015-16 PLA reforms. At times, when I asked questions about certain issues, people would simply respond that they didn’t know — until the reform package clarified, there was a lot of uncertainty.

As a group, the Chinese defense and strategic studies community isn’t fundamentally different from what you’d expect in the United States. There’s diversity in terms of gender, age, and ideological perspective — some more conservative than others. However, it’s a relatively small community, similar to the United States, where many experts know each other well and are familiar with each other’s views.

Jordan Schneider: One more question about the defense community. In a recent paper, you noted that the civilian defense consensus argued against extensive nuclear modernization, but Xi disregarded this view, and modernization proceeded anyway. However, before 2019, expert discussions often aligned with eventual PLA actions across various dimensions. What do you make of this shift?

Fiona Cunningham: One conclusion might be that the expert community has less influence or interaction with decision-makers than previously. Some have drawn this interpretation regarding nuclear policy. I have work in progress — referenced when we discussed China’s nuclear modernization drivers — showing that perspectives within China vary on why nuclear capabilities should be modernized and what specific actions should be taken.

In that paper, I focused on one segment of the PRC community involved in arms control that expressed concerns about threats to China’s retaliatory capability but didn’t advocate for substantial arsenal expansion. However, other voices within China’s strategic and defense community do support a larger arsenal — not based on intricate calculations about warhead targeting or force posturing to make credible threats, but simply because more nuclear weapons provide political leverage. This isn’t connected to military campaigns or outcomes but shows the psychological impact of greater capability. It could even be viewed as status-related, with nuclear prominence offering instrumental advantages.

This mentality helps explain the gap between what the arms control community might recommend and China’s actual behavior. More research on this is forthcoming.

Jordan Schneider: Standing invitation — we could host shows in Chinese if experts want to join us, with Fiona and me co-hosting discussions about specialized aspects of PLA doctrine.

Mike Horowitz: I look forward to the translation of that episode.

Mary Clare McMahon is an incoming Schwarzman Scholar (‘26) and former Winter Fellow at the Centre for the Governance of AI, where she researched compute governance and U.S.-China AI competition. Previously, she worked in the National Security and Cybercrime Section of the U.S. Attorney’s Office for the Eastern District of New York.

Last month, reports emerged that DeepSeek was running a distilled version of its R1 reasoning model on Huawei’s Ascend chips. While DeepSeek trained its model on Nvidia Hopper series chips, Huawei’s deployment of Deepseek R1 underscores a broader strategic question: to what extent can Huawei erode Nvidia’s dominance in the global AI chip market?

Nvidia’s position rests on what has called a “three-headed hydra” of leading hardware, networking capabilities, and, most importantly for this piece, a deeply entrenched software ecosystem. At the center of that ecosystem is CUDA, a proprietary programming framework that allows developers to efficiently map computations onto Nvidia’s GPUs. CUDA’s value lies not only in its performance but in its reach: an expansive set of libraries, optimized workflows, and tight integration with widely-used machine learning frameworks make it the industry standard. And, crucially, CUDA can only be used with Nvidia GPUs. That makes CUDA a core component of Nvidia’s competitive advantage, otherwise known as Nvidia’s moat.

This article explains Huawei’s attempt to replicate and bypass that moat. For now, Huawei appears to be advancing the following three-pronged strategy:

1. Building out its own software stack, including a proprietary parallel programming model and surrounding tools that developers rely on to write, optimize, and deploy code efficiently.

2. Deepening integration with PyTorch, the most widely adopted open-source machine learning framework for model training.

3. Investing engineering resources in developing the Open Neural Network Exchange (ONNX), an open standard for machine learning models that enables portability across hardware platforms, to support the deployment of non-Ascend-trained models on Ascend chips.

Huawei is not the only actor seeking to erode Nvidia’s software lock-in — AMD has made similar efforts with ROCm, and Google has a software stack fitted to run Google TPUs. However, Huawei remains the most significant challenger in the Chinese market. The core question is not whether Nvidia’s dominance is being contested, but whether Huawei’s software strategy can mature enough for a full-stack transition away from U.S. hardware. This article proceeds in two parts: part one provides background on Nvidia’s software moat and how it was constructed; part two analyzes Huawei’s evolving response.

Nvidia’s Software Moat

The roots of Nvidia’s software moat can be traced back to the late 2000s, when CEO Jensen Huang made a long-term bet on CUDA, Nvidia’s proprietary parallel computing platform. In 2007, Nvidia released CUDA as a programming model for scientific computing. At the time, the dominant paradigm for scientific research (and most other computing applications) was CPU-based computation; GPUs were considered niche accelerators, primarily designed for graphics rendering in video games. CUDA’s launch was an explicit attempt to invert that paradigm by positioning the GPU as a general-purpose compute platform.

CUDA allowed developers to write code in familiar C/C++ syntax that executed directly on Nvidia GPUs, thereby accessing the highly optimized functionality of these GPUs. But creating a new computing model meant overcoming a classic chicken-and-egg problem: developers needed hardware to test their software on, and customers needed software to run on their hardware — neither would commit without the other. Nvidia addressed this by seeding the market for CUDA with its consumer gaming cards, which already had a broad base of installation. It made CUDA freely available (without open sourcing the code), created a global developer conference, and worked directly with scientists and researchers to port algorithms to the GPU. As Huang later recalled in a speech at National Taiwan University, “We worked with each developer to write their algorithms and achieved incredible speedups.” This engagement strategy eventually paid off; in 2012, AlexNet was trained on CUDA and Nvidia GPUs.

Subscribe now

As Nvidia’s software and hardware stacks became popular with deep learning researchers, Nvidia continued to invest in — and improve — CUDA. Nvidia created an extensive suite of libraries, such as cuDNN for deep learning, which dramatically lowered the time and expertise required to deploy high-performance models. In short, CUDA became more than just a programming model — it became the foundation of a full-stack software ecosystem.

For the next decade, CUDA continued to improve and attract more developers. And it is still improving to this day — though CUDA is closed source, Nvidia welcomes and often incorporates developers’ feedback. Nvidia also maintains online forums for developers to answer and ask questions about CUDA.

Thus, the CUDA ecosystem embeds substantial switching costs. Developers who migrate away from CUDA usually must rewrite large portions of code — by forgoing access to Nvidia’s finely tuned libraries, developers are forced to substitute with less mature equivalents, if any replacements exist at all. Further, developers also lose support from the large troubleshooting community that has grown up around CUDA.

Today, many machine learning developers do not code directly in CUDA. Instead, they write code in Python, a higher-level and more user-friendly language, using frameworks such as PyTorch and JAX. But even here, CUDA remains central: it acts as the backend bridge between PyTorch and Nvidia’s GPU architecture.

We will discuss PyTorch in greater detail in a later section. For now, it is enough to note that CUDA’s value lies not only in its impressive performance (which has improved continuously for nearly two decades), but also in the ecosystem that has formed around it. That is the essence of Nvidia’s moat — challengers with competitive hardware must also replicate an entire software environment if they want to compete.

Huawei’s Software Strategy

Undermining Nvidia’s software moat requires more than performance parity with Nvidia GPUs — it demands a credible alternative to the tightly integrated CUDA ecosystem. Huawei appears to be pursuing such an alternative. Its strategy consists of three interrelated prongs, each aimed at reducing the friction of switching away from Nvidia.

First, it is expanding its native software stack alongside a growing suite of tools designed to mirror the utility of CUDA’s broader ecosystem. Second, Huawei is deepening integration with PyTorch, the most widely adopted machine learning framework and one that, by default, pairs seamlessly with CUDA. By building backend support through adapters like torch_npu, Huawei is attempting to position Ascend as a drop-in hardware alternative. Third, Huawei is investing in ONNX (Open Neural Network Exchange), an open standard for cross-platform model representation, to allow models trained on non-Huawei hardware to run inference efficiently on Huawei chips. Together, these efforts seek to replicate the full-stack developer experience that has made CUDA so difficult to displace.

1. Huawei’s Software Alternatives

Huawei’s most direct challenge to CUDA comes in the form of CANN (Compute Architecture for Neural Networks), its proprietary programming environment for Ascend NPUs. CANN sits at the same level of the software stack as CUDA, providing the tools needed to execute high-performance machine learning models on Huawei hardware. Paired with CANN is MindSpore, Huawei’s high-level deep learning framework, conceptually analogous to PyTorch. Together, these tools form Huawei’s native alternative to the Nvidia-centric PyTorch + CUDA stack.

​​CANN has been in development since at least 2019, the year Huawei was added to the US entity list. Huawei’s 2024 Annual Report highlighted (on four occasions) the release of CANN 8.0 in September of 2024, promoting this development as a significant step in advancing AI computing capabilities.

However, developers cite serious usability issues with CANN. According to the Financial Times, one Huawei researcher complained that CANN made the Ascend chips “difficult and unstable to use.” One developer described the process of using the Ascend 910B as “a road full of pitfalls” (踩坑之路), sharing the following reflections on Zhihu, a Quora-like Chinese website for academic discussion, in February 2025:

“I have been interning in the company for the past six months. Due to the shortage of computing resources, interns can only use Ascend 910B for training and development… Looking back, every time I encountered various problems and bugs, it was difficult to find the corresponding solutions on the Internet. Some problems were finally solved with the help of Huawei's operation and maintenance engineers. Therefore, I hope that this article, in addition to summarizing my own staged engineering experience, can help more Ascend NPU developers and help the development and progress of the domestic computing ecosystem.”

426 other users upvoted the post. One commenter responded, “It seems that it will take until 2027 for CANN to be truly mature, stable, and easy to use.”

The absence of a robust developer community for CANN further increases the onboarding burden for new developers. Unlike Nvidia’s developer forums, which benefit from community-maintained documentation and rapid peer troubleshooting, Huawei’s Ascend developer portals — both in English and Chinese — exhibit low engagement, with sporadic posts and limited public debugging activity. According to another Zhihu article posted in June of 2024, “When I first started exploring Ascend, I felt quite overwhelmed. Although there is a lot of documentation available, it feels quite disorganized. When encountering problems, the limited user community means you probably won’t find a corresponding solution, which leads to frequent frustration.”

While the Nvidia CUDA Programming and Performance Developer page had multiple live threads posted just days before the screen capture above, the most recent posts on the Huawei CANN developer pages were from January 2025.

Adapting models to run on Huawei’s platform is also onerous. According to that same Zhihu article from June 2024, “Any public model must undergo deep optimization by Huawei before it can run on Huawei's platform. This optimization process is heavily dependent on Huawei and progresses slowly.” By contrast, after testing the Nvidia H100 and H200 for model training applications, Semianalysis reported, “Nvidia’s Out of the Box Performance & Experience is amazing, and we did not run into any Nvidia specific bugs during our benchmarks. Nvidia tasked a single engineer to us for technical support, but we didn’t run into any Nvidia software bugs as such we didn’t need much support.”

To try to increase adoption, Huawei has adopted a strategy reminiscent of Nvidia’s own CUDA rollout in the 2000s: embedding engineers directly into customer sites to assist with code migration. According to reporting from the Financial Times, Huawei has deployed engineering teams to Baidu, iFlytek, and Tencent to help reimplement and optimize existing CUDA-based training code within the CANN environment​. This mirrors the anecdote recounted above, where Jensen Huang described how Nvidia “worked with each developer to write their algorithms and achieved incredible speedups” during CUDA’s early years. Huawei is now attempting to replicate that strategy, pairing onboarding with high-touch technical support in the hope of accelerating ecosystem uptake.

In parallel, Huawei is also trying to improve its native software stack. DeepSeek engineers have reportedly said that the Ascend 910C can achieve up to 60% of the inference performance of the H100, and potentially more with CANN optimizations. As Kevin Xu noted on a prior episode of ChinaTalk, DeepSeek engineers have proven adept at “work[ing] below CUDA to maximize their Nvidia GPU.” If similar techniques were applied within the Huawei ecosystem, they could help close the performance gap between Ascend and NVIDIA hardware.

One particularly intriguing way to close that gap involves using AI to accelerate software optimization. If AI systems themselves can be leveraged to improve kernel optimization, develop the CANN and MindSpore stack, and reduce performance inefficiencies, it could meaningfully shift the competitive landscape. Sakana AI has already demonstrated a version of this approach with its “AI CUDA Engineer,” an agentic framework that translates standard PyTorch code into highly optimized CUDA kernels. According to Sakana, the system achieves 10—100x speedups for AI model training. If comparable AI-driven optimization techniques could be adapted for Huawei software, it would represent a significant step toward enhancing performance within the CANN ecosystem. Developer loyalty might follow.

Despite its investment in a native software stack, though, Huawei appears to recognize that displacing CUDA with CANN is not feasible in the near term. As a result, it has shifted part of its strategy toward interoperability rather than replacement. Nowhere is this more evident than in Huawei’s growing involvement with the PyTorch ecosystem.

2. Huawei and PyTorch

As part of its strategy to reduce friction in migrating away from Nvidia, Huawei has prioritized compatibility with PyTorch, the dominant open-source machine learning framework used across academia and industry. Originally developed by Meta’s AI research lab in 2016, PyTorch was released publicly in 2017, then transitioned to being governed by a wider network of companies under the Linux Foundation in 2022. The resulting PyTorch Foundation is governed by a consortium of premier members, including Meta, Microsoft, Google, Amazon, AMD, Intel, Nvidia, and, as of October 2023, Huawei.

PyTorch enables developers to define, train, and deploy machine learning models using concise and intuitive Python code. The framework's popularity stems from its "eager execution" model, which allows each operation to run immediately, making it easier to debug, prototype, and iterate than other alternative frameworks (like Google’s TensorFlow).

From the outset, PyTorch was optimized for Nvidia GPUs. New operators and features are still tested and tuned against CUDA first, and performance benchmarks are routinely conducted on Nvidia’s hardware. Installing PyTorch via Python’s package manager automatically sets it up to run on Nvidia GPUs. This makes the framework effectively Nvidia-native, and any effort to use it on non-Nvidia hardware requires not just backend substitution, but complete ecosystem engineering.

The challenge for Huawei, then, is not only to make PyTorch run on Ascend hardware, but also to make it run well enough that developers don’t notice they’ve switched ecosystems.

Huawei’s primary technical achievement has been enabling the execution of PyTorch models on its Ascend NPUs through an adapter called torch_npu. Torch_npu bridges PyTorch with Huawei’s low-level NPU drivers and CANN backend. Huawei developers publicized this development at the 2024 PyTorch Shanghai Meetup, pictured below.

Huawei’s torch_npu adapter allows Huawei's AI accelerators to interface with PyTorch, though it exists separately from PyTorch’s main codebase. (The torch_npu adapter uses PyTorch’s PrivateUse1 mechanism, an interface that lets hardware makers test new accelerators without immediately merging their code into PyTorch.) At the 2024 PyTorch meetup in Shanghai, a Huawei engineer noted that devices maintained outside PyTorch’s core, like Huawei’s, often face stability issues because changes in PyTorch's main code aren't automatically tested for compatibility. This challenge is widely recognized by the community.

For this reason, Huawei’s forked version of PyTorch is still less effective than Nvidia’s CUDA-native implementation, and developer feedback points to persistent challenges in runtime reliability and documentation. In a Zhihu thread with more than 700,000 views, senior software engineer “Mingfei” wrote that, “It’s worth emphasizing that plugins [referring to the forked version of PyTorch] are not native” and “several unavoidable issues arise,” including version compatibility; third-party extension support; and test coverage challenges. Another Zhihu contributor noted, “Ascend chips provide poor support for third-party frameworks like PyTorch and TensorFlow, making it extremely challenging to adapt to the latest large-scale models and use them effectively.” Note that the developer seems to be referring to the challenges of deploying models on Ascend chips, not training new models.

While Huawei’s patches have not yet been fully integrated upstream, there are reasons to believe that Huawei might be able to garner political support within the PyTorch Foundation to formalize its contributions. The PyTorch Foundation’s official announcement of Huawei’s status as a premier member noted that Huawei “provides easier access to the PyTorch ecosystem for more hardware vendors… [which] aligns with the PyTorch Foundation’s mission to develop AI as part of a sustainable open source ecosystem and produce inclusive technological feats.” This quote seems to suggest that PyTorch wants to support other hardware options besides Nvidia’s. Further, Huawei’s status as a premier member of the PyTorch Foundation grants it a seat on the Governing Board, as well as a formal role in setting foundation-wide policies and technical priorities. This membership was unanimously approved by existing premier members, signaling at least tacit acceptance of Huawei’s contributions by Meta, Nvidia, AMD, and Google. Finally, Huawei appears to be strongly committed to contributing to open source projects. The company’s 2024 Annual Report highlighted that Huawei is “a firm supporter and major contributor to open source communities” and explicitly mentioned its membership in the PyTorch Foundation.

In sum, Huawei is executing a long-term strategy to allow developers to use PyTorch with its Ascend series of chips. Its success will depend on the company’s continued technical contributions, the size of its developer community, and whether the PyTorch Foundation will incorporate the torch_npu and other Huawei contributions into its main code base.

3. Huawei and OXXN

While Huawei’s PyTorch integration aims to reduce friction in model development, it does little to solve the harder problem of model portability — that is, how to take a model trained on Nvidia hardware and deploy it on Huawei’s Ascend chips. To address this, Huawei has turned to a complementary approach, optimizing the Open Neural Network Exchange (ONNX) format to serve as a bridge between software ecosystems.

ONNX (Open Neural Network Exchange) is an open-source format originally developed by Meta and Microsoft in 2017 to enable model interoperability across deep learning frameworks. It allows developers to export a model trained in one framework, such as PyTorch with CUDA, and run inference in another runtime environment — or on different hardware entirely. It also helps optimize models, allowing them to run faster than they would if they were directly deployed from PyTorch. ONNX operates under the umbrella of the Linux Foundation AI & Data, of which Huawei is a premier member.

Put simply, ONNX is like the PDF of AI models. Just as documents created in Microsoft Word or Google Docs to be exported into a portable, fixed-format PDF file that can be opened and viewed across operating systems, ONNX allows models trained in PyTorch or other machine learning libraries to be exported into a standardized format that can then be run on different hardware platforms.

Huawei has embraced ONNX Runtime, the engine that executes ONNX models. The company maintains a public Ascend ONNX Runtime, available on GitHub, which includes optimized kernels and execution instructions tailored to CANN and Ascend chips. According to the ONNX Runtime documentation, Huawei’s ONNX Runtime page is “community-maintained,” meaning that it is maintained by Huawei rather than by the core ONNX Runtime team, and that it is Huawei’s responsibility to ensure ongoing support for the library.

Subscribe now

Huawei’s goal here is straightforward: to enable developers to train models on non-Huawei hardware, export the files to ONNX, and deploy the models on Ascend chips, all without rewriting core logic. This workflow has clear appeal in the Chinese market. Model developers could still train on Nvidia Hopper chips or train models through the cloud, then shift deployment or inference workloads to Huawei hardware.

It’s important to note that running a model on hardware, even if using an ONNX file, can introduce bugs or compatibility issues. Some PyTorch operations don’t export cleanly to ONNX, while others need rewriting. ONNX models may also need custom operations that the hardware backend has to support. That said, Huawei’s investment in ONNX offers a practical path to inference decoupling. In contrast to the CUDA-first development loop, which binds training and deployment to Nvidia hardware, ONNX gives Huawei a way to insert itself at the deployment stage, even if training remains CUDA-bound.

Huawei’s Future

Nvidia’s enduring dominance in the AI chip market is not due to superior hardware or networking architecture alone — it’s also a function of Nvidia’s deeply integrated software ecosystem. This ecosystem — anchored by CUDA, high-performance libraries, and seamless compatibility with PyTorch — offers a robust developer experience and an active community that reinforce Nvidia’s lead. Huawei’s strategy is to build a competitive stack of its own.

Model deployment may be Huawei’s most immediate opening. Already, it has demonstrated that models trained on Nvidia hardware, like DeepSeek’s R1, can be run in distilled form on Ascend chips. If the US were to ban the export of Nvidia H20s to China, this workaround could become standard. In that scenario, indicators of improvement in the Huawei software stack would manifest not as headlines, but as reduced developer complaints, more seamless deployments, and fewer distinctions between fallback option and first choice.

Huawei isn’t there yet, though. As noted by the exasperated programmers quoted above, working with Ascend 910B chips still requires debugging without community support. But Zhihu threads where developers vent frustrations can eventually become a troubleshooting resource that contributes back to the Huawei ecosystem. With enough developers dedicated to advancing that new ecosystem, the result could be a slow, durable shift away from CUDA. That shift won’t happen overnight — remember, it took Nvidia 18 years to build the CUDA ecosystem of today; building a competitive software ecosystem is a multi-year effort even under pressure. But what started as necessity may, over time, harden into habit — and eventually, into infrastructure that can compete with Nvidia’s software stack.

Special thanks to Jeff Ding and Kevin Xu for thoughtful feedback on prior drafts.

What’s happening to America’s science and technology ecosystem? How is China interpreting the current state of US research, and how is it working to build its own science and technology base in response? And what can we learn from China's war mobilization exercises?

To explore these questions, we're joined by Divyansh Kaushik and Alex Rubin, who both work at Beacon Global Strategies. Divyansh holds an AI PhD from Carnegie Mellon, and Alex spent the past decade at CIA focusing on China and emerging technologies.

We discuss…

• The historical origins of the US R&D model, and the division of labor between universities, government, and industry,

• How budget cuts will impact the NSF, NIH, NIST, and DoD basic research,

• Why and how China attempts to emulate US research institutions,

• What a leaked wargame exercise from Guangdong province can tell us about China’s grand strategy,

• How institutions like ChinaTalk complement the IC with fresh, independent research.

Listen now on iTunes, Spotify, or your favorite podcast app.

How America Has Won With R&D

Jordan Schneider: Let’s do a 101 on the broader American research ecosystem. What does the interaction between universities, government funding, and corporations look like in the 2000s?

Divyansh Kaushik: To better understand today’s landscape, we need to trace our steps back about 70 years and examine how the American research ecosystem was conceptualized. The original model positioned universities to conduct curiosity-driven research funded by the federal government, while American industry focused on transforming that research into applications.

There were certain industrial monopolies created by the government that also conducted basic research, which Alex can address more comprehensively. However, the overwhelming majority of basic research happened in academia — universities created as land-grant institutions or those existing before the war. This system served us remarkably well, as basic research developments from the 1950s, 60s, and 70s bore fruit 10, 20, 30, or 40 years later. The nature of basic research doesn’t necessarily have an immediate application, but applications may emerge years down the line.

Consider this example: During the 1970s and 80s AI winter, when nobody was funding neural networks research because it was viewed as a dead end without viable applications, the National Science Foundation — created to fund basic research through the federal government — was funding Geoffrey Hinton’s work on neural networks, being the only entity supporting this research at the time. Fast forward 40-50 years, that work has fundamentally shaped how we view AI today. It’s the foundational technology behind all the large language models currently in use.

NSF also funded Andrew Barto’s entire PhD. Barto, together with Richard Sutton, established the field of reinforcement learning at a time when there were few practical applications. Today, reinforcement learning is a critical component behind LLMs, AlphaFold, and similar technologies.

This exemplifies how America has pursued basic research. Currently, there’s considerable criticism about research projects like "shrimp on a treadmill" or "fish on cocaine," questioning why such studies receive funding. While these projects have legitimate scientific purposes, to the general public they appear to be wasteful uses of federal research dollars.

Agencies like the National Institutes of Health fund more applied research on medicines and can point to tangible outcomes — specific drugs developed with NIH funding. The NSF, conversely, funds basic research that may not demonstrate tangible benefits for decades, as happened with neural networks.

We’re now engaged in a deeper conversation about what constitutes waste or abuse of federal research dollars and how to allocate those funds more effectively. Is industry-funded research the optimal approach? Does the federal government have — or should it have — a role in the R&D ecosystem? What about public-private partnerships, which were a cornerstone of the CHIPS and Science Act in creating the Technology Innovations and Partnerships Directorate at the National Science Foundation?

This significant conversation emerged particularly this year, as the National Science Foundation, Department of Energy, Department of Defense, and National Institutes of Health find themselves at the center of questions regarding the appropriate allocation of federal dollars toward research.

Jordan Schneider: Let’s explore more of this history, because I think we can’t ignore the broader defense community’s role in funding R&D over the past 75 years. Alex, would you like to address that?

Alex Rubin: This is best illustrated through an interesting case study. We’re currently focused on the semiconductor industry due to its substantial economic and strategic implications. The foundation of the semiconductor industry can be traced directly back to funding from the Air Force and NASA for both the Apollo program and ICBMs.

One interesting example demonstrating the federal government’s role involves what we call the "Valley of Death" — the challenge of bringing novel research from the laboratory to market. The federal government, particularly the defense procurement establishment, has excelled at intervening at this crucial stage in the R&D cycle by providing customers and markets for these technologies. This allows companies to expand production, build scale, and reduce costs, making it feasible for them to enter commercial markets.

The modern semiconductor industry wouldn’t exist as it does today without those initial purchases from the Air Force, NASA, and other government entities. It’s extraordinarily difficult to transition from the lab to the market, especially when costs are high. Finding consumers willing to purchase these products when they’re expensive is challenging, which is where government plays a particularly important role in advancing that cycle.

Subscribe now

When discussing the federal government’s role in early-stage basic research, there are instances where industry has undertaken this responsibility. The AT&T system — the Bell system — during the 20th century provides a classic example of industry conducting exactly the type of research we’re discussing. However, certain unique characteristics made this possible for the Bell system.

Throughout the 20th century, the Bell system comprised several key units: corporate headquarters in New York, its manufacturing arm (Western Electric) producing equipment for telecom networks, the regional Bell operating companies that eventually became companies like T-Mobile and Verizon providing local phone service, long-distance service, and crucially, Bell Labs.

Extensive literature documents Bell Labs’ history and impact. Innovations including the transistor, discovery of cosmic microwave background radiation, and cell networks all emerged from Bell Labs. The list of remarkable technologies and innovations originating there is extensive.

Bell Labs could invest in both basic and applied research because of their consistent funding stream from what was essentially a government-regulated monopoly. In the 1910s, the government granted AT&T monopoly status over the telecom industry with certain conditions, including a requirement to license any inventions outside the telecom industry under generous terms or without fees.

Bell Labs operated with a consistent revenue stream from their telecom service monopoly, maintained a direct relationship with manufacturers allowing them to troubleshoot as products moved to manufacturing, and had direct connections to customers through their operating companies, enabling them to identify market demands.

Most importantly, they had long-term consistent funding — precisely what makes federal research dollars so crucial for basic research. This consistent funding allows investment in projects that might not yield deliverables for 10, 15, or 20 years. This differs dramatically from typical corporate investments seeking returns within a couple of years.

Companies justifiably need to demonstrate return on investment, which becomes incredibly difficult without consistent market support. Industry can indeed support basic research, but it requires specific enabling characteristics similar to those that enable basic research funding from the public sector.

Jordan Schneider: Divyansh, could you address the university’s role in this ecosystem?

Divyansh Kaushik: Universities today aren’t limited to basic research — they conduct significant applied and industry-funded research as well. However, an often overlooked aspect of universities’ contribution is their role in creating talent pipelines. The researchers going into industry are those who received federal funding at universities.

These individuals enter graduate programs where they develop intellectual curiosity through curiosity-driven research and by solving interesting problems without immediate pressure to generate revenue. They cultivate this intellectual curiosity and bring it with them when they join industry, ultimately driving the industrial innovation we witness.

Universities play a crucial role in regional innovation and economic growth. The spillover effects include startups and jobs created as a result of research funding. Numerous economic studies demonstrate multiple dollars returned for every dollar of federal R&D spending at universities.

Universities also advance national security objectives. Carnegie Mellon University, my alma mater, works with the Department of Defense on several projects directly impacting warfighters. Universities host Federally Funded Research and Development Centers (FFRDCs) and integrate DoD personnel into their research programs.

Universities therefore have a broader ecosystem-driving role, not just a narrow focus on quarterly profits. Both approaches are valid — they’re complementary rather than substitutive. While most companies don’t conduct basic research, some do. Microsoft, for instance, spent nearly ten years developing the Majorana chip for quantum computing, made possible by consistent funding.

If we reduce consistent funding for universities, we’ll see fewer PhD students enrolling, fewer PhDs granted, fewer qualified individuals joining companies like Microsoft, and ultimately fewer innovations like the Majorana chip. This affects the entire ecosystem.

Jordan Schneider: Alex, can you discuss how envious the rest of the world is of the ecosystem America has built?

Alex Rubin: A couple of decades ago, China looked at the U.S. R&D ecosystem and essentially said, “We want that,” and began working to replicate it. They’ve invested considerable resources — money, time, and high-level attention. President Xi Jinping regularly emphasizes the importance of basic research, improving China’s STEM education system, and developing talent as key enablers of China’s technological development and growth as an economic and global power.

China recognizes that the U.S. model is incredibly effective at generating innovations, bringing them to market, and establishing dominance and first-mover advantages in critical new technologies. China’s approach to its R&D ecosystem and education system focuses on three main categories, emphasizing generational investment.

Jordan Schneider: Let me provide some context with numbers. The U.S. spends approximately $50 billion annually on basic R&D, with another $50 billion coming from universities and businesses — totaling around $100 billion yearly.

American firms represent 80% of the world’s global technology market capitalization. Additionally, 80% of science Nobel Prizes over the past 50 years have included winners with U.S. affiliations. These three factors are interconnected.

This ecosystem produces the most advanced companies, which then provide cutting-edge technologies to the national security establishment. It’s a beautiful, self-reinforcing system. The best scientists work at American universities, attracting the best students worldwide. Despite providing only 25% of global basic funding, the U.S. spends it so effectively that the greatest minds globally want to come here and work on these topics.

Alex Rubin: This is indeed a generational investment. It’s no coincidence that U.S. companies initially led in the semiconductor industry, then in personal computers and other computing applications, and now in AI. These advantages build upon each other.

If you establish early leadership in one industry and continue making long-term investments, it naturally positions you advantageously for the next generation of technology. Conversely, if we make decisions now that underinvest in research or otherwise hinder the development of these talent ecosystems, the real impact might not become apparent for 10-15 years.

Unfortunately, once those impacts become visible, it’s often too late. Recovering lost ground requires significant time. The research investments and decisions we make now — whether regarding grants or graduate programs — will show their consequences a decade or more from now.

Jordan Schneider: That concludes our cheerleading session for the American R&D ecosystem. Divyansh, what has been happening over the past 100 days that concerns all three of us — developments that may risk the world-historical R&D golden goose America has built since Vannevar Bush wrote to FDR, envisioning a glorious future made possible by the collaborative efforts of corporations, research universities, and the U.S. government?

Divyansh Kaushik: I would actually broaden the timeframe to consider the last year or so. During this period, we saw the NSF budget cut by approximately 8% from the previous year, the NIST budget cut by about 13%, and the DoD basic research budget reduced by roughly 4%.

Early in the administration, NIH changed its policy on Facilities and Administrative (F&A) benefits, unilaterally reducing them to 15% — a decision that faced legal challenges. This was followed by layoffs at several federal funding agencies.

More recently, DoD and DOE followed NIH’s approach by capping F&A at 15%, which merits separate discussion. Additionally, NSF terminated approximately 400 previously awarded grants. We also witnessed the resignation of the NSF director amid rumors of potential additional layoffs at the agency.

Jordan Schneider: A former Trump appointee, mind you.

Divyansh Kaushik: Correct, and unanimously confirmed by the Senate. Further concerning developments included leaked information about the President’s budget request, suggesting NSF could face approximately 55% budget cuts, with NIH potentially facing similar reductions.

These issues are foremost in the minds of both academic and industry researchers. Brad Smith recently wrote a blog post about quantum computing where he emphasized the importance of basic science funding for workforce development. This concern is widespread.

Several former national security leaders, including former Trump appointees such as his former Homeland Security advisor and others, signed a letter to Congress highlighting the importance of funding basic science research at this critical juncture. As Alex mentioned, China has increased its basic R&D spending by 10% year-over-year for the past seven years.

The CHIPS and Science Act established a vision to double our federal basic R&D spending over the next decade. Instead, we’ve failed to meet this moment. Approximately $50 billion of authorized funding from the CHIPS and Science Act remains unappropriated for the science component.

Regarding talent, certain universities received letters terminating visas for some PhD students — a decision the administration later reversed. This situation weighs on the minds of universities and researchers, raising questions about broader implications on the global stage.

France, Australia, and China have attempted to capitalize on this uncertainty by establishing specific programs to attract U.S. researchers, offering long-term stability, funding, and residency benefits.

Jordan Schneider: You missed one aspect, Divyansh. We’re also seeing targeted actions against specific universities, with significant conflicts involving Columbia and Harvard. Beyond the 400 NSF grants canceled due to DEI considerations, research is being canceled simply because researchers happen to be PhD students or professors at Harvard.

Divyansh Kaushik: That certainly occurred. Interestingly, as we record this, President Trump just announced he’s naming Secretary of State Marco Rubio as interim National Security Advisor and nominating NSA Mike Waltz for UN Ambassador. Developments are unfolding rapidly.

Jordan Schneider: Wow, really? That’s the best possible outcome. I was preparing for the Laura Loomer National Security Advisor era.

Divyansh Kaushik: I mention this because the National Security Council plays a crucial role in this conversation by emphasizing the national security importance of federal R&D. Alex understands this well from his previous position, particularly regarding the critical benefits it provides.

Jordan Schneider: Wait, we need to focus on this for a moment. We haven’t seen this dual role since Kissinger. Is that right, Alex? Has there ever been another person serving in both capacities simultaneously?

Alex Rubin: No, I believe Kissinger was the only one.

Jordan Schneider: This is remarkable. There’s been considerable discussion about NSC reform, as it’s not a fixed organization. I wonder if this presents an opportunity for such reforms, though Rubio’s State Department reforms appear less developed than anticipated or discussed.

I recognize we’re speculating beyond our expertise, but this breaking news deserves attention. There’s a certain Nixon-era quality to these developments. From my perspective, this appointment represents a positive direction — the situation could have been significantly worse than Marco Rubio.

Alex Rubin: Regarding technology policy, Rubio has been at the forefront on issues concerning investments in technology and its centrality to competition with China. During his Senate tenure, his team produced a report examining Made in China 2025. Technology represents the key battleground in this competition. Extending that metaphor, researchers, scientists, and engineers serve as the frontline contributors to American power in this space.

Jordan Schneider: This is astonishing. Earlier today, New York Times articles suggested Waltz had been dismissed — likely someone attempting to shape that narrative. Rubio presents an interesting case, given the contrast between his decade-plus Senate career and his more recent MAGA-aligned positioning. JD Vance seems to embody that perspective more naturally than Rubio. We’ll have to observe how this develops.

Returning to science and technology — before discussing positive aspects, Divyansh, we should address the challenges facing the university ecosystem regarding talent and funding. International students constitute a crucial funding component since most pay full tuition. Government funding represents another vital revenue stream.

Only about 10-15 universities possess multibillion-dollar endowments that would enable them to withstand major external shocks such as losing international students or significant funding cuts without drastic measures like closure or acquisition by private equity firms.

You briefly mentioned immigration concerns, but the situation created genuine alarm among many students who feared leaving the country. Although courts have temporarily reversed certain policies, I worry these uncertainties will linger in the minds of parents worldwide and PhD students considering where to establish their careers.

Divyansh Kaushik: We’ll soon see how this affects enrollment as universities release their yield data. With May 1st approaching and April 15th being the deadline for students to accept or decline offers, that information will become available shortly.

We’ve already observed an 11% decrease in international student enrollment between March 2024 and March 2025, mirroring a similar trend between March 2016 and March 2017. We must monitor this data closely.

Regarding the importance of this population, people often overlook that international students comprise 60% of Computer Science and AI PhDs, and approximately 50% of all STEM PhDs and Masters students. Replacing this talent with domestic students would require considerable time and concerted effort.

Notably, the number of domestic computer science undergraduates pursuing graduate degrees has remained unchanged since approximately 1990-1995. China, with four times the U.S. population, produces twice as many STEM PhDs, twice as many STEM Masters, and four times as many STEM Bachelors graduates.

We can no longer credibly claim that their STEM education or research quality is inferior — they excel in both quality and quantity. Based purely on numbers, our only viable competitive strategy involves recruiting talent globally. China has substantially more human resources to dedicate to complex problems than we do, a critical factor in this discussion.

Jordan Schneider: Let’s return to your "On the plus side" perspective, Divyansh.

Divyansh Kaushik: Consider Michael Kratsios’ remarks at the Endless Frontier retreat approximately 16-17 days ago, on April 14. He described an emerging golden age for America, speaking of "the early light of this new golden age," "American hope," and "the possibility of progress through science and technology."

He emphasized that this golden age will materialize only if we actively choose it, then outlined his approach. He discussed how ours was the atomic age and how we must fight to restore that inheritance. He proposed rethinking federal R&D spending through smarter methodologies.

The Biden administration implemented numerous pilot programs in this direction, but making those approaches the primary R&D strategy would represent a significant achievement for the current administration. New experimentation and prize competitions would be particularly beneficial. We must consider how to optimize every dollar spent on R&D.

Grazio emphasized that beyond a protective agenda to maintain American dominance, we need a promotional agenda. We must create a funding environment that clearly articulates our national priorities, enables scientists to develop new theories, and empowers engineers to implement them. Using advanced market commitments would multiply the impact of government-funded research.

His address contained numerous positive elements that create opportunities for the administration to scale these efforts. Now is the ideal time for those with bold ideas to advance them.

Jordan Schneider: We observed the DOGE approach during the first hundred days — not implementing reforms to unlock a better version of government, but simply making cuts. As the DOGE era concludes, we recognize you can’t forcibly impose creative meta-science reforms, though these organizations do need restructuring.

The current energy, insight, and understanding acknowledge that conditions aren’t ideal. Breakthroughs have become less frequent and more expensive relative to expenditures compared to the 1950s-70s. This presents an opportune moment to experiment with new approaches. However, these efforts become significantly more challenging with half the funding and without international talent — risks created by the budgetary constraints, visa restrictions, and confrontational stance toward universities we’ve witnessed in recent months.

Alex Rubin: Yes.

Divyansh Kaushik: My friend Caleb Watney offers a valuable perspective: viewing federal R&D through a venture capital lens, given the substantial VC presence in government. We should measure performance by return on investment rather than by minimizing expenditure. The critical question is how to maximize outcomes from our investments.

Regarding reforming and restructuring agencies, these institutions are generally receptive to change. The National Science Foundation created the Technology, Innovation, and Partnerships (TIP) Directorate before Congress even passed the CHIPS and Science Act, despite some quiet resistance from other directorates. The agencies welcome innovation.

Consider how Department of Energy national laboratories are experimenting with OpenAI’s models as scientific peers for brainstorming. These represent fascinating initiatives by research agencies to reinvent their approaches to research and funding. If the administration pursues this direction, they’ll likely find substantial support from within the agencies themselves, as well as from universities and industry.

Jordan Schneider: We should campaign for Irwin as NSF Director!

To conclude our discussion on America’s research ecosystem, my assessment is that the Vannevar Bush “Endless Frontier” model has, over the past 75 years, delivered some of humanity’s greatest benefits. Setting aside national power considerations — which should be self-evident given that this system helped overcome the Soviet Union and created history’s wealthiest nation — this ecosystem could benefit from reforms. However, it represents the quintessential golden goose that we’ve managed to develop through work, consistency, and some fortunate circumstances.

This represents a national treasure, and what disturbs me most is the risk of crossing thresholds we cannot reverse. Ecosystems like this, when supported, demonstrate remarkable resilience. However, they contain inherent vulnerabilities related to institutions, funding streams, and talent that require continuous replenishment to maintain previous levels of success.

We’ve covered this extensively over the past eight years, following the excitement surrounding and ultimate passage of the CHIPS and Science Act — a period when bipartisan consensus seemed to favor increased investment in basic research. Watching immigration restrictions, culture war issues, and DOGE priorities converge to create perhaps the greatest threat this ecosystem has faced in decades is deeply concerning and something we’ll monitor closely in the coming months and years.

Divyansh Kaushik: American R&D is globally envied, and we should intensify our commitment to it. Universities serve as powerhouses in this system. Simultaneously, they aren’t blameless in many respects and need to engage in introspection regarding why our commitment to academia and universities faces questioning today. I hope many institutions will undertake this self-reflection and emerge stronger.

Alex Rubin: I’d like to add that while we’ve focused extensively on laboratories and academia and higher education, the R&D ecosystem extends beyond these components. It encompasses community colleges, vocational schools, and technical training programs that produce technicians who operate equipment in these laboratories — an absolutely crucial function.

Many major technology companies, particularly those with significant manufacturing operations, primarily employ community college graduates or individuals with technical training rather than PhDs. The semiconductor industry, for instance, has a substantial veteran population, recruiting former mechanics with relevant skills to maintain equipment. These aren’t PhDs, but they possess essential skills for equipment maintenance.

The final component, which speaks to generational investment, is K-12 education. Truly enhancing the quantity and quality of graduates from PhD and master’s programs begins at these early stages. We’ll discuss China’s approach in this area later, but the foundation lies in K-12 education, gradually building technical literacy so that by the undergraduate or graduate level, students’ mathematics and science skills match global standards.

Divyansh Kaushik: The administration recognizes this priority, evidenced by the recent AI in K-12 executive order, which aims to integrate AI education throughout K-12 curriculum to develop an AI-ready workforce in the coming years. Alex’s observation is entirely accurate, which further supports my optimism regarding future opportunities.

Monitoring Chinese Innovation

Jordan Schneider: Let’s discuss China. Alex, when fundraising for the ChinaTalk Institute, which has enabled me to hire exceptional talent tracking China’s developments in AI and biotech, several funders questioned the necessity of such an organization. They assumed the U.S. government adequately monitors China’s commercial technology through open-source intelligence. As someone who has spent the past decade primarily following Chinese science and technology in the commercial sector, how would you respond to that assumption?

Alex Rubin: My response is that it’s fundamentally a team sport. Different organizations — the intelligence community, other government agencies — have comparative advantages in what they monitor. However, when discussing commercial technology and areas where the primary actors aren’t governments but companies, universities, and laboratories, many strategically significant developments emerge from industry rumors and corporate insights.

Effectively monitoring these developments can’t be limited to individuals like myself in my previous role, working in secured environments to examine these issues. It requires a comprehensive approach that incorporates companies and universities.

The space for organizations like ChinaTalk involves engaging the general public. During the Cold War, nobody questioned why developments within the Soviet Union mattered — there was an inherent understanding of their connection to the American economy, jobs, and security. We need to establish similar connections today, explaining why developments like Huawei creating an advanced GPU matter to average Americans.

This is precisely where podcasts like ChinaTalk and similar outlets contribute value — bringing perspectives well-understood in Washington and disseminating them throughout the country.

Jordan Schneider: It’s interesting how you frame this through tactical, operational, and strategic perspectives when analyzing these questions. The flexibility available in think tanks, academia, or whatever category ChinaTalk occupies allows for different approaches.

Alex, what are the Chinese government’s long-term strategic intentions regarding science and technology?

Alex Rubin: I’ve settled on what I believe is the most accurate characterization, paraphrasing Matt Damon in “The Martian” — Xi Jinping plans to “science the hell out of China.” That’s his fundamental approach — an all-in bet on science and technology.

Whether examining the economy, military, or internal stability, technology permeates everything. Looking at the economy, Xi’s new catchphrase is "new quality productive forces" — a reinterpretation of classic Marxist-Leninist productive forces theory that essentially asks how technology can improve economic performance.

Key components include upgrading traditional manufacturing through robotics and AI automation. Another focus involves eliminating technological choke points by making China more self-sufficient through innovation and R&D investments, enabling Chinese companies to develop domestic alternatives to technologies they currently source from foreign providers. A classic example is photolithography, where significant investment is directed toward Chinese companies like SMEE to reduce dependence on lithography systems from the Netherlands.

The third component focuses on future industries. Last year, China identified six broad categories and numerous specific technologies for targeted support in their Future Industries Development Action Plan. Some might seem far-fetched, including humanoid robots, quantum technologies, artificial general intelligence, and brain-computer interfaces.

They’re absolutely serious about leveraging these technologies for economic benefits. China recognizes that the United States, through its R&D ecosystem, positioned itself to dominate high-revenue sectors of the modern economy. China aims to dominate these sectors moving forward and is investing accordingly.

Regarding social concerns, Chinese leadership prioritizes issues like social stability that could potentially undermine the Party’s control. Their solution involves technology — AI-based tools to enhance surveillance through facial recognition, gait recognition, voice recognition, and predictive analysis. These technology-based solutions monitor and control the population.

Throughout Chinese history, food security has represented the leading cause of revolutions and rebellions. For 22 consecutive years, the first document issued annually by the State Council and CCP Central Committee has addressed rural policy and agriculture — reflecting their significant concern about food security partly due to limited arable land and pollution. Again, their solution involves technology-based approaches to improve agricultural output.

Examining China’s strategy from the reform and opening period to the present reveals consistent prioritization of scientific and technological investment, seeking to leverage these advancements across multiple objectives. So yes, the aim is to “science the shit out of China.”

Divyansh Kaushik: To add to what Alex was saying, China has openly stated in many documents how they want to copy the US system. I was testifying last year to Senate Energy and Natural Resources on this topic. The Chinese 13th Five Year Plan explicitly identifies Argonne, Los Alamos, and Lawrence Berkeley national labs as crown jewels of US innovation. China aims to mimic the US national laboratory system to focus on national goals, strategic needs, and target international technological frontiers — all the points that Alex highlighted.

Jordan Schneider: Alex, could you tie that to the basic research ecosystem?

Alex Rubin: Everything I’m discussing and everything China is attempting to do is fundamentally based in the basic research ecosystem and the talent flowing into it. Xi Jinping himself has talked about how basic research is the foundation of China’s technological progress and how talent is the key enabling factor for their development.

There’s a recognition within senior leadership circles in China that to succeed in dominating future industries and technologies, they must start with investments in basic research. They’re facing challenges in shifting their investments away from applied research toward basic research, given their long-standing investments in applied areas. However, there is broad recognition that to be competitive as a global technology leader, you must invest in early-stage research, basic research, and crucially, train people to staff those facilities.

When we discuss basic research, we often focus on building infrastructure — whether purchasing GPUs or constructing data centers to train models. However, you can build the best infrastructure in the world, fill it with the best equipment, and provide unlimited funding, but if you lack people who know how to use that equipment and what to do with it, it accomplishes nothing. It inevitably comes down to having the right people working together with the appropriate training, experience, and connections to advance science and technology.

Jordan Schneider: Alex, what’s your take on the argument that China is looking for "good enough" technology as opposed to Nobel Prizes and truly frontier research?

Alex Rubin: China is essentially pursuing both approaches. China has a different interpretation of what it means to be a technology leader than the US does. The US defines technology leadership as having the most advanced technology and leading cutting-edge research. China defines it as that plus dominating markets and owning most of the world’s markets for key technology products.

For that second part of their definition, you don’t necessarily need the most advanced technology. What you need is technology that achieves perhaps 80% of the capability at 80% of the cost. When you’re looking at dominating markets, you’re considering Sub-Saharan Africa, Southeast Asia, Latin America — places that aren’t necessarily capable of affording the most advanced technology but still want the benefits advanced technology can provide.

A classic example is Huawei circa 2019. For a very long time, from its early stages, Huawei wasn’t seeking to be the world leader in telecommunication equipment. Instead, it developed technology that was "just good enough" and offered it to countries that couldn’t afford the best American options. Eventually, once the US exited the telecom equipment market, Huawei competed with European offerings by providing discounted prices and generous financial incentives — possible because of government funding, subsidies, and state support.

If you track the rollout of telecom networks through the 2000s and 2010s, from 2G networks to 3G and 4G networks, Huawei’s secret sauce was entering emerging markets in the Global South, offering technology that was perhaps not as good but cheaper, thereby gaining a foothold in these network buildouts. This strategy gave them significant revenue and market share, which they reinvested in research and development.

By 2019, they not only owned most of the world’s 4G network infrastructure but had leveraged their profits from markets like Sub-Saharan Africa to invest in 5G technology, which at the time was both better and cheaper than competitors’ offerings. This put the US in a difficult position because, arguably for the first time in modern history, it faced a major critical infrastructure buildout without a US company in the running, confronted by a Chinese company offering equally good technology at a lower price.

When we talk about "good enough" technology, it’s about broadening our definition of what it means to be a global technology leader. It emphasizes that leadership isn’t just about cutting-edge innovation but also about scale and presence in markets worldwide.

Jordan Schneider: Eva Dou, who wrote the excellent The House of Huawei, is on maternity leave, but we’ll get her on ChinaTalk at some point. One of the fascinating lessons from the Huawei story is that even though the government was pushing firms to do more R&D, the decision to spend an absurdly high percentage of revenue on R&D was Ren Zhengfei’s decision, not a government mandate. This made Huawei an outlier compared to rivals in China like ZTE, which invested only 5-7%. It demonstrates the interaction between government support, domestic scale, and visionary founders who see the long game. These founders understand that to build the most advanced technology company on the planet, you need to do the work yourself — you can’t just steal it.

Alex Rubin: You can steal your way to parity, but you can’t steal your way to leadership.

Jordan Schneider: Totally.

Alex Rubin: Another key point we mentioned earlier is the interconnection between research, customers, and manufacturing. That’s exactly what Huawei built for itself after studying success stories in the US and elsewhere. Huawei functions as both manufacturer and designer with a secure domestic market, where government support was particularly crucial in its early days. This creates an interplay that makes for a very successful, efficient model for advancing R&D.

The concerning part is that we’re seeing these same dynamics play out across multiple sectors today. If we’re not careful, we could find ourselves in the same position later this year or within two years, where critical infrastructure sectors are either reliant on Chinese technology or forced to choose between a Chinese supplier or paying more to be less competitive by going elsewhere.

Jordan Schneider: The broader American media and political ecosystem is only starting to process that China will be ahead in major commercial technologies over the next five years. We’ve already seen it in drones, telecom, and electric vehicles. Regardless of where the Chinese macroeconomic environment or American science funding goes in the coming years, we’re entering a new dynamic.

The trade-offs of keeping these technologies out of the US — which is broadly what we’ve decided to do for drones, telecom, and cars — creates a strange situation. Another important part of the China story was export discipline and the fact that many of these firms, at least in their early days, really had to compete to achieve scale domestically, both with other firms from different provinces and against companies like Apple and Tesla.

From a policy perspective, we need to remember that just because we don’t see these cars here doesn’t mean they don’t exist. They’re getting better, winning in third markets, and forcing GM, Ford, and Tesla to improve.

Alex Rubin: Even if what a Chinese company offers right now isn’t as good or is more expensive than what a US company offers, they’ve consistently shown they can leverage legacy technology to eventually move up the stack and position themselves to achieve world-leading technology. Focusing solely on where they are right now and the current quality of their offerings misses the future risks of their ability to leverage "good enough" technology to eventually generate world-leading technology, whether through theft or innovation.

Jordan Schneider: Divyansh, do you want to say anything about this?

Divyansh Kaushik: Look at where Huawei is now compared to where it was in 2019, as Alex pointed out. It’s expanding everywhere — building data centers, producing cell phones and laptops, operating undersea cables, and investing in EVs. It’s no longer just a telecom company. The same pattern is true for many other Chinese companies.

Alex’s point about projecting forward rather than just looking at a static moment in time is extremely critical, especially as we try to implement more "protect and promote" actions. We should consider where these companies want to be. They’ve laid it all out openly. Made in China 2025 was not a hypothetical document — they met every objective. The AI 2030 plan was not hypothetical either — they’re on track for their 2025 goals.

We’re sometimes overconfident about how significant our lead is. We have an uncanny ability to underestimate China’s capacity to out-hustle everybody. This is something people should be careful about.

Jordan Schneider: When I was fundraising for the ChinaTalk Institute, which now exists and is still taking donations — we’re doing great work around Chinese AI, biotech, and strategic competition — a number of funders asked if the Intelligence Community already has all this China technology information covered. What would you say about what the IC can and can’t do, and the utility of people writing about these topics independently from the government and publicly?

Alex Rubin: Fundamentally, we’re talking about commercial technologies and commercial entities — companies and academia. Many of these industries are relatively small where key players know each other. There’s an inherent need for platforms like ChinaTalk and a key role for industry participants.

Many key insights that might be technological in nature but have significant strategic implications are rumors circulating within industry or insights that companies gain from talking to customers or partners, whether in China or elsewhere. There’s significant value in that information.

Unlike analyzing the Chinese military or leadership, which is a very different challenge, the targets and developments you’re looking at here are fundamentally different. There must be a role for entities that monitor the open-source ecosystem. Otherwise, you risk missing significant trends and developments.

Last point — while we spend a lot of time discussing how China is different and how Chinese companies operate differently, they still have profit motivation, even if somewhat reduced or circumventable when needed. Many of these companies actively publicize their developments in industry press and within their ecosystem because they want visibility. They want to broadcast their technological advancement and development.

There’s so much available in open sources — small technological developments with massive strategic implications. Something like China making progress toward more advanced semiconductor manufacturing is highly technical but has enormous strategic implications for U.S. export controls and AI policy.

Jordan Schneider: I think my answer is that if you’re comparing military to commercial intelligence, maybe it’s 80/20 or 90/10 on the military side — the interesting information requires hunting and digging in an intelligence community way. On the commercial side, it’s the inverse, where 90-95% of what you need — maybe not in a specific tactical way, but at a larger strategic level of what it means for America, industrial policy, or science and technology policy — you can get by just reading publicly available information.

The relative openness of the Chinese media ecosystem when discussing commercial technologies versus operational military plans is completely different because companies need to win domestic market share, hire people, get workers excited about their companies, and raise money from investors. All of that happens under a journalism ecosystem which is, for the most part, fairly free. It’s valuable to surface this information if you have the language skills and context to process it and share it with an English-speaking audience.

Subscribe now

Alex Rubin: China’s war plan for tech is essentially their industrial policies, which they release constantly. Made in China 2025 was a very specific tech dominance plan that detailed their goals down to controlling specific percentages of industries or producing certain percentages of components. You can’t get more detailed than that.

Jordan Schneider: Before this episode, Alex, I asked if there was one document you wanted to discuss, and you pointed me to this mobilization war plan. We’ll link it in the show notes, but could you give listeners some context on why you think it’s interesting and important?

Alex Rubin: The document is from May 2022 — it’s a leaked transcript from a tabletop exercise, a war mobilization simulation in Guangdong province. What makes it interesting is that it includes representatives from the party, military, and government, all brought together in one room.

The scenario presented is essentially: "We’ve decided to invade Taiwan. What does the province do?" The focus isn’t necessarily on military movements like positioning naval vessels, but rather on how to mobilize the population and economy. In very detailed fashion, it discusses converting civilian manufacturing industries to wartime production, specifically calling out the shipbuilding sector, drone manufacturing, and other high-tech industries.

This provides a fascinating example of China preparing for potential major conflict with the US, not at the national level, but at the provincial level. They’re thinking through how to leverage their economy in wartime. If this sounds familiar, it’s basically similar to the US approach to war mobilization during World War II — that’s the scale and framework they’re considering.

They’re planning to leverage the benefits of China’s decades-long investment in expanding manufacturing capacity to essentially outproduce the US in the event of a conflict. The transcript is surprisingly detailed and covers all their considerations, from mobilizing reserves and recruiting people to converting maritime industries, aerospace repair yards, and organizing militias — everything is covered.

Jordan Schneider: When I read this, I thought it might be somewhat performative — Americans do nuclear war games for entertainment, after all. There’s something about the history of the Chinese Communist Party where national mobilization is portrayed as the most exciting time to be alive. But your sense is that I shouldn’t dismiss this entirely. Convince me otherwise, Alex.

Alex Rubin: You can find evidence of these activities in local Chinese press — I found examples just by using ChatGPT to search for relevant articles. There are numerous instances at county and prefectural levels, as well as provincial levels, of similar exercises being conducted. These are part of a comprehensive national system called the National Defense Mobilization system, which establishes cross-party government-military committees at national, provincial, county, and prefectural levels.

For example, in November 2020 in Chongqing Municipality, they conducted a mobilization exercise where civilian manufacturing companies temporarily switched their production lines to make ATVs. While not particularly advanced technology, it demonstrates them testing their capabilities.

To put this in a US context, these county and prefectural level exercises would be equivalent to officials in Fairfax County or New York City planning how they would mobilize to support a national-level conflict in the Pacific. It shows the scale and depth of their preparation and system-building.

Jordan Schneider: For our next episode on ChinaTalk, we’ll Twitch stream America’s national mobilization war plan.

Mood Music—Divyansh’s request as the song capturing the essence of China’s S&T strategy

Created by the United States Congress in 2022, the National Security Commission on Emerging Biotechnology (NSCEB) is an independent, bipartisan body tasked with assessing biotech’s national security implications. Modeled after the earlier AI Commission, the NSCEB recently released its report of recommendations to ensure the US's future leadership in biotechnology.

Their report signals something important: biotech is moving from policy background noise to strategic priority. But key parts of the discussion remain murky. The term “biotech” is used broadly, without disaggregating research from regulation and pharmaceuticals from agriculture. “China” is everywhere in the report — proving the need for additional analysis of the real ins and outs of Beijing’s biotech system. This article attempts to provide that supplemental understanding.

Let’s start by unpacking the first problem: what is biotech?

What is biotech?

The NSCEB defines biotech as “the design and engineering of biological systems, devices, and parts,” which intersects with a range of sectors: defense, industrials, consumer goods, healthcare, agriculture, and energy.

A whole-of-nation, holistic approach to biotechnology makes sense — to a point. Gene editing tools, bioengineering platforms, and computational models of biology do have broad cross-sector potential. But to craft policy, allocate funding, and set regulatory guardrails, the catch-all term “biotech” is too blunt to be useful. For each biotech sector, the stages of development, degrees of maturity, funding dynamics, regulatory environments, and end-user stakes vary so widely that generalizing becomes a liability.

Consider a few examples:

• Health biotech: a mature, well-capitalized sector about one-half of the total biotech market. It operates within a highly structured regulatory system, is driven by both public research and large-scale private investment, and faces challenges with cost, access, and long timelines.

  • Examples: mRNA vaccines, biologic drugs for cancer treatments

• Agricultural biotech: an established sector with capital markets less than one-tenth the size of pharma. It operates in its own silo of regulation and is highly influenced by public perceptions.

  • Examples: genetically modified crops, biofertilizers

• Industrial and biomanufacturing biotech: an emerging field that faces major scale-up challenges. Success depends less on regulatory approval and more on economics: whether these products can compete with traditional alternatives.

  • Examples: biofuels, bioplastics

• Defense biotech: mostly in the R&D and prototyping phase. Funds come almost entirely from government sources, goals are strategic rather than commercial, and products are often dual-use or classified.

  • Examples: biosurveillance,1 combat medicine

There’s no easy way to capture everything biotech touches in a single report, and the NSCEB’s efforts to visualize the big picture are important. Still, the complexity and rapid evolution of the field is exactly why specificity matters. Effective biotech governance requires a modular, sector-specific approach that aligns policies with actual economics, risk profiles, and social impacts.

Tailoring policy to the needs of each biotech sector will also sharpen how we evaluate progress. One of the benchmarks the US is already using is China — so it’s all the more important to take stock: what’s happening across China’s biotech landscape?

What’s going on in China?

China is the organizing principle of the NSCEB report. Chinese government and private sector activity shaped policy recommendations, defined success metrics, and fuelled a sense of urgency.

But if the Commission wants to meet the challenge it outlines, US policy must be grounded in a more detailed understanding of China’s biotech system: its strengths, its weaknesses, and its trajectory. Misreading the landscape risks building American strategy on flawed assumptions, allowing public resources to flow into the wrong problems or reinforce the wrong incentives.

So, let’s turn headlines into hypotheses. What follows is a toolkit for transforming common claims about China’s biotech sector into sharp, curiosity-driven questions – the kind that demand deeper analysis but can inform good policy.

1. The Money Question

Claim: China is spending billions on biotech.

Question: How much is China investing in biotech, and how well is it being spent?

China has identified biotechnology as a strategic priority. Budget allocations reflect this shift: China spent roughly US$34.5 billion (RMB 249.7 billion) on science and technology in 2024, and earmarked US$55 billion (RMB 398.12 billion) for 2025. However, disaggregating biotech from broader S&T initiatives — and parsing where spending falls across research, development, and commercialization — remains difficult. For instance, China’s 2024 investment of US$4.17 billion in bio-industrials and biomanufacturing is hard to categorize by stage. What’s clear is that the bulk of China’s biotech policy continues to prioritize biopharma, the most dominant part of the sector.

The efficiency of China’s public biotech funding ecosystem also requires investigation. One Chinese professor observed: “Money is no longer a major issue… the problem is with the funding system.” Government ministries, national foundations, and provincial and municipal governments all deploy biotech-linked funds, and a lack of coordination and transparency between actors can lead to redundant or wasteful spending. Fund allocation is also uneven, often favoring large, top-down projects aligned with national priorities — an approach that can be ill-suited to biotech, where breakthrough directions are inherently unpredictable. High competition and the short-term structure of grants further limit incentives for more high-risk innovation.

Private capital is equally important to understand. In key application areas like pharmaceuticals and agriculture, private investment plays an outsized role in shaping what gets developed and what reaches the market. China was the second-largest destination globally for biopharma venture funding in 2018–2019, raising around US$60 billion. Such a metric is less impressive in comparative terms: the US raised US$212 billion in the same period, and China’s innovative capabilities remain limited by its less robust private sector. But since the COVID-19 pandemic, China’s capital markets have cooled, and many biotech firms have scaled back R&D. The survival rate and long-term viability of these Chinese start-ups remain undecided.

What should the rest of the world do?

Following China’s actions, other nations could also identify biotech as a national priority deserving serious investment. The NSCEB’s primary recommendation for the US government to spend “minimum of US$15 billion over the next five years” on biotech is a necessary first step – especially at a time when US R&D spending as a share of GDP has declined significantly2 and federal science funding faces mounting pressure from grant freezes and budget cuts.

At the same time, China’s “whole-of-nation” approach to biotech comes with clear challenges: bureaucratic fragmentation, inefficient capital deployment, and misaligned incentives. For the rest of the world, effective biotech policy need not require the same holistic approach. For instance, the US already has a strategic advantage in its robust, innovation-driven private sector – the question is how to unlock that capacity more systematically. The NSCEB’s proposals for an ‘Independence Investment Fund’ to support high-potential startups, advance market commitments (AMCs) and offtake agreements to smooth demand signals, and the restoration of full R&D expensing under the tax code are smart practical levers to promote biotech.

Strong policy deploys public capital to make innovation easier: incentivizing early-stage R&D, derisking scale-up, and building stronger bridges between public goals and private execution.

2. The People Question

Claim: China is building a massive biotech workforce of domestic and global talent.
Core Question: What is the size of China’s talent base, and how sticky is it?

Over the past two decades, China has modernized its education system with the goal of cultivating world-class science and technology talent. Government education expenditure as a share of GDP rose from 2.5% in 1998 to over 4% since 2012 and has stayed above this benchmark through 2024. Such investment spans all levels of education: primary and secondary schools are integrating more STEM into their curriculums, and the number of STEM undergraduate and PhD degrees awarded in China grows each year. 45% of China’s STEM PhDs now come from top “double-first class” universities, indicating their quality.

However, academic programs may still fall short when it comes to preparing talent for biotechnology. In one survey, a third of Chinese biopharma companies reported persistent R&D hiring gaps, citing academic curricula that lag behind industry needs. Industry experience, including the ability to translate research into commercial products and to build and lead globally competitive biotech firms, remains underdeveloped. This gap in managerial and translational expertise has become a core constraint on ecosystem growth. While China’s high-skilled STEM workforce is expanding, it still represents a relatively small share of the overall population. The imbalance between supply and demand has led to intense competition for talent — reflected in an 18.0% active turnover rate in pharmaceutical R&D in 2020.3

Beijing is also seeking to bring back Chinese professionals educated or employed overseas. The National Science Fund for Distinguished Young Scholars supports scientists conducting basic research, provided they spend at least nine months a year in China. The Thousand Talents Plan offers returnees signing bonuses, high salaries and funding, housing assistance, and family support. Such programs, while not unique to China, intend to strengthen the country’s domestic research and innovation capacity.

These programs have mixed effects. Government statistics report a growing proportion of returnees each year. At the same time, both the flow of Chinese students going abroad and the rate at which they stay overseas after graduating have held steady. As of 2019, for instance, over 90% of Chinese AI talent educated in the US have chosen to remain in the US. While China has become more attractive for returnees — thanks to rising living standards, a growing private sector, and increased R&D investment — many of the original push factors remain. Concerns about academic and political freedom, limited job prospects, and digital censorship continue to shape decisions to stay abroad. Some of those who do return are frustrated with lower-than-expected salaries, shortages of postdoc positions and jobs, and reverse culture shock.

Demographic pressures are pushing China to diversify its talent base. China’s college-aged population is declining and has been on a marked downward trend for more than a decade. Ongoing labor market challenges are likely to further drive down Chinese postgraduate enrollment. Beijing has made efforts to attract global talent, including a series of immigration reforms in 2017, but those reforms have yielded limited results. Since 2017, China has risen only modestly in the Global Talent Competitiveness Index4 and still ranks below the top 35 countries. Fewer than 7% of the country’s PhD enrollments are foreign. Political, professional, linguistic, and cultural barriers continue to limit China’s appeal to global researchers.

What should the rest of the world do?

As the NSCEB observes, demand for biotech talent is growing faster than supply. China’s comprehensive workforce strategy — from early education to postgraduate opportunities — should push countries like the US to invest just as broadly in domestic and global talent. And with clear cracks in China’s own approach, now is the moment for others to act.

Fewer than 30% of American public school biology classes include molecular biology content,5 the foundation of most biotech today. Undergraduate biology curricula can be overly rigid and siloed, failing to match biotech’s interdisciplinary nature. Federal and state investments in modern lab infrastructure, teacher training, and interdisciplinary STEM curriculum development could begin to close the preparation gap.

Undergraduate and graduate research opportunities also remain unevenly funded, and too many STEM students face degree-to-career dead ends. In a field as technical as biotech, talent needs to accumulate experience and credentials in a way that stacks. That means investing in well-paid postdocs, lab-intensive training, and academic–industry bridges that allow talented people to do innovative work. One analyst notes this is where the US holds a comparative advantage over China: postdoctoral programs that offer real research experience and career growth.

At the same time, the US’s ability to attract and retain global talent is one of its greatest strategic assets. Over 75% of international students who earn STEM PhDs in the US stay for at least a decade, contributing to vital sectors like AI. Chinese officials regularly cite such US retention of Chinese talent as a key obstacle to their national goals. The NSCEB report cites Jeremy Neufeld, who emphasizes the importance of STEM immigration for the defense-industrial base, writing:

Existing restrictions on STEM immigration — and the resulting backlogs and waiting times for STEM talent — hamper the defense industrial base’s growth. Without reducing the barriers to high-skilled immigration, efforts to onshore and strengthen critical industries in the United States will face significant hurdles, and may fail altogether.

To maintain an edge in biotech, the US needs to double down on reinforcing the pull factors that make it the world’s premier destination for scientific and technical work while addressing the growing push factors that drive talent away. Staying attractive for global talent requires investing in research and building secure career pathways. At the same time, streamlining visa transitions and expanding green card availability for advanced STEM grads can help send a clear cultural and political signal that global talent is welcome.

3. The Supply Chain Question

Claim: The US is dangerously dependent on China for pharmaceutical supply chains.
Core Question: How concentrated is China’s role, and how relevant is this to biotech competition?

Pharmaceutical supply chains are commonly cited as a source of concern when it comes to China and biotechnology. However, available data does not support the narrative of overwhelming US dependence on China for pharmaceuticals. While China plays a significant role in upstream chemical manufacturing, pharmaceutical supply chains are globally distributed, multi-step, highly fragmented, and frequently outsourced — making US exposure indirect and difficult to isolate.

For example: in 2024, China was responsible for less than 6% of US imports of finished pharmaceutical products.6 As for APIs (the component of a drug responsible for its therapeutic effect), current data suggests that between 2014 and 2022, 17% of APIs used in medicines in the US were imported from China. In 2024, that share fell to 8% for prescription medications. The oft-cited “80% of APIs come from China” is an unsubstantiated misinterpretation of a government statement that has since been clarified. India and Ireland are both larger exporters of APIs and finished medicines to the US than China.

China’s footprint in pharmaceutical supply chains is primarily upstream, producing key chemical inputs and APIs that pass through multiple global intermediaries before reaching consumers. This prominence is largely in small-molecule generic drugs, which are manufactured through chemical synthesis. This matters because chemical manufacturing and biomanufacturing are fundamentally different in terms of core technologies and processes. Much of what’s described as “biotech supply chain risk” is actually about commodity chemical inputs for generics — a sector adjacent to, but distinct from, modern biotechnology.

China’s footprint in biomanufacturing requires investigation into biologics (medicines derived from living cells, rather than through chemical synthesis; typically large and complex molecules such as proteins and antibodies). In biopharmaceutical supply chains, China’s presence is limited but its capabilities are growing.

Chinese biopharmaceutical manufacturing capacity represents about 10% of the worldwide total, third globally behind the United States and European Union. Much of this capacity was built after 2016, following a major regulatory change7 that opened China’s market for contract manufacturing organizations (CMOs) that provide small-scale and large-scale drug production services to pharmaceutical companies. Multinational CMOs like Boehringer Ingelheim, Lonza, and Samsung Biologics now operate facilities in China. Among domestic CMOs, WuXi Biologics is widely considered a biomanufacturing leader by the industry, though it has recently faced setbacks due to political scrutiny. ​Following Beijing's recent pledge to accelerate biomanufacturing, China's capacity in this sector is poised for significant growth.

What should the rest of the world do?

De-risking pharmaceutical supply chains from China starts with recognizing their complexity. Exposure to China is real but often indirect. When it comes to small-molecule generics in particular, the economics of onshoring rarely add up on their own. These drugs operate on razor-thin margins, giving manufacturers little incentives to invest in quality or resilience – even with tariffs in place. Building more resilient supply chains at home will require accepting trade-offs such as higher costs for payers and consumers. Meaningful incentives for domestic production will require government intervention through policy tools such as targeted subsidies, publicly-funded manufacturing infrastructure, and expansion projects such as BARDA’s work to build out domestic biomanufacturing for vaccine production and distribution. And as supply chains are restructured, regulatory oversight will need to scale as well. In the US, a stronger, better-resourced FDA will be essential to maintaining quality, consistency, and compliance.

On the biotech side of pharmaceutical supply chains, the focus should be on reinforcing homefield advantages and lowering the barriers that keep domestic firms from scaling. One key advantage is talent: biomanufacturing depends on engineers, technicians, and regulatory specialists with deep expertise. Strengthening training pipelines — through programs like the National Institute for Innovation in Manufacturing Biopharmaceuticals, a public–private partnership highlighted by the NSCEB — can help expand this critical workforce. Regulatory credibility is another asset the US shouldn’t take for granted: the FDA still carries far more global trust than its Chinese counterparts. Meanwhile, high startup costs for facilities, equipment, and compliance systems remain a major barrier for domestic firms, requiring targeted investment to lower the threshold for entry.

In Sum!

“China” is often invoked as if it were a unified actor with perfect coordination and seamless execution. Reality is more complicated. Behind some impressive numbers and case studies are real growing pains: capital inefficiencies, inexperienced management, limited regulatory expertise, gaps in early-stage funding, underreported failures, uneven enforcement, and dependence on global partners.

In this respect, China is not an exception. Innovation is rarely a simple narrative of linear progress. But as more eyes turn to China, and as real changes unfold within its biotech ecosystem, careful observation and accurate analysis become all the more essential. Meeting this moment requires asking the right questions to understand how capital is deployed, how talent systems function, and which parts of the supply chain actually matter.

The point isn’t to downplay China’s rise in biotech. It’s to understand it on its own terms — and in doing so, craft smarter, more grounded policy.

2

6

How is Trump changing America’s extended deterrence regime? I got Polymarket to make a market on whether a US ally will acquire nuclear weapons in 2025. It’s currently trading at 6%. Are we buyers or sellers?

To discuss, I interviewed Vipin Narang, professor at MIT, who served as Acting Assistant Secretary of Defense responsible for nuclear deterrence policy during the Biden administration; Pranay Vaddi, a senior fellow at the Center for Nuclear Security Policy at MIT who worked on arms control and non-proliferation on Biden’s National Security Council; and Junichi Fukuda, senior research fellow at Tokyo’s Sasakawa Peace Foundation.

We get into…

• The historical development of the American nuclear umbrella, including the “software” and “hardware” components of deterrence,

• The probability that an American ally will proliferate by 2030, and which countries are the most likely candidates,

• Why France proliferated despite US objections,

• How the world might respond to nuclear ambitions from Poland, Japan, or Saudi Arabia,

• China’s nuclear modernization and deterrence strategies for a multi-polar world.

Listen now on iTunes, Spotify, or your favorite podcast app.

Thanks to the US-Japan Foundation for sponsoring this episode.

Foundations of Non-Proliferation

Jordan Schneider: Let’s start with a history of American extended deterrence. Where did this concept originate, and what has it achieved over the past 75-plus years?

Vipin Narang: Extended deterrence is the notion that the United States will use its range of capabilities — including nuclear weapons — to defend its formal allies and partners against nuclear attack from a common adversary. It also covers high-consequence non-nuclear strategic attacks, which is a mouthful but really means an existential threat to our allies.

The concept developed in the mid-1950s after World War II. European states had been devastated by the war and could not defend themselves against the emerging Soviet threat. The United States maintained its forward conventional deployment, but our huge advantage over the Soviet Union was nuclear weapons. The Eisenhower administration quickly realized that against a conventionally superior Soviet Union, we might need to increasingly rely on nuclear weapons to defend our forces and our allies, which were still rebuilding.

We never implemented extended deterrence out of altruism or charity. It benefited the US by preserving the openness of Western Europe and our East Asian allies, particularly South Korea and Japan, after the Korean War and World War II. This allowed us to keep the economies and political systems of our allies free, democratic, and capitalist, which created open markets for the US.

Over time, extended deterrence became important for another reason that is often overlooked — it became our best non-proliferation tool. We could prevent our allies from pursuing their own nuclear weapons if they believed our nuclear umbrella would credibly deter common adversaries. Kennedy predicted that we’d have 25 nuclear weapon states by the end of the 1960s — that was largely prevented by a credible extended deterrence architecture that the United States developed and maintained.

We worked with allies not just to deter common adversaries — China, Russia, North Korea — but to assure allies that we would come to their defense. Making this credible requires convincing allies that we can fight on their behalf without losing American cities or allied cities. The core of making extended deterrence work is being able to substantially limit the damage that Russia or China can impose on both the ally and the US homeland.

Regarding the non-proliferation point, extended deterrence has a remarkable record of preventing ally proliferation. Of our 34-35 formal allies, France is the only state that consciously decided to acquire nuclear weapons because it believed the US umbrella wasn’t credible for France’s defense. We can discuss some corner cases, but France remains the only example where perceived lack of credibility in extended deterrence caused a formal ally to develop nuclear weapons. Until now, it’s only been one case, and we hope it remains that way.

Jordan Schneider: Let’s explore this “Berlin-for-Boston” dynamic. The assumption throughout the Cold War and after was that no president would really trade Boston for Berlin. What was needed was a defense deterrence package that could plausibly threaten to keep Berlin free while avoiding an escalatory spiral that would leave Cambridge, Massachusetts in thermonuclear flames. What strategies has the US used over the past 75 years to make deterrence credible both to Moscow and to our friends in Asia and Europe?

Pranay Vaddi: You can’t separate the military dynamics and strategy from this discussion. What matters is the relationship between the head of state in the United States — the nuclear guarantor — and the head of state in the allied country receiving the nuclear umbrella. They need to be aligned. There must be a close political relationship between the United States and the 35 allies that Vipin referenced as nuclear umbrella allies.

When you put that relationship at risk or question whether the US will honor its defense arrangements — for example, if the United States declares at the presidential level it won’t defend an ally unless that ally spends a certain percentage of GDP on defense — it creates problems. Regardless of whether the United States maintains its nuclear modernization program, keeps forward-deployed dual-capable aircraft in Europe, or shows off B-2 bombers every time North Korea tests a missile, these mechanics can continue. However, every ally knows that it’s the US President who decides whether to use nuclear weapons in defense of an ally.

When the US President isn’t committed to an extended deterrence strategy as we have been for the past 70-80 years, it creates doubt. This is what we’re seeing today — countries wondering if the US will actually honor its guarantees. Because of these doubts, Poland and South Korea in particular have started discussing how to hedge against potential US abandonment of nuclear guarantees.

Just this week, we’ve seen continued statements from French officials about what a European deterrent could look like — something that Vipin and I have written about extensively. Poland has indicated it would welcome both US and French weapons. But ultimately, you need to combine military strategy with political relationships.

One example of this approach is when Vipin and I, in our previous roles, worked closely with South Korean counterparts to establish the Washington Declaration at the presidential level when President Yoon visited Washington in April 2023. This required extensive work below the surface — Vipin leading discussions at the Defense Department with his counterparts in South Korea’s Ministry of Defense, alongside White House interactions with the presidential office in Seoul. The goal was to create public language that both countries could agree to, demonstrating a renewed and enhanced commitment to extended deterrence with commitments from both sides.

South Korea needed to recommit to the Non-Proliferation Treaty, while we needed to recommit to South Korea’s defense. We agreed to do so more transparently, giving South Korea greater insight into US nuclear planning and more input into how nuclear operations could be conducted. We needed to address both military and political aspects to provide reassurance in that bilateral relationship.

Jordan Schneider: Let’s stay on history a little more. We’ve got the credible military component — “I can actually bomb places” — and we’ve got the credible political aspect — “I actually care about your country enough to do something if it gets attacked.”

Vipin Narang: We called it “hardware” and “software.” The hardware piece is actually easier because it’s about having flexible capabilities to limit damage to the ally and the homeland. The theory is relatively simple, but implementation is not. This is something the Department of Defense spends considerable time contemplating.

Pranay’s point is really important — we underemphasize the importance of the software piece both historically and today. The only change from the Biden administration to the Trump administration is the software piece. That alone explains why we’re having this podcast. It’s almost like a blue screen of death for some allies. They don’t question our capability — they question whether, when the time comes, we will actually attempt to limit damage to the ally and US homeland and fight on their behalf.

Jordan Schneider: Let’s explore the history a bit more. What have been the sketchiest software moments since 1950?

Vipin Narang: The Eisenhower administration raised concerns right out of the gate. The plan was to reduce our conventional footprint, but that conventional presence in Western Europe reassured allies because we had skin in the game. Without skin in the game, the fear was we wouldn’t pull the trigger if the Red Army came crashing through.

The debate during the transition from the Eisenhower administration to the Kennedy administration centered on whether we had flexible enough options for a theory of escalation management. There have been various ups and downs, but what we’re seeing today is new: questioning whether we even care about our allies. Article 5 is a hallowed pillar of NATO and essentially our commitments to South Korea, Japan, and Australia.

For the first time, we’ve had an administration that basically says we might not honor Article 5, and allies might be on their own if they don’t pay enough. One of the bigger implications of the Signal chat leak was revealing that in private, the Vice President essentially said, “I hate Europe.” If I were Europe, I might question the willingness — not the ability — of the United States to step in when necessary.

Junichi Fukuda: Regarding the crisis of extended deterrence, we should consider the case from the early 1970s and late 1960s. At that time, the Guam Doctrine, or Nixon Doctrine, resulted in conventional forces withdrawing from the Indo-Pacific area alongside the withdrawal from the Vietnam War. This created a genuine crisis.

There was no clear nuclear extended debate then. Most countries were deciding whether to sign the Non-Proliferation Treaty (NPT) around 1969-1970. At that time, the Japan-US alliance didn’t explicitly mention the nuclear umbrella. The term “nuclear umbrella” was first used in 1975, yet Japan decided to sign the NPT during détente. The tension between the US and Soviet Union had decreased, allowing us to join the NPT.

Officially speaking, there was no extended nuclear debate at that time. This provides an analogy for our current situation — the United States withdrawing forces while the credibility of extended deterrence appears to be in crisis. We must learn from the experiences of the early 1970s.

Jordan Schneider: Perhaps the distinction between then and today was détente. You didn’t really see nuclear war on the direct horizon. China was busy with its Cultural Revolution, and the Soviet Union was being friendly. It wasn’t the most frightening period in thermonuclear history. Does that change the dynamic compared to today?

Junichi Fukuda: Because it was the détente era, tensions were reduced, so we didn’t have to deeply consider the implications of the extended deterrent crisis — we simply accepted it. In our current situation, US-China relations have deteriorated, and US-Russia relations are poor. Tensions are increasing while the United States appears to be withdrawing its engagement with the outside world. This creates a crisis.

Pranay Vaddi: Another historical element worth mentioning — something I always try to write about when Vipin and I collaborate, though he always deletes this paragraph — concerns extended deterrence from the early 1960s. There’s an interesting story about when a German Chancellor visited JFK after his White House victory.

Konrad Adenauer represented the CDU — the center-right in West Germany — and was essentially JFK’s opposite. Adenauer was about 85 years old while Kennedy was this young, new president. Kennedy was deeply concerned about allied proliferation, as Vipin outlined regarding developments in the 1950s and 60s.

According to declassified documents recording their meeting, Adenauer essentially said, “I hear you have weapons in West Germany, but I’ve never seen them. I’ve been repeatedly told those warheads are there, but I haven’t been shown them.” He wanted his Defense Minister to understand what the US was actually doing in their territory — what they had based there.

Kennedy assured Adenauer it made sense and promised that Defense Secretary Bob McNamara would brief him and his defense minister on the US stockpile operational plans — when we would actually use them if the Warsaw Pact attacked.

To me, this marked the beginning of the types of requests we in the US Government would regularly hear from allies — “What are you actually doing in a nuclear scenario?” As mentioned regarding the Washington Declaration and US work with South Korea, Japan, and NATO, we conduct tabletop exercises, scenario-based discussions, and simulations to demonstrate that the US has thoroughly considered extended deterrence scenarios. This marries hardware and software to make extended deterrence effective. Its origins date back to that conversation.

Retrenchment vs Burden Sharing

Jordan Schneider: If we’re using a relationship analogy, we’re saying, “We’re going to give you a tour of our place. We’ll let you bring some toothbrushes into the bathroom.” The US has been doing this 75-year dance with its allies, broadly trending toward reassuring everyone so they feel safe and secure enough not to develop their own nuclear weapons.

The question now is, what does the decay function of that look like? The Trump administration isn’t going to destroy our B2 bombers anytime soon, so the capability isn’t disappearing. However, the trust has developed some real hiccups, particularly on the software side.

Where do we start with this? What’s the right conceptual framework to understand the signals that the Trump administration is sending to the world?

Vipin Narang: Let me quickly address the hardware aspect. We shouldn’t overlook the hardware adjustments that were ongoing or being laid as a foundation at the end of the Biden administration. This includes ensuring the modernization program for each leg of the nuclear triad — we’re replacing each leg simultaneously for the first time in history, and we’re facing challenges.

The Sentinel ICBM is facing delays. Attention and focus are needed to prevent the modernization program from falling further behind schedule. The current administration hasn’t issued an executive order on nuclear forces, though it has addressed missile defense architecture, which is related but represents a very long-term strategy.

The immediate concern is our progress on this modernization program and how we’re dealing with China’s emergence — something we didn’t anticipate when we sized the modernization program 15 years ago. We don’t need forces matching the combined strength of our adversaries, but we might need more than what we planned when discussing New START numbers in 2010. At that time, Russia was considered a partner, and China and North Korea hadn’t begun their expansions.

For an ally examining the hardware aspect, there are mixed signals. The president has stated he doesn’t want to allocate more funds to nuclear weapons, which might be concerning. However, the more pressing issue, as Pranay and Fukuda-san mentioned, is that the software component isn’t experiencing a decay function — it’s a complete discontinuity. The posture toward our European allies has transformed dramatically since early January 2025.

If I were in the Indo-Pacific region — and I’d welcome Fukuda-san’s thoughts on this — there’s considerable uncertainty. The administration has declared China the pacing threat and pledged to focus on China, yet simultaneously continues to pressure Japan and South Korea economically in ways that may undermine the credibility of our commitment to defend them against existential threats from China or North Korea. This credibility is arguably much weaker than it was in January 2025.

Pranay Vaddi: The fundamental question is whether this is about right-sizing burden-sharing between the United States and its allies or pursuing a retrenchment ideology. If you’re an ally like Japan, you can approach the United States saying, “We understand you want everyone to pay their fair share. You want Japan to prioritize China as your primary concern, and we’re contributing to the extended deterrence strategy that Indo-PACOM needs for a Taiwan Strait crisis."

Japan is developing air and missile defense capabilities, long-range conventional strike capabilities, and a blue-water navy. They’re willing to open their shipyards for replenishment and repair of US naval vessels. In return, the US needs to maintain a strong nuclear umbrella and guarantee. That’s the basic exchange, but we should develop this strategy together.

If it’s primarily a burden-sharing question from the US, that’s a strategy to present to the White House: “Let’s determine the appropriate payment amount and capabilities within a bilateral strategy — or trilateral if South Korea is included, or quadrilateral with Australia as another extended deterrence ally in the Indo-Pacific."

However, if this is about pursuing a retrenchment foreign policy — if that’s what “America First” means — which I believe is an unseen battle within the administration, then allies have limited options. As an extended deterrence ally, you can only be so persuasive because ultimately, you need a United States willing to engage with allies, be transparent, and fulfill the software element of extended deterrence. Being credible on the software side seems contradictory to a retrenchment strategy.

Allies need to determine whether this is about recalibrating burden-sharing in the new political climate or if the United States is reducing its international commitments. The second scenario is much more challenging and drives allies to consider hedging strategies.

Jordan Schneider: It’s complicated because the answer is both. Pranay pointed out that some people believe one thing and others believe something else. We have a president who frequently changes his mind. So what’s the appropriate response?

Junichi Fukuda: Regarding the hardware issue, I’m not overly concerned about the modernization of US strategic nuclear forces. The United States is modernizing its strategic nuclear weapons, and there’s an option for uploading nuclear warheads on existing platforms.

My concern lies more with the future trajectory of developing theater nuclear forces, such as sea-launched cruise missiles from submarines. These types of theater nuclear weapons would be essential for Japan’s defense in the future. Given the current administration’s position on cutting the defense budget, I question whether such new nuclear capabilities will ultimately be realized.

On the software issue — the political relationship between the United States and Japan — there are several possibilities. First, there’s conflict in the trade area. The United States has increased tariffs on Japan, and we’re currently negotiating this matter. Japanese people might find it difficult to rely on the United States for security while we’re engaged in economic disputes.

Simultaneously, the Trump administration is attempting to address the challenge posed by China. Most security experts in the administration appear to be “prioritizers” — they would withdraw from the European theater but concentrate on strengthening capabilities in the Pacific to counter China’s threat. This approach provides some reassurance to us.

The situation remains difficult to analyze fully at present. Overall, we remain confident in continuing the alliance relationship for now, though the future is somewhat unpredictable.

Jordan Schneider: It’s interesting, Junichi, because you’ve seen this incredible earthquake in European politics triggered by just a handful of speeches. It seems to me that only people like you are paying close attention, not necessarily the entire Japanese or South Korean political establishments. Am I wrong? Do you think it would take something as explicit as a J.D. Vance or Pete Hegseth speech saying, “We don’t care about you guys,” to trigger the same kind of awakening that seems to have happened in Poland and Germany?

Junichi Fukuda: That’s a nightmare scenario we hope to avoid. Currently, there appears to be conflict within the administration. One faction focuses on confronting China, recognizing the necessity of maintaining US hegemony in the Indo-Pacific region. The other faction consists of isolationists who advocate withdrawing US forces from every theater worldwide, preferring to concentrate resources on domestic politics. The future direction of the current administration remains uncertain, with potential internal struggles ahead.

Jordan Schneider: The very existence of this debate is unprecedented from a post-1945 perspective. Perhaps now is a good time to discuss the politics of nuclear armament in both South Korea and Japan. If we remain in this awkward, subterranean debate — or move toward the more alarming scenario of J.D. Vance expressing indifference toward Japan — how might this affect Japan and South Korea’s nuclear postures?

Junichi Fukuda: Japan’s nuclearization has traditionally been an unthinkable issue, not widely discussed. However, as someone studying nuclear deterrence, I’ve had many opportunities to address this topic. Last month, I attended a conference where American experts frequently discussed the possibility of Japan nuclearizing in the future.

The primary condition under which Japan might consider nuclearization would be a lack of confidence in the credibility of US extended deterrence. A secondary scenario would involve South Korea pursuing nuclearization first, which might prompt Japan to follow suit. The specific circumstances that would trigger such scenarios remain unclear to me.

Pranay Vaddi: From a US perspective, it’s difficult to imagine any scenario where the United States benefits from ally proliferation. We’ve seen comments by the Vice President questioning whether stationing US nuclear weapons in Poland makes sense — indicating that proliferation of US weapons to other allies isn’t necessarily on the table. This struck me as an off-the-cuff reaction rather than the product of an interagency process.

The challenge for the Trump administration is reconciling several objectives — preserving the status quo regarding the number of nuclear-armed countries, upholding the Nuclear Non-Proliferation Treaty, and renegotiating burden-sharing agreements or pursuing retrenchment. These goals conflict with each other.

Japan is a technologically advanced nuclear country with nuclear technology and reprocessing capability. They’ve committed to neither possessing, producing, nor introducing nuclear weapons into their territory as part of their peace constitution. They remain an importantUS partner in nuclear energy, extended nuclear deterrence, and nuclear disarmament. Japan uniquely embodies all the contradictions of nuclear policy in one bilateral relationship.

Countries like Japan, South Korea, or Poland cannot simply decide to acquire nuclear weapons overnight. The process requires specialized facilities, legal and regulatory changes, withdrawal from the NPT, and consideration of potential US export controls and sanctions. This timeline gives adversaries opportunity to react — not just economically. If North Korea observed South Korea pursuing nuclear weapons while US extended deterrence guarantees appeared shaky, they might preemptively strike those facilities or initiate conflict. The risk of conflict increases when an ally with weakened US security guarantees pursues nuclear capabilities — an important consideration for countries weighing this decision.

Vipin Narang: Those who identify as “America first” — which includes all of us, as I reject the notion that only the Trump administration puts America first while the Biden administration didn’t — should oppose allies acquiring nuclear weapons for at least three reasons.

First, we have historically opposed allied proliferation in Washington because we wanted to minimize the number of independent decision centers for initial employment or escalation decisions. This is a sanitized way of saying we didn’t want France starting a nuclear war we’d have to finish. The French — whom I genuinely respect — had a strategy where their independent nuclear capability would be employed when the French President determined vital interests were at stake. However, they couldn’t finish such a conflict, leaving the US to manage the aftermath against the Soviet arsenal. This thinking persists today when discussing the credibility of European deterrence. France can initiate a conflict, but cannot conclude it.

Similarly, other allied nuclear powers could never develop damage limitation capabilities comparable to the United States independently. An allied proliferator would acquire nuclear capabilities that could initiate a nuclear war and chain-gang the US into it — precisely what the US has always sought to avoid.

Second, while Kenneth Waltz famously argued “more may be better” from a retrenchment perspective, the risk of accidents increases substantially. Smaller states with limited arsenals facing more capable nuclear adversaries with head starts might develop itchy trigger fingers or lack the institutional infrastructure to effectively manage these weapons. The accumulated risk of accidents grows with each additional nuclear-armed state.

Third, as Pranay and Fukuda-san mentioned, the threat to the NPT is real, as is the potential cascade effect if one ally — particularly South Korea, given the growing political consensus there for at least developing a hedging capability — acquires nuclear weapons. I doubt the NPT could survive a democratic state in good standing withdrawing to pursue nuclear weapons. Japan might follow, along with Poland, Ukraine, Germany, Saudi Arabia, Turkey, and a number of other countries if the floodgates opened.

Pranay’s point is important: these countries would pursue nuclear weapons in an environment where their adversaries have both a vote and a head start. Historically, no state has acquired a credible nuclear weapons capability in less than five years from the decision point, and a survivable arsenal typically requires a decade or more.

Regarding probabilities — there’s a zero percent chance an ally proliferates by the end of 2025? The timeline doesn’t work. By 2030? I’d wager the odds are more than 50 percent, though as an academic I don’t have much money to bet — that at least one ally will have a credible latent nuclear capability by 2030-2035 if current trends persist.

Jordan Schneider: Just wait until you get all those Polish and Saudi contracts to help write their nuclear doctrine.

Vipin Narang: You can reach Pranay, my agent, at prvaddi@mit.edu.

Jordan Schneider: Let’s talk about France. They were exceptionally difficult throughout the Cold War, no?

Vipin Narang: I love them, but one France is enough.

Jordan Schneider: This is precisely the point about “America First.” Nuclear weapons are incredibly powerful — they provide tremendous leverage, allowing you to dictate terms to other countries, constrain their actions, and influence whether they initiate conflicts.

Consider how the US effectively prohibited Chiang Kai-shek from invading mainland China in 1962 because he lacked nuclear weapons. Our ability to maintain global stability and, as Vipin mentioned, prevent conflicts that we would ultimately have to resolve actually enhances American security. The alternative makes the world a lot more dangerous.

The notion that Putin would remain passive for five years while Poland develops nuclear weapons is unrealistic. We’ve already witnessed this scenario play out with Israel, Syria, and Iran. Russia would undoubtedly take action. Imagine the consequences of a Polish nuclear facility being bombed by Russian aircraft — the global ramifications would be catastrophic.

Vipin Narang: Poland remains a NATO member, so that scenario involves Article 5 and risks war with Russia over a counter-proliferation strike. These are low-probability events.

Pranay Vaddi: Attempts at retrenchment could actually accelerate the path to great power confrontation that you’re trying to avoid. Regarding our discussions with French colleagues — you don’t necessarily need to implement extended deterrence exactly as the United States has done. We’ve pursued an expensive strategy heavily reliant on nuclear capabilities that requires an enormous defense budget.

Perhaps in their Euro-deterrent approach, the French don’t need 3,800 nuclear warheads or a triad similar to ours. They might achieve more with conventional forces and missile defense systems. However, they ultimately need an approach that accomplishes the same objectives: reassuring eastern flank allies on Russia’s borders that you could meaningfully engage in conflict and either deter or defeat Russian aggression against those allies.

This is extraordinarily challenging. The United States has invested over seven decades in continuously refining this approach. The work on extended deterrence is never complete. As Fukuda-san knows well, the meetings between the United States and Japan throughout the year on extended deterrence are extensive, regularly scheduled, and ongoing. It requires dedicated staffing and expertise on both sides. This isn’t something another country can simply adopt if the United States withdraws those commitments.

Jordan Schneider: What frustrates me about much of the MAGA-influenced foreign policy is its simplistic thinking — reminiscent of middle school debate logic.

There are reasons why the world exists in its current state. I thought conservatives valued the principle of Chesterton’s Fence — not dismantling established systems without understanding their purpose. The transition to a new world order could be dangerous and catastrophic.

Spending an extra $500 billion annually to prevent 20 countries from acquiring nuclear weapons actually represents an excellent deal.

Vipin Narang: It is indeed an excellent deal.

Jordan Schneider: A historically unprecedented deal. You would have accepted this arrangement 100 times out of 100 if you were present at Alamogordo in 1945.

Vipin Narang: This represents a bipartisan consensus. Republican or Democrat — Pranay and I often find ourselves aligned with traditional Republican views on strategic deterrence and extended deterrence, sometimes surprising ourselves with how much common ground we share. Those traditional voices are being marginalized, if not completely excluded, from the Trump administration.

This represents a different approach than anything we’ve seen in the post-World War II era. There was bipartisan consensus that we were committed to extended deterrence, with debates focusing on implementation methods rather than questioning the fundamental commitment itself. That’s why this moment seems distinctive.

To be fair, I don’t know where the Trump administration will ultimately land on this issue, as Fukuda-san and Pranay both noted — there are various competing voices within it.

Vipin Narang: At the core of extended deterrence credibility lies predictability and consistency. The very existence of this debate creates problems for the credibility of extended deterrence.

Jordan Schneider: Junichi, how are you feeling about all this?

Junichi Fukuda: At the administration level, I cannot predict the future because President Trump is unpredictable. However, at the congressional level, there exists concrete bipartisan agreement. The United States must continue extended deterrence toward its allies. Two years ago, a congressional report from the Bipartisan Commission — the Strategic Posture Report — emphasized continuing and even increasing the credibility of extended deterrence. At the congressional level, bipartisan agreement will persist in the future. Administrations last just four years, but Congress continues.

Jordan Schneider: I appreciate your optimism. I’m curious, Junichi — when you first entered this field and began studying nuclear extended deterrence and nuclear policy, what were the most extreme scenarios being discussed? How much more extreme is our conversation over the past 45 minutes compared to anything you’ve encountered before?

Junichi Fukuda: I began studying international relations in the late 1990s. At that time, the most severe concerns involved bioterrorism around 1997 or large regional civil wars. Nuclear war between great powers was definitely not considered a possibility. Now we’re discussing nuclear war. The situation has changed dramatically, forcing us to contemplate the unthinkable.

Pranay Vaddi: For the past 30 years, nuclear policy as a subject area was limited to specialists and experts. Now this debate about extended deterrence reminds us how deeply connected US nuclear strategy is to other elements of US grand strategy and foreign policy goals.

After the Trinity test, as nuclear strategy developed, it became integral to American grand strategy for confronting and containing communism, rebuilding decimated countries including close allies and former adversaries like Germany and Japan. This created a remarkable market for American goods and innovation through the Marshall Plan and other foreign aid efforts.

If you disconnect extended nuclear deterrence from foreign policy and our economic strategy, you can have simplistic conversations about whether we should continue extended deterrence or maintain forward-based troops. The reality is these elements have been interconnected for over 70 years.

Vipin and I have written about how much effort the United States put into rebuilding the world in the 1950s and 60s, ensuring that the United Nations, economic integration, and globalism — which nationalist trends in every democracy now resist — would help prevent another world war. Now the administration is attempting to remove the economic pillar while simultaneously questioning sacred principles of extended nuclear deterrence.

Our assumption when we wrote about this in January was that the economic benefits to the United States would eventually lead back to a coherent extended nuclear deterrence strategy. However, when economic interdependence and the accompanying political relationships diminish, I might question the American commitment to extended nuclear deterrence if we become less reliant on international trade due to efforts to retrench and reshore everything from industry to military capabilities.

These elements are all connected. Retrenchment, if implemented comprehensively, can severely damage US national security.

Deterrence with Chinese Characteristics 玉石俱焚

Jordan Schneider: We’ve progressed quite far into this discussion without mentioning China’s efforts to update its nuclear arsenal. This factor seems irrelevant in the Trump calculation, but it’s worth discussing how countries in Asia perceive this development. Junichi, could you provide a brief overview of what has happened in China over the past decade, and what this modernization means for countries in the region as they evaluate their strategic environment?

Junichi Fukuda: Historically, China’s nuclear assets were limited to approximately 200 nuclear weapons. At that time, analysts described China’s nuclear strategy as “minimum deterrence” — though this is a Western concept that might not align with China’s actual thinking.

China has gradually changed its approach to nuclear weapons. Recently, China has rapidly increased its nuclear arsenal, adding approximately 100 nuclear warheads annually. In the long term, China’s strategic nuclear arsenal may reach 1,500 warheads, potentially achieving parity with the United States and Russia.

The purpose behind this expansion remains unclear, but several possibilities exist. One apparent reason is to gain national prestige and achieve the same international status as Russia and the United States by creating parity with them.

Another potential reason involves creating a strategic stability-instability paradox in the Taiwan Strait contingency. By increasing its strategic nuclear arsenal, China might deter United States intervention against Chinese aggression toward Taiwan.

Their final objective, however, remains unknown. China has stated that by 2029, they intend to build a “world-class military,” which likely means achieving the same scale as the United States. This suggests their nuclear forces will eventually match the current US strategic nuclear arsenal.

Jordan Schneider: A question from Japan’s perspective — does the difference between 200 versus 1500 nuclear weapons actually matter?

Junichi Fukuda: Currently, it doesn’t matter as much, but we’re concerned about what would happen in the next decade. Our nuclear concern is focused on the medium to long term, particularly around 2030-2035.

Jordan Schneider: Why? How do more nuclear weapons with better delivery capabilities matter strategically?

Junichi Fukuda: If China possesses 1000 or 1500 nuclear weapons, they might be able to deter the United States from intervening in a Taiwan Strait contingency, creating a decoupling and instability paradox. This would be a major concern for us. In the short term, we’re more concerned about China’s increasing conventional capabilities, which are changing local military balances and might enable China’s invasion of Taiwan by 2027. But regarding nuclear issues, our concerns focus on the 2030-2035 timeframe and beyond.

Vipin Narang: China’s expansion is real and occurring faster than we anticipated. It’s designed to challenge the United States’ force sizing principles. While we don’t fully understand their motivations, we must prepare for the possibility that China is developing a nuclear posture and strategy to provide cover for regional aggression and coercion against US interests and allies.

We took this very seriously in the Biden administration. For context, we have these formal documents like the Nuclear Posture Review, which is unclassified and available to everyone. Following this parent document is the Presidential Nuclear Weapons Employment Guidance, which I worked on with Pranay. There’s an unclassified report to Congress known as the “491 Report” that I would recommend reading closely.

When the previous employment guidance was issued during the Trump administration around 2019, China was just emerging as a nuclear concern. There was recognition that China would challenge the US force posture, but the specific nature of that challenge wasn’t yet clear. The current document represents an evolution, with continuity between the Trump and Biden administrations’ guidance. However, it takes seriously, for the first time, the prospect of multiple peer strategic challenges, where China can no longer be treated as a “lesser included case.”

An important point that’s often overlooked is that it’s not just about the numbers. China will reach approximately 1,000 nuclear weapons by 2030 and potentially 1,500 by 2035, which represents quantitative parity with New START treaty levels. What’s more concerning is the composition of their force development.

If China had remained consistent with its long-standing assured retaliation strategy and focused solely on a sea-based force similar to the UK, we wouldn’t need to adjust our nuclear force posture. Instead, they’re developing hundreds of hardened silos in remote areas, perfectly spaced, which serve as nothing other than counterforce targets for the United States. Given our strategy to limit damage to the US homeland and our allies, China knows we must target and hold those forces at risk.

Under New START, we only have 400 ICBMs with single warheads and a limited number of SLBMs (Submarine-Launched Ballistic Missiles) at sea at any given time. Hundreds of Chinese ICBMs in remote regions would consume a large portion of our strategic forces and could make us vulnerable against Russia. We’re observing collusion and coordination between Russia and China. If we’re engaged in conflict with Russia and have expended some nuclear forces to maintain strategic deterrence because of their actions in Ukraine, China might see an opportunity to act because we no longer have sufficient forces to maintain strategic deterrence against them.

Their force structure appears carefully and intelligently designed to undermine our capabilities and sizing principles. We recognized the need to respond in ways that enable us to maintain both strategic and regional deterrence in a multiple-peer world across multiple theaters, especially given the coordination between Russia and China — and we shouldn’t underestimate North Korea either.

We conducted a year-long process to assess adjustments to the US force posture for strategic and regional deterrence in this new reality. As Fukuda-san mentioned, we’re implementing changes because Congress mandated it, but these changes also offer deterrence advantages. The security environment continues to deteriorate, and we can no longer deny the value of having a purpose-built maritime regional deterrence capability that frees up strategic platforms for strategic deterrence.

This challenge demands our attention. I’m concerned about the administration potentially rushing to reach a deal with Russia on arms control to extend New START without considering the implications for China. During the first Trump administration, Marshall Billingsley actually attempted to coordinate arms control strategies. I’ll defer to Pranay on this since he knows more about it than I do. But that approach is crucial — these issues must be coordinated within our own strategy. We need to consider what we’re willing to accept with Russia while keeping China in mind, and vice versa.

Pranay Vaddi: I’ll just highlight one additional point. Beyond everything that Vipin and Fukuda-san have raised regarding China’s changing posture — the warhead buildup, the decision to place nuclear weapons in silo ICBMs (which are vulnerable to nuclear strikes and therefore need to launch quickly in a nuclear war scenario), and their approach to survivability that the US would have to contend with — the Chinese are also modernizing cyber capabilities, counterspace technologies, and conventional precision strike systems. These capabilities could be categorized as counter-intervention measures since the US is also arming itself for a potential conflict in the Indo-Pacific. Alternatively, they could be used aggressively as part of a strategy to coerce under the nuclear umbrella.

The reality is that much of this will depend on one person’s decision-making and whether China decides to maintain or change its historic nuclear doctrine established since Mao’s time — characterized by restraint, minimum deterrence, and a no-first-use policy. If any aspects of this doctrine are changing, we probably wouldn’t be the first to know, as we’ve been unable to sustain diplomatic dialogue with China on strategic stability and nuclear policy issues for the past decade or two.

The Biden administration held just one consultation with Chinese colleagues on this particular topic. As with many other national security issues in the US-China relationship, China’s willingness to discuss nuclear matters depends on the political climate between our countries. If conditions seem favorable for US-China discussions across various topics, they might engage on nuclear issues as well. However, if relations are unfavorable — as they have been for several years due to incidents involving surveillance balloons, Nancy Pelosi’s visit to Taiwan, or tensions during the Trump administration — they won’t discuss these matters with us.

This creates a fundamental problem — how are US policymakers supposed to understand the nuances of China’s policy regarding its modernizing nuclear posture if we can’t meet for discussions? Setting aside any attempts at arms control agreements or diplomatic efforts to limit China’s buildup, progress is impossible unless both sides can engage in dialogue. Despite having one meeting during the Biden administration, all our attempts to encourage, incentivize, or facilitate a second meeting were unsuccessful.

There have been some positive developments. Before leaving office, Presidents Biden and Xi established a joint understanding regarding the use of artificial intelligence in nuclear employment decisions, essentially agreeing that humans should remain in the loop for any nuclear deployment or termination decisions made at the presidential level. Additionally, the two countries now exchange notifications before intercontinental ballistic missile tests. While these are small confidence-building measures, they don’t address the central issue — China is the only nuclear weapons state under the Non-Proliferation Treaty that is quintupling its arsenal.

The United States and China must discuss this issue, or else, in worst-case scenario planning conducted by the Pentagon, we’ll have to assume that China intends to build an arsenal qualitatively equivalent to that of the United States — one designed to undermine US nuclear strategy, as Vipin mentioned. This would force the US to respond with its own buildup, creating a world that benefits no one.

Jordan Schneider: Junichi, you’ve been the most positive and sanguine out of all four of us throughout this conversation. Would you like to conclude on a high note? What are you looking forward to over the next few years?

Junichi Fukuda: Actually, I’m not particularly optimistic. Having nuclear weapons in Japan is politically and economically challenging, with very high barriers. We simply don’t want to consider this option, which is why we’re always concerned about strengthening the credibility of extended deterrence and maintaining our alliance relationship.

The future is unpredictable, especially under the Trump administration, so we must consider various scenarios. However, nuclear weapons are not the most desirable path for Japan. The least desirable option would be to bandwagon with China, which we must avoid at all costs. If we face a truly existential situation where we lose confidence inUS extended deterrence, we might decide to pursue nuclear weapons despite the damage this would cause to the Non-Proliferation Treaty. But this is certainly not our preferred outcome.

Jordan Schneider: Why are you so confident that nuclear weapons would be the preferred path versus an accommodation with China?

Junichi Fukuda: Bandwagoning with China would require compromising Japanese sovereignty. Looking at Japan’s long history with China, there is no precedent for such a decision. Japan has at least 1,500 years of recorded history, and during that time, our country was occupied by a foreign power only once — from 1945 to 1952 by the United States. We have never experienced Chinese occupation of our land.

This represents a fundamental difference between South Korea and Japan. South Korea has a long history of being part of China’s tributary system, but Japan has no such experience. We cannot align with China under any circumstances.

To avoid that scenario, we might need to consider independent nuclearization if we lose confidence in the alliance.

Jordan Schneider: Vipin, Pranay — we can’t end on that note.

Vipin Narang: Let me try to be more optimistic. There’s considerable uncertainty right now. We transitioned from an administration that was very careful with its words to avoid sending mixed signals or giving the impression of internal divisions. For the most part, we spoke with one voice and maintained discipline on these issues.

The Trump administration has a different style. The president speaks frequently, and his statements aren’t always consistent. He doesn’t like nuclear weapons, but if that’s true, he should appreciate extended deterrence — though that connection isn’t always obvious. There’s certainly a battle within the Trump administration over the direction of foreign policy and grand strategy, as Pranay mentioned.

We should take a moment and wait to see which direction this goes. The sky isn’t yet falling on extended deterrence or NATO. The NATO ambassador was very clear about the US commitment to NATO. It would be prudent for allies — and this was true during the Biden administration as well — to contribute more, which they have done. The burden-sharing issue is real, bipartisan, and consistent. This focus isn’t necessarily bad for American security, regardless of which administration is in office.

However, there’s a possibility that things could move in a different direction. It’s important for allies to consider what a world without American leadership in the alliances might look like, while also recognizing that the bipartisan continuity we’ve seen on these issues might prevail in practice. It’s still early — only 90 to 100 days have passed — so I remain in wait-and-see mode. I continue to hope for the best, but it isn’t imprudent to prepare for a very different world.

Subscribe now

Pranay Vaddi: This issue is central to US national security. The more people outside the nuclear space learn about extended nuclear deterrence — its history, our current strategy, what we’ve done and why — from people like Vipin who’ve held important positions at the Pentagon or from allies who have experienced being recipients of extended nuclear deterrence, the better.

Every time I see foreign policy generalists learn about extended nuclear deterrence, a light bulb seems to go on. They understand that this is fundamental to what the United States has built in the world, benefiting the average American. I’m somewhat optimistic that as more people become aware of these issues and critically examine foreign policy decisions that might undermine extended nuclear deterrence, they’ll speak out in support of not just maintaining the status quo but improving our approach.

I want to thank you, Jordan, for having us on. Podcasts, articles, news coverage — all of these help people understand why we approach extended nuclear deterrence the way we do and recognize that having more states with nuclear weapons is inherently detrimental to the United States. I believe people in the current administration instinctively understand this as well. The key will be connecting these dots to help them conclude why specific actions in extended nuclear deterrence are necessary. That’s the source of my optimism.

Jordan Schneider: Let’s close with some reading recommendations. How about two from each of you? Jun, would you like to start with two books?

Vipin Narang: I’d like to recommend Under the Nuclear Shadow by one of my former students, Fiona Cunningham, which just came out.

Jordan Schneider: You can’t recommend that one! We already did a show with her.

Vipin Narang: I highly recommend it. My second recommendation is the updated version of Managing Nuclear Operations. This is quite technical. While Fiona’s book is high-level and accessible for anyone interested in China, Managing Nuclear Operations is on the detailed end of the spectrum. Charlie Glaser, Austin Long, and Brian Radzinsky re-edited this classic with updated chapters from experts like Franklin Miller and James Miller. It provides an excellent introduction to how US nuclear strategy and policy are developed. If you’re interested in that topic, I highly recommend it — it’s a very good update to the classic 1987 volume edited by Ash Carter, Steinbrenner, and a third co-editor whose name escapes me.

Jordan Schneider: Are we telling people not to read Annie Jacobson’s Nuclear War?

Pranay Vaddi: I haven’t read it, so I can’t speak to whether it should be recommended or not.

Jordan Schneider: You guys are so diplomatic. I’ll say it directly — don’t read it. It reads like a journalist who doesn’t really know what they’re talking about. I caught one or two historical errors just from my limited knowledge. It gave me the impression of someone with too much of an agenda to be careful with their sources.

Vipin Narang: I appreciate the original sources. I recently taught my class on extended deterrence and discovered an obscure article by James Schlesinger from 1962 explaining why we extend deterrence to Western Europe and NATO. Remarkably, the arguments haven’t changed between 1962 and 2025. It’s a RAND chapter that’s quite difficult to find. I was stunned when I read it carefully, realizing how much we’re simply reinventing the wheel. [Link here!]

I’m very humble about the fact that scholars like Schelling, Brody, Schlesinger, Earl Ravenhall — all published in International Security — along with Ash Carter, Frank Miller, and others who preceded us, had already thought through all these issues. We’ve forgotten many of their insights because we had a 35-year interregnum when we didn’t have to consider the importance of extended deterrence and nuclear deterrence more broadly. For the audience, I recommend going back to the original texts — you’ll be surprised how little has changed.

Jordan Schneider: Pranay, Junichi — any recommendations come to mind?

Pranay Vaddi: I have two recommendations. First is a report by the American Academy of Arts and Sciences authored by Tong Zhao and Dmitry Stefanovich. They write from Chinese and Russian perspectives about why those countries are concerned about US missile defense and its implications for strategic relationships. For those of us who focus on arms control policy and try to understand how the administration might approach hypothetical denuclearization talks with Russia and China, this is essential reading. Both authors are colleagues who have studied this area extensively. It’s not a book, so you can gain valuable insights from two genuine experts in just 50-60 pages.

The second recommendation is an older book from the late 1980s by Richard Betts called Nuclear Blackmail and Nuclear Balance. I found it particularly interesting as the Ukraine conflict progressed while I was still at the NSC. It reminds nuclear policy experts that nuclear weapons aren’t magical solutions. Attempts to use them in world crises — either to force adversaries to back down or to coerce for other military objectives, as in the Vietnam War — have produced mixed results at best. Understanding this history is important for our current era, where we’ve seen Russia issue almost daily nuclear threats in the context of a conventional war it initiated. Those are my two recommendations.

Junichi Fukuda: I’ve chosen two books. The first is another classic, Glenn Snyder’s Deterrence and Defense: Toward a Theory of National Security. This book is quite important because it addresses the problem of the stability-instability paradox, which directly applies to the current situation in the Pacific. If you want to study the stability-instability paradox, I recommend reading Graham Allison’s work on deterrence and defense.

The second recommendation is Brad Roberts’ The Case for US Nuclear Weapons in the 21st Century from 2015. This book is widely read in Japan as it directly discusses options for strengthening extended deterrence for allies in the Indo-Pacific. Many Japanese experts interested in this subject have read this book, so if you read it, you’ll have common ground for discussions with those experts in Japan. I highly recommend it.

Thanks again to the US-Japan Foundation for sponsoring this episode. It’s the first in a series about Japan-adjacent topics. I promise they won’t all be this dark.

Mood music:

Lennart Heim is the Compute Team Lead at RAND’s Technology and Security Policy Center and a Professor of Policy Analysis at RAND School of Public Policy.

China will likely match U.S. AI model capabilities this year, triggering inevitable concerns about America’s technological edge. However, this snapshot comparison misses the bigger picture. While Chinese models close the gap on benchmarks, the U.S. maintains an advantage in total compute capacity — owning far more, and more advanced, AI chips. This compute advantage, if leveraged strategically, will play an extraordinary role in driving economic transformation, securing technological leadership, and shaping the global AI ecosystem. U.S. policymakers risk squandering this edge by focusing on the wrong metrics and overreacting to predictable Chinese advancements.

TSMC's Secret Pipeline to China

Central to the U.S.-China AI competition are U.S. export controls that restrict China from importing advanced AI chips, acquiring semiconductor manufacturing equipment to build indigenous advanced AI chips, and using leading chip manufacturers such as Taiwan Semiconductor Manufacturing Company (TSMC). Despite these measures, a massive failure occurred in September 2024: TSMC, lacking basic due diligence, breached export restrictions by producing advanced AI chips for Huawei through a Chinese proxy company. This violation allowed Huawei to secure approximately 3 million chip dies using TSMC’s 7nm process, enabling the production of China’s best AI chips, the Huawei Ascend 910B and the upcoming 910C. This dwarfs even the highest estimates of smuggled chips, which typically involved tens of thousands of units, not millions. Although these chips trail the U.S. state of the art by about four years, they collectively provide China with computing power equivalent to approximately 1 million export-controlled Nvidia H100s (Nvidia's previous-generation chip from 2023) — substantial AI compute capacity that compensates for China's lack of indigenous production capabilities.

A Pattern of Failures, Not an Exception

The U.S. government responded to TSMC’s illicit production issue with a harsher rule in January 2025 and a likely investigation, but this incident is just one of many failures.

The 2022 chip controls contained specification errors that allowed Nvidia to slightly modify existing chips, creating the A800 and H800, which enabled DeepSeek's rise. Entity-listed Chinese companies built literal bridges to access advanced chipmaking equipment in unlisted facilities. Chinese firms stockpiled enough high-bandwidth memory (HBM) to meet their needs for the next few years, after the industry leaked the December 2024 restrictions in July. Until recently, AI chips with deployment-optimized performance, such as Nvidia's H20, continued to flow to China despite evidence that they enable new reasoning capabilities, with U.S. government officials taking months to act despite warnings. In every case: too little, too late.

What to Expect from Export Controls—And What Not to Expect

Despite these implementation failures, export controls remain a valuable strategic tool. While export controls do not create absolute barriers, they impose costs that have substantially slowed China's progress in AI and semiconductors, thereby maintaining America's lead.

But why is China still producing competitive models? It's harder for export controls to affect individual training runs than an entire ecosystem. Think of AI compute like factory equipment: Having fewer production lines doesn't prevent you from manufacturing a single product—you might still create that flagship product—but it severely limits your production capacity and market reach. Without the latest machinery, your production costs are higher and efficiency lower while competitors optimize their operations. What's more, the economics of scale work against you: while your competitor with ten factories can spread fixed costs across massive production volumes, driving down unit costs and enabling experimentation with new product lines, you're stuck with higher per-unit costs and limited ability to diversify. Similarly, when China has less total compute, they can still develop competitive models. After all, even in the U.S., companies only spent a fraction of their compute on training frontier models. But China misses out on the economies of scale that allow the U.S. to deploy AI broadly, experiment with more approaches, support a diverse ecosystem of leading AI companies, and continuously reinvest efficiency gains across their entire AI economy. Critically, this compute disadvantage also limits China's ability to project soft power globally—whether by providing AI services to international markets or by having the capacity to export advanced AI chips to other nations.

However, if current trends in AI training continue—shifting from two chips in 2012 to clusters of hundreds of thousands in 2025—export controls may eventually bite harder. China would need to build significantly larger clusters to compensate for less powerful chips: a cluster of 100,000 Nvidia B200s (a leading U.S. AI chip) might require a Chinese equivalent of 300,000 Ascend 910Cs (China's leading chip), resulting in higher energy consumption and greater engineering complexity for distributing AI workloads. Compounding this hardware gap is Nvidia's sophisticated software and networking ecosystem—crucial for orchestrating massive AI chip clusters and currently unmatched by Chinese alternatives.

10x More AI Workers: The Power of U.S. Compute Dominance

This focus on preventing model parity misses the fundamental question: What are the true metrics of AI leadership? We reacted to DeepSeek because it narrowed the gap between U.S. and Chinese model capabilities as measured by benchmark leaderboards and model rankings. However, the fixation misses the broader technological competition.

For example, think of AI systems as “virtual employees” that can perform cognitive labor, replacing remote colleagues you’ve never met in person. The number of AI employees an economy can deploy hinges directly on its compute resources. With roughly 10 times more compute capacity than China, the U.S. can field proportionally more AI employees across its economy, in sectors as diverse as drug discovery, logistics optimization, industrial robotics, and AI research itself. This compute advantage therefore compounds into a broader economic advantage over time. As AI systems drive productivity growth and innovation, they strengthen the underlying factors of national power. While China may catch up in individual model quality for now, the true metric of AI leadership might lie in deploying and integrating these systems at scale. In that sense, America’s true moat isn’t just better models—it’s the capacity to deploy and integrate AI in the economy at scale.

Subscribe now

Maintaining and Leveraging America's Compute Advantage

Both the Trump and Biden administrations deserve credit for recognizing early on that compute would be central to AI competition, taking key steps like restricting extreme ultraviolet (EUV) equipment exports to China in 2018 and banning AI chips in 2022—before ChatGPT captured public attention. But to stay ahead, the U.S. government must now build capacity to act quickly when reinforcing export controls while developing a clearer understanding of their actual impact. Rather than hoping to prevent temporary model parity, the U.S. strategy should leverage its substantial compute advantage to transform the broader economy and establish leadership in AI deployment worldwide. China may achieve competitive individual AI models this year, but this narrow benchmark gap is neither permanent nor strategically decisive. China will continue to tout breakthroughs to rattle U.S. confidence, but getting distracted and abandoning America's fundamental compute advantage would be a profound mistake.

Mood Music

How should America maximize its odds when competing with China? Can Trump’s approach to alliances succeed in strengthening deterrence? And what will it take for China to diplomatically capitalize?

To find out, ChinaTalk interviewed Rush Doshi, author of The Long Game: China’s Grand Strategy to Displace American Order. Rush served as deputy senior director for China and Taiwan on Biden’s NSC and is now at CFR. His new article with Kurt Campbell, entitled “Underestimating China: Why America Needs a New Strategy of Allied Scale to Offset Beijing’s Enduring Advantages,” I sincerely hope becomes a seminal document for American grand strategy in the 21st century. This is the most important show I’ve recorded all year.

Our conversation covers…

• Strategies for countering China’s strengths while accurately assessing weaknesses like demographics, debt, and slow growth,

• Historical lessons for US-China competition, from the newly-industrialized UK to Gorbachev’s USSR,

• Capacity-centric statecraft as an underrated type of international partnership,

• Persuasive versus coercive approaches to alliances,

• The evolution of China’s grand strategy and how Beijing’s diplomatic overconfidence in the face of tariffs could backfire.

Listen now on iTunes, Spotify, or your favorite podcast app.

America’s Odds

Jordan Schneider: Let’s start with net assessment, that is, understanding where the US is relative to China. Why is this important for effective strategy?

Rush Doshi: You have to compare yourself relative to your adversary or competitor in order to understand exactly what your strategy should be. Without good net assessment, you can’t really form a strategy. There used to be something called the DoD Office of Net Assessment. It’s since been eliminated.

Our assessment of China has swung widely over the last few years. Back in 2020, there was a perception that the US was in decline. We’d had January 6th, we weren’t handling the pandemic particularly well, and we had alienated our allies and partners. At the same time, China’s economy was booming. It seemed to be handling Covid well. In that moment, there was a perception that America was on its way out.

Just two years later, however, the zeitgeist changed completely. Russia invaded Ukraine in 2022 but achieved only limited success. The removal of zero Covid in China didn’t bring huge prosperity or reignite the economy. Later that year, America debuted ChatGPT to the world. It seemed like we had a rabbit we could pull out of our hat again — we could always rely on American innovation to sustain our lead.

These factors created a new zeitgeist — America ascendant. “America’s got this, and China can’t possibly keep up.”

But it’s difficult to strategize around such vastly differing conclusions about where we are relative to our competitor.

In our article, Kurt and I try to present a nuanced evaluation of our position relative to China.

Jordan Schneider: This is a fascinating case study of the past four years because it almost felt like these swings occurred faster than during the Cold War. Back then, they took longer and the triggering events were bigger than COVID policy. Events like the Vietnam War — America losing a conflict — would cause you to reassess relative trajectory, or decades-long realizations that the Soviet system was not delivering the technological and economic growth that people thought it might have been in the 30s, 40s, and 50s.

The big lesson is that it’s not necessarily a flows question — it is a stocks question. The US and China each have roughly a quarter of the world’s GDP and are likely going to continue having roughly a quarter of the world’s GDP each in the decades to come.

Rush Doshi: It’s important to note that two things can be true simultaneously. China could be slowing economically and have serious economic problems, but it’s also possible that it has strategic advantages and technological advantages in the timeframe that matters most for geopolitical competition.

Poor assessment simply concludes, “They are facing tough economic times, therefore, strategically they’re going to be on their way out.” It never ends up being that neat. The fundamental point of our piece is that you have to think about how particular advantages and disadvantages in one sector translate strategically to great power competition. This isn’t intuitive. It requires some degree of guesswork, but also careful analysis to understand correctly.

Jordan Schneider: There are cones of expectation for how well the US will do and how well China will do from a national power perspective. If you base your assessment on the third standard deviation performance of one country over another, you’re probably not picking the right strategy for the vast majority of futures. The way the US and China are discussed often defaults to either “this country has it figured out” while “that country is doomed,” or vice versa.

Rush Doshi: Thinking about it probabilistically is exactly the right approach.

Jordan Schneider: What do you think about the psychology of the American public and our leaders in Washington? Do you predict a preference for one direction or another, or is there appetite for a nuanced middle stance?

Rush Doshi: There’s a fundamental sense that things are changing — that China is formidable but also has a number of fragilities. The stakes of getting this question right are very high. This leads people to promote narratives they want to believe.

Sometimes there’s a market for saying, “Everything’s fine, don’t worry. American advantages are enduring, stable, real, and decisive.” On the other side, people might suggest things are in tough shape because they have a particular domestic policy agenda to advance. We saw that with the trade war. These assessments have political valence and salience.

Currently, China isn’t even a top priority for most American strategic thinkers. The focus continues to be on Russia-Ukraine or the Middle East. China lurks in the background rather than occupying the foreground where it deserves to be. That’s partly why there’s not always much attention on getting these assessments right.

It was important to have an office that did this kind of work in the US Government, and it would be monumental if that function was restored in the Trump administration. Hopefully, they make that decision soon.

From a government perspective, you almost never have anybody doing true net assessment, because the intelligence community isn’t authorized to consider American strengths and weaknesses relative to an adversary. They can talk about “red” (the adversary), but they won’t necessarily talk about “blue” (us). This causes complications.

Jordan Schneider: I agree that net assessment is important. It’s striking that so much of the commentary about the US relative to China comes from people who aren’t necessarily China experts. It seems that there’s more appetite to use China as a rhetorical talking point, on either the left or right, rather than to make accurate assessments about China.

Rush, you point to scale as the key metric for evaluating the US and China on a decadal or multi-decadal horizon. Can you explain your US versus UK analogy?

Rush Doshi: We only had 5,000 words for our piece, but a full net assessment should be much longer. Not all large countries achieve great power status. There’s a distinction between size and scale. Scale is the ability to translate size into meaningful outcomes. Scaling up is that process.

The Asian tigers had an incredible ability to scale on a small foundation. When you take those techniques and apply them to a foundation as large as China’s, the consequences are world-shaking.

The UK found this out the hard way. The UK had a first-mover advantage in the Industrial Revolution, and a small island in the northwest corner of Europe dominated the entire world for a long time — which geographically is quite remarkable. But that first-mover advantage wasn’t permanent. British industrial methods eventually transferred to Germany, the US, and Russia.

The British knew this was happening. Lord John Seeley wrote a book in 1883 about the rise of Great Britain that ended with a warning. He worried that just as Florence was surpassed by the great nation states of Europe, Great Britain would be surpassed by larger countries — the US, Germany (which had a much larger population than the UK), and Russia. He believed the British methods for industrializing and creating wealth could be transferred to others. When that occurred, they would have what he called “scale” — unimaginable scale relative to Great Britain — and Great Britain would decline.

During that period, these countries leveraged scale to push Britain out of key markets. They used their larger domestic economies to drive down marginal costs and outcompete in third-country markets. Today, that dynamic resembles the US and China.

As the US grew in scale and became more efficient and competitive, our industrial base exploded. By 1910, before World War I even happened, we were manufacturing four or five times as much steel as Great Britain. By World War II, our size and scale were even greater.

Hitler warned that the US was an unimaginable productive power. Yamamoto thought Japan could succeed in the first six months of war, but eventually American scale would defeat them. Even the Italians feared a contest of stamina would favor the United States.

That fundamental scale advantage we had shaped the thinking of our adversaries in the run-up to World War II and during the conflict itself. It wasn’t a hidden fact of politics — it was well understood by our rivals. Now that sense of daunting scale probably belongs to China.

Jordan Schneider: You make the point that even when Nazi Germany was at its peak from 1938 through 1941-42, the relative industrial weight of the Axis powers doesn’t hold a candle to what China can bring relative to the US today.

Rush Doshi: Hitler called the US a “giant state with unimaginable productive capacities.” In many ways, he was hoping to create that kind of productive scale for Germany, which obviously led to disastrous consequences. The fact of American manufacturing prowess was well established and understood around the world.

Jordan Schneider: For a couple of decades in the middle of the 20th century, America was building everything for everyone. Then globalization arrived, which was beneficial as more people gained wealth. We prevailed over Great Britain, Nazi Germany, and Japan, and later over the Soviet Union. But we’re in a very different industrial race today. Rush, let’s look at China’s statistics.

Rush Doshi: Every great power competitor the US has faced previously lacked the size and scale that China possesses. The Soviet economy was much smaller than the US economy, less productive, and its absolute manufacturing capability was inferior to America’s. The US was stronger than Germany and Japan combined in World War II. However, China represents the first competitor with true size and scale advantages against the United States.

Consider manufacturing shares. About 25 years ago, the US share might have been 30% while China’s was 6%. Within two decades, China’s share has quintupled to nearly 32%, while the US share has fallen by half to 15%. According to the UN, by 2030, China’s manufacturing share will be four times that of the United States — about 40% to our 11%. This represents a surprising turnaround in just 30 years.

The last comparable shift occurred between the US and UK. From 1870 to 1910, the British share of global manufacturing fell by 50% — the same proportion we’ve seen fall for the US, except it took us only 20 years rather than 40.

China’s current output is remarkable: twice American power generation, three times American car production, 13 times American steel production, 20 times American cement production, and approximately 200 times US shipbuilding capacity overall (though only three times our capacity for warships).

China produces about half the world’s chemicals and ships, 67-70% of the world’s electric vehicles, more than three-quarters of the world’s batteries, 80% of the world’s consumer drones, 90% of the world’s solar panels, and 90% of the world’s refined rare earths.

They’re also betting on the next industrial revolution. The US is installing many industrial robots, but China is installing seven times more. Half of all robot installations in 2023 happened in China. Regarding nuclear power for fueling the AI revolution — China leads in commercializing fourth-generation nuclear technology that we invented, and it’s planning 100 nuclear reactors in the next 20 years.

In science and technology, people claim China can’t innovate, but they exceed us in active patents and top-cited publications. You can question these numbers and argue they manipulate statistics. There’s some truth to statistical manipulation, but the trend line is incredible. They’re probably about even with us in those categories, and in 10 years, they’ll be well ahead of us, even accounting for statistical manipulation.

This manufacturing power translates into two forms of advantage. One is military advantage, as we discussed regarding World War II. The second is technological advantage — innovation from the factory floor, tacit knowledge, process knowledge, and special production capabilities that improve over time. This creates enduring advantages for China.

We should recognize this because it mirrors American innovation. When we were becoming a manufacturing power, we lagged behind Europe in science and technology. We didn’t have the Nobel Prize numbers that we have today. Manufacturing was the leading edge, and scientific recognition came later. We’re seeing the same story with China.

Jordan Schneider: Let’s not underestimate China — I’m with you this far, Rush. What’s the answer? One approach from the administration is that we need to reindustrialize. The problem is that even with perfect policies, you might gain only 0.5-2% per year in global manufacturing production. Setting reindustrialization and manufacturing growth as long-term goals aside, the trends you’ve outlined aren’t going to reverse anytime soon.

Rush Doshi: These built advantages are very sticky. Poured concrete is poured concrete — it exists. China’s supply chains won’t immediately relocate. They’re also more resilient than people think.

Some argue that China has macroeconomic and demographic problems, questioning why manufacturing statistics matter. But these problems are probably overstated. Consider their economy — it’s smaller than ours in nominal dollar terms, and many take comfort that it’s shrinking relative to the US economy. However, much of that is due to a strong dollar.

If you remove that financial advantage and look at purchasing power parity, China’s economy probably surpassed the US economy 10 years ago and is 30% larger today. Adjusting for purchasing power has limitations, but it’s a good attempt at capturing the local price of key strategic inputs — infrastructure, weapons, government personnel — which are priced in local terms. While nominal GDP might be important for quality of life, purchasing power adjusted GDP better captures factors that generate strategic advantage.

Regarding demographics, China is aging, but when will that matter? By 2100, their population will have fallen by half based on current trends. However, the strategic harm of aging won’t materialize for another 20 years or so. Between 2010 and 2020, the under-14 share of China’s population increased in absolute terms and as a percentage of the total population. The population is still aging, but the younger share grew due to a Mao-era baby boom — these are the grandchildren. Demographics are lumpy, which can buy time.

China’s dependency ratio won’t be as problematic as Japan’s current ratio until 2050, giving them time. Their investments in industrial robotics and embodied AI could help address labor problems from a declining population. Chinese factories like BYD plants already operate with few assembly line workers.

Regarding debt, China’s level is serious — it’s 300% of GDP when combining government, household, and corporate sectors. But that figure matches the United States. The composition of debt matters, as does who holds it, but this aggregate statistic provides a sense of overall indebtedness. If comparable, China might weather this challenge by properly recapitalizing local governments.

Finally, people note that American companies have high market capitalizations — the most valuable companies on earth are American with the highest profit shares in technology. But that market cap partially reflects investors wanting dollar assets and a strong dollar. Profits matter, but American companies maximize profits while Chinese companies maximize market share. They’re willing to operate at a loss to achieve market dominance. They’re optimizing for something different that might matter strategically — if they can remain solvent longer than our companies, they can put them out of business and deindustrialize us.

China has weaknesses — aging population, economic slowdown, potential deflation, high youth unemployment. All are important, but they may not matter as much in the timeframes that determine great power competition. That’s what we’re addressing with the question of scale.

Jordan Schneider: You’ve convinced me about the unlikelihood of second or third standard deviation downside outcomes for China. You’ve addressed many bearish arguments about China’s 20-year outlook, but not political instability. Can you discuss that? Xi will eventually die.

Rush Doshi: This is a critical point. My assessments focused on macro indicators, demographics, and financial factors because those are commonly cited when people claim China might never catch up to the US. Political leadership is challenging to evaluate.

If there’s a path to Chinese failure — an inability to succeed in the 21st century — it runs through a failed political transition from President Xi to his successor. Interestingly, Deng Xiaoping set up 20-plus years of succession by selecting Jiang Zemin and Hu Jintao, but he didn’t pick Xi. We lack that same long-term planning now.

China is living in the house that Deng Xiaoping built. Xi Jinping is changing the architecture, and we don’t know if these changes will cause the house to collapse or strengthen it against external pressures — to use terminology Xi often applies about strengthening the house against the elements.

China’s ability to maintain political stability is extremely difficult to assess. The Chinese Communist Party has endured the Great Leap Forward, Cultural Revolution, Tiananmen Square massacre, and globalization, remaining powerful throughout. However, elite instability at the top can threaten to bring down the entire party, as seen at various points in China’s history — such as when Deng Xiaoping was repeatedly imprisoned during the troubled transition from Mao to his successor.

We don’t know if Xi Jinping will select a successor or be too nervous to do so, given that he’s alienated vested interests, powerful families, and parochial groups that historically exercised power more freely than they do under his leadership.

This represents one potential path to Chinese problems. However, you can’t count on systemic failure or institutional collapse, especially when China has studied the Soviet collapse intently and tried to avoid those same mistakes. We must give them credit for considering paths to decay and attempting to close off as many as possible. They won’t get everything right, but that effort can’t be ignored.

Jordan Schneider: Even if you give Kyle Bass as much credit as possible — say a 50% chance of being right — that still leaves a 50% chance that you have an incredibly legitimate superpower breathing down your neck. It makes sense to buy the insurance plan to implement all the policies needed to live in a world where China actually succeeds.

Rush Doshi: That’s a great way of putting it. In our piece, we discuss “allied scale” as the alternative grand strategy for the United States. You’re right, it is an insurance plan. You can hope that China somehow gets in its own way — which wouldn’t be great for 1.4 billion Chinese people, but geopolitically, you could hope for that. However, betting on China’s collapse, governmental system change, or a botched leadership transition isn’t sensible. You need to buy insurance by investing in an American path to scale. Few great powers have achieved scale when confronted with a bigger rival, but the US can do it through its allies and partners.

Beyond Traditional Alliances

Jordan Schneider: Let’s talk about capacity-centric statecraft. What must Washington do?

Rush Doshi: Working more closely with allies sounds like a cliché. Kurt and I argue that we need to revolutionize how we work with allies by putting capacity building at the center — flowing both from the US to its allies and from allies to the US. If done right, the benefits are substantial.

When quantifying allied scale, the US with its allies represents approximately three times China’s nominal GDP, twice China’s purchasing power adjusted GDP, and more than twice China’s defense spending, even using high Chinese estimates that aren’t publicly available. We would have 1.5 to 2 times China’s share of manufacturing, and we would dominate in patents and top-cited publications.

China may be the top trading partner for 120-140 countries worldwide, but an allied consortium would be the top trading partner for virtually everyone except perhaps North Korea. This represents an incredible scale advantage that’s readily available.

The central task of American foreign policy in this era is transforming that theoretical advantage into reality — realizing that scale. Too often, we think of allies in hierarchical terms based on post-World War II and Cold War habits. We focus excessively on military issues and categorize alliances as tripwires, protectorates, vassals, or status markers. Our contention is that alliances must function as platforms for building capacity.

We could envision Japanese and Korean investment in American shipbuilding, where they have three to five times more productivity per worker. We could provide more technology to our allies; AUKUS exemplifies this approach. Kurt and I worked closely on launching AUKUS, which involves the US and UK helping Australia acquire nuclear submarine capability — our most sensitive military technology — to build meaningful Australian capacity in the Indo-Pacific.

We should consider novel joint military formations. The British and French created a joint brigade; the US and Japan or the US and South Korea could create anti-ship cruise missile battalions to train each other on the best shoot-and-scoot tactics, learning best practices from the Marine Corps under General Berger.

Allied scale also means allies helping each other, with the US facilitating collective action and connections. South Korea could help Europe rearm with its incredible defense industrial base. Scandinavian countries with excellent anti-ship cruise missiles could sell them to Southeast Asia. France, with expertise in LEU nuclear-powered submarines, could assist India, which has the same nuclear submarine program.

This vision of allied scale doesn’t always place the US at the center, but the US provides gravitational pull to facilitate collective action. And that’s just the security dimension — there’s much more to discuss on the economic and technological fronts.

Jordan Schneider: Let’s stay on the security side for a while. You were in government for the past four years trying to implement these ideas, which didn’t emerge from nowhere. The fact that the US plus its friends will be 2-3 times larger than China and its friends over the next few decades — assuming the balance of alliances remains stable — isn’t a new insight. As you and Kurt pushed for this approach with a president who was extraordinarily ally-focused, what roadblocks did you encounter from both US and partner perspectives?

Rush Doshi: Let me first address whether allied scale is intuitive. It is intuitive — we collectively know that if we don’t hang together, we’ll hang separately. What’s been missing from most analyses is the sense of urgency. People say working with allies is good — like apple pie and motherhood — but don’t explain why it’s essential.

It’s essential because we’re underestimating China’s scale. Great Britain failed as a great power because it couldn’t achieve scale. Lord Seely looked to unite with British colonies to create scale that could rival other great powers of that age, but those efforts were too little, too late, and everyone drifted apart. The question for the US is whether we’ll face the same fate.

We have an advantage: instead of an empire, we have allies who are independently capable. That’s a huge advantage compared to the UK. What Kurt and I are arguing in our piece is that this is urgent — we must get this right or cede the century to China. This isn’t the usual framing around allies. People typically see allies as beneficial, not as a four-alarm fire necessity. Part of the challenge in implementation is that people don’t perceive it as urgent, but rather as conventional common sense.

Jordan Schneider: To be clear, the US and China independently will likely run neck and neck — or too close for comfort — in the vast majority of future scenarios over the next few decades. Therefore, America’s most important strategy is to leverage the rest of the world, which is much more likely to partner with us than with a CCP-controlled PRC. That’s how you change the balance of power — the most straightforward way to dramatically shift the balance over the coming decades.

Rush Doshi: I’ll add that we may not be neck and neck in all metrics that matter for strategic value. We’re not neck and neck in manufacturing — we’ll be one-quarter their size. Comparing the US versus China on many metrics, we’re behind. We don’t have scale. We have several advantages: capital markets, immigration, innovation, talent, and general political stability. But we lack scale.

My concern is that without allies, the US won’t run neck and neck with China. The only way we can compete on metrics for great power competition is with allies. That’s why this is urgent. Without them, we’ll be like the UK was to the US and Russia, or like Florence was to the UK — one-quarter their size.

As productivity equalizes around the world — partly due to institutions and technology — China will remain productive. They already outmanufacture us and lead in many industries. The US urgently needs its own path to scale, which comes through allies.

The British couldn’t achieve scale because they failed to unite with their colonies. What differentiates the US is that we have capable, independent, sovereign allies who share our values — not colonies or subjects. If we join together with them, it’s not even a contest with China. We completely outscale them.

Jordan Schneider: This is a really important point. Modern-day Japan and Germany are not comparable to India circa 1890. We’re talking about very different allies in terms of their latent ability to contribute to the metropole’s national power.

Rush Doshi: Even in Lord Seely’s case, he excluded India, considering it a resource pit — obviously wrong about India’s potential. He focused on Canada, South Africa, Australia, and New Zealand — the white settler colonies. But compare Japan or Korea to any of them. The American alliance system contains incredible performers — powerful, technologically advanced, economically impressive nations who also make excellent military equipment and broadly share our values. That’s an extraordinary advantage in world politics. Almost no one else has this.

Don’t take my word for it — take China’s. China believes our single greatest asymmetric advantage is our allies. In 2017-2018, their delight at rising populism stemmed from one thing: the sense that populism was breaking American alliances apart. They’ve always seen alliances as our critical advantage, and they understand what we often don’t — that alliances bring economic and technological power to bear in competition.

The Biden administration had a theory for achieving scale with good and bad elements. The Trump administration also has a theory with strengths and weaknesses. Our piece isn’t partisan — we’re not advocating for one approach. What we’re saying is that the destination must be scale. There can be a path to allied scale with “Trumpian characteristics” — great, let’s pursue that. My concern is unilateralism that resolves bilateral issues without considering the larger strategic context vis-à-vis China.

The Trump administration has an opportunity to secure several bilateral deals worldwide. My hope is they’ll do so in a way that focuses on pooling our shared capacity and building scale against the PRC.

Jordan Schneider: The J.D. Vance perspective of promoting the AfD in Germany and suggesting that if the German right doesn’t win an election, they’re no longer our allies, represents a very insidious mindset. As you said, Rush, we forget how fortunate we are to have so many countries essentially aligned with our worldview.

Rush Doshi: We fought two world wars against Germany and one against Japan. These were the preeminent industrial powers of their regions in the 20th century and part of the 19th century. Today, they’re stalwart American allies. We have disagreements, but they’re on our side and share our values. That represents an incredible accomplishment and success story.

The idea that culture war issues should dictate our approach to these allies on the greatest geopolitical challenge America has ever faced is like cutting off your nose to spite your face. It doesn’t make sense.

That said, the Trump administration could still achieve scale. One critique of our piece is that it’s an idealistic dream that made sense five years ago but not now. I strongly disagree. There is a Trump path to scale, likely less focused on persuasion — which President Biden emphasized — and more on coercion. You could criticize the Biden administration for insufficient coercion.

Subscribe now

My point is this — you probably need to mix and match different tools depending on whom you’re dealing with. But we all know where we need to go. It cannot be the US versus China one-on-one. A unilateral America retreating to its own sphere would be poorer, less powerful, with deteriorating quality of life and diminished global influence, including on factors that make us prosperous. That’s obviously not a good path.

Jordan Schneider: Rush, you’ve outlined the two failure scenarios. The Trump administration would attribute one to the Biden administration — being too soft, allowing allies to avoid defense spending and ride our coattails into oblivion. I’ll let you elaborate on the contrasting errors.

Rush Doshi: One critique the Trump administration makes of the Biden administration is that it focused too much on persuasion and not enough on pushing allies — what you might call coercion. This resulted in free-riding. There’s a good counter-argument: progress occurred with AUKUS, the Quad, European involvement in Asia, and Asian involvement in Europe, which began moving toward allied scale.

However, the Trump administration’s critique that more needed to be done has merit. Kurt and I also believe more scale is immediately necessary. AUKUS was essentially a down payment on the kind of allied approach we need to institutionalize across the board. As someone deeply involved in AUKUS in 2021, working on negotiations for the White House under Kurt Campbell, I recognize it was novel then. We need to reach a point where such arrangements become routine.

Conversely, the Trump administration’s potential error would be excessive coercion — thinking you can bully everyone into compliance. Another error would be believing allies don’t matter, that America can go it alone, and that relations with Japan or Korea should focus solely on bilateral trade deficits and Treasury purchases.

Consider the economic picture: If we want to ensure scale for our industries when China is more competitive on a marginal cost basis, protecting only the American market isn’t enough. Without access to allied markets and allies’ access to our market, our companies cannot achieve scale relative to China. We need pooled market share.

The Trump administration has discussed erecting a tariff wall. In my view, a regulatory wall might be even better against certain Chinese exports. Beyond that, reducing trade barriers among allies would facilitate greater scaling up within the protected zone. Achieving this requires a mix of coercion and persuasion — combining Biden and Trump approaches.

Currently, President Trump has taken the coercive route with Liberation Day, sparking negotiations and deal-making opportunities. Hopefully, these deals will incorporate persuasive elements. Ideal deals would involve allies investing more while we reduce trade barriers — perhaps to zero-zero tariffs as some on Trump’s team have suggested. Allies would increase defense spending or work with us to improve their military capabilities. Additionally, coordinated steps would prevent China from dominating allied markets, including shared tariffs and regulatory policies to build a protective moat for all participants.

This kind of deal is achievable. As the Trump administration pushes these negotiations, I hope that’s their model. Some say they need to negotiate 75 deals, which seems excessive. I believe they need 8-10 key agreements. If they get those right, they can build allied scale differently from President Biden’s approach. They can claim victory and say they succeeded where others failed — that’s fine. The point is to achieve scale, and if they accomplish that, I’ll applaud them.

Jordan Schneider: Let’s stay on the Germany example for a second. The Biden administration was not able to get Germany to break its debt covenants and spend $1 trillion on defense. On the other hand, it basically took administration officials saying things so intense that Germany and Poland decided they can’t really trust extended deterrence anymore.

I don’t think the wake-up moment has necessarily come for Taiwan, Japan, and South Korea, but we seem to be in a very different timeline in the European context.

Rush Doshi: The risk of the current approach is that it’s so intense it pushes Germany away from being a core US ally on issues we care about. There’s a lot of flirtation in Europe right now with different forms of strategic autonomy, which could be interesting on their own. But if driven by dissatisfaction with the US, it might mean they’re not on our team when we need them.

Consider a concrete example: in 2020, the Europeans and Chinese negotiated the Comprehensive Agreement on Investment (CAI). This was fundamentally negotiated because Europe was concerned they couldn’t rely on the US. The Biden administration was upset — we had just taken office. That agreement eventually fell apart, but there’s a world where it doesn’t.

What worries me now is seeing Europe signal they want to hedge with China. That’s not sensible policy, but great powers can “commit suicide for fear of death” — a political science phrase I’m applying to this context. When you push an ally too aggressively, you can change their domestic politics to make it more anti-American. You can activate questions of face and respect that lead them to make poor decisions. Countries don’t always act in their rational self-interest, especially in times like these.

You can reasonably critique the Biden administration for not pushing allies hard enough, and the Trump administration for pushing too hard in ways that undermine transatlantic unity on China — which is extremely important for preserving America’s position. The truth is you need to do both. You need a nuanced approach, which either administration could achieve if they recognize the urgent need for scale and exercise more humility in American statecraft.

The capacity-centric statecraft we’re advocating is humble. We acknowledge there are capacities America has that its allies need, which we can provide. But we also recognize there are capacities America has lost that only our allies possess, and we need to regain them. How easy will it be to get Germany or Japan to manufacture in America if their domestic politics turns anti-American and their governments feel that accommodating the United States contradicts their political self-interest? That’s the risk — we might lose the transatlantic relationship we need if we’re not careful.

Jordan Schneider: In dark timelines, how does Trump 2.0 turn into America’s Gorbachev moment?

Rush Doshi: That was a provocative line in our piece. Personally, I’m hoping that doesn’t happen. Kurt and I intended to provide a strategic theory of how America should conduct itself with alliances — different from past approaches and valuable for the Trump administration. It’s not meant to be partisan.

The Gorbachev reference relates to two things.

1. There’s a risk that we neglect our allies’ agency, respect, face, and autonomy, causing them to move away from us if pushed too hard. Mikhail Gorbachev didn’t assume every Soviet republic would abandon his experiment, but that’s what happened. The same could happen to the American system if we renegotiate foundational elements, become unpredictable, interfere in allies’ domestic politics by supporting specific parties, or dismantle tacit bargains that underpin the American alliance system. This fragmentation would be catastrophic.

2. Successfully competing with China requires state capacity. Gorbachev’s reforms, and especially Yeltsin’s shock therapy, destroyed Soviet and then Russian state capacity, making it harder to accomplish geopolitical and domestic aims. I worry we’re dismantling our institutions. I’m not saying they shouldn’t be reformed or renewed — there were different approaches possible for both DOGE and the trade war. What concerns me is that these critically important initiatives weren’t implemented in the most strategic way.

Consider the trade war as a thought experiment. Imagine if, before Liberation Day, the Trump administration had presented a coherent theory of the case — policy papers explaining they would raise tariffs on China, with everyone else on notice. Imagine they had outlined what would happen to other countries in 90 days if they didn’t strike a grand bargain covering economics, trade, finance, security, and relations with China. What if they had already worked out preliminary deals with a few countries before the announcement, surprising China with this collective approach? And what if markets saw a logical grand design behind this strategy?

Would yields be spiking? Part of what’s happening is that markets don’t think we have a plan. The rationalization the Trump administration created after the fact about the trade war — imagine if it had come first. We’d be looking at a very different strategic situation.

You don’t go to war without preparation, as Russia demonstrated in Ukraine. That applies to global trade war also. What worries me is that there appears to be no document that articulates exactly what we hope to achieve globally, bilaterally, and multilaterally in the trading system. Without that framework, we’re improvising — and that’s frightening.

Jordan Schneider: As much as I want the Rush Doshi version of the Trump administration, I’m not optimistic that we’ll get it.

Rush Doshi: The purpose of my thought experiment is to imagine if we had done this differently. This could have been a moment of strategic advantage where the United States finally changed the global trading system to better reflect what’s needed regarding China’s participation. Instead, we have a unilateral, diminished American position with the entire world angry at us. I don’t believe that was the optimal approach.

You’re right — there were people with different proposals. My argument is that they should have made a different case, and I’m not certain anyone made this case. This is what they’re saying now — their post-hoc rationalization. Rewind the clock two months to the transition period. Had they presented this case then and informed the press — even if they ultimately imposed tariffs on allies — at least there would have been a coherent narrative before implementation. That story never emerged.

Jordan Schneider: Assuming the bear case from an American policymaking perspective for at least the next four years, we face several potential futures. One scenario involves democratic allies realizing they can’t count on the US, but needing to band together, put their houses in order, increase defense spending, and establish beneficial agreements between countries like South Korea and Germany. Let’s explore this path first. What might the global order look like with China checked out, while America and the rest of the developed world — uncomfortable with Chinese ascendancy — attempt to establish a new equilibrium?

Rush Doshi: China currently appears quite excited. President Xi is traversing different countries and striking deals. They clearly see an opportunity. I recently met with Chinese academics who indicated that China’s propaganda department is ecstatic, believing they haveways to position China as a better partner than the United States. Nobody should readily accept this narrative — skepticism is warranted. Nevertheless, China is actively attempting to drive wedges between the US and its allies, using trade as their primary mechanism.

My concern revolves around the bear case scenario. If we fail to reconstitute our alliances, experience greater transatlantic ruptures, and cannot establish meaningful agreements with our East Asian partners, America might retreat to the Western Hemisphere, attempting to make its stand there. This approach is doomed to failure. We cannot cede the entire world to China’s industrial power and expect that an America retreating westward can somehow escape the Mackinder trap. Mackinder warned against allowing any power to control the Eurasian “world island” — America’s pullback to the Western Hemisphere would enable precisely that outcome. This strategy was dismissed during World War II and makes even less sense now.

It’s particularly illogical in our globalized world where, even if the trading system fragments, it will inevitably reconstitute itself with another power at its core — potentially China. In that scenario, America becomes frozen out globally, unable to access the innovation, technological progress, and efficiencies developed elsewhere. America as the Galapagos Islands represents a bleak future. We must maintain our global presence and advocate for a world order reflecting American interests.

This isn’t merely “America First” rhetoric. For those embracing an America First perspective, this approach actually aligns with those principles by safeguarding American prosperity, not just geopolitical leadership. We’re discussing the quality of life Americans enjoy, which stems from constituting just 5% of the world’s population yet achieving extraordinary wealth — a reality made possible by a system designed to sustain this quality of life. We cannot simply demolish this system hoping everything will work out. That resembles Yeltsin’s shock therapy approach, which proved disastrous for the Soviet Union and Russia. My fear centers on an America that withdraws, creating a world that becomes China’s to lose — an outcome we must avoid.

Jordan Schneider: Do you believe the Trump team has learned lessons about alliance building from missed diplomatic opportunities during the first administration? What potential playbooks might China be implementing at this moment?

Rush Doshi: We should never discount China’s capacity to undermine itself diplomatically — essentially tripping over its own feet. Your observation is accurate, Jordan. There’s considerable confidence in Beijing. In my previous work on Chinese grand strategy, I’ve noted that the most important variable shaping their approach is their perception of American power. Beginning in 2016, this perception shifted dramatically. They concluded the world was experiencing unprecedented changes, leading to more aggressive and assertive Chinese policies, even toward countries they should have been cultivating.

One of history’s greatest strategic puzzles is China’s persistent alienation of India. China consistently pushes India toward the United States, when India might naturally prefer a more balanced position. China simply makes such neutrality impossible. They could certainly repeat these mistakes.

However, Jordan, we must acknowledge that our current approach differs fundamentally from anything attempted during the first Trump term. That administration conducted a trade war with China — arguably overdue — and attempted to renegotiate certain security agreements with allies. The current approach, however, essentially applies “Control-Alt-Delete” to the entire global trading system. We have a 90-day pause, but also a 10% blanket tariff — itself a revolutionary act. If we considered this tariff in isolation, without the chart the President unveiled, we would be shocked by it. Now, because our comparative reference is that chart, we’re virtually ignoring the importance of the 10% tariff.

To answer your question directly: American policy has become much more revolutionary than during the first Trump term, creating greater opportunities for China. The question becomes: even if they execute their strategy imperfectly, but marginally better than in the past, they could still accumulate substantial advantages. That prospect deeply concerns me.

Jordan Schneider: This connects to our earlier point about political instability. They might mishandle the opportunity, or they might not. Even granting them a 50% chance of failure still leaves several undesirable futures where smaller countries worldwide must navigate coordination problems without the United States providing necessary cohesion and facilitation. That’s frightening.

Rush Doshi: Precisely. A divided world where the US abandons its role as facilitator of collective action presents serious challenges. In political science, collective action represents one of the fundamental puzzles — what enables people to cooperate despite incentives to pursue individual interests? America functions as an anchor for collective action, making cooperation possible. If we withdraw, everything potentially fragments. Allies become divided and conquered. We might currently be in the “divide” phase of China’s “divide and conquer” strategy, but we’re inadvertently assisting them with the division — the conquest follows later.

Jordan Schneider: Every once in a while people ask me, “Jordan, why don’t you become the Matt Levine for China policy?” My answer is fundamentally that Matt Levine thinks securities fraud is funny, whereas I consider this stuff deadly serious. You too are someone who cannot look at the stakes of the conversation we’ve had for the past hour, shrug your shoulders, laugh about it, and simply move on with your life. Do you have any advice, Rush? It feels very helpless when the potential geopolitical errors seem larger than they’ve ever been in either of our lifetimes. Give me the pep talk. How do I keep putting one foot in front of another with the podcast?

Rush Doshi: There are a few things to keep in mind.

First, the Trump administration can negotiate agreements with allies and partners in ways that actually achieve scale. This remains conceivable. There are signs that some individuals running the negotiations are thinking in these terms. We know many people staffing the Trump administration at the mid-level, the key principals — not specifically the Cabinet, but many people below that level — believe in allied scale. They may not have called it that, but they understand its logic. That group exists, and there remains the possibility of landing the ship in that spot. I cannot predict the odds, and I do worry, but we shouldn’t be despondent.

Part of what must happen is continued discussion of these issues. One reason Kurt and I wrote our piece was to argue why allied scale matters. I hope as that argument circulates and others reinforce it, it may help shape some of the Trump administration’s thinking. The piece wasn’t meant as a partisan attack but as a strategic framework any administration could adopt.

The last thing I’d note is that this administration won’t last forever. Congress might flip in two years. America constantly changes. For allies and partners, they will find ways to work with the United States — whether through the legislative branch, pockets in the executive branch, or a suddenly enlightened executive that negotiates better deals. It’s not over.

Great power competitions can be lost in short periods, but China faces many challenges, too. They often struggle to consolidate and press their advantages. America possesses real advantages we shouldn’t neglect: two oceans, abundant resources, 70% of the world’s capital markets, the ability to attract the best talent from around the planet, and general political stability. These represent huge advantages that won’t disappear overnight. Some might be damaged or tarnished, but others might even strengthen.

We must remember this competition won’t resolve itself in two or four years — we must make investments for the long haul. Congress can do much. Everything the Biden administration did to build allied scale received bipartisan support, funding, endorsement, and praise from Congress. We must remember Congress wields power in this domain as well.

Jordan Schneider: One last question, Rush. I remember DMing you during the administration saying, “They’re not going to let you talk about this book, right?” And you replied, “Yeah, sorry,” but here we are now. We’ll have you back to do a deeper dive and an updated discussion about it. I’m curious — assuming nearly everyone in our audience has read the book — how did your experience being on the inside, reading intelligence, and interacting with the Chinese government affect some of the conclusions you reached toward the end of your research for that project?

Rush Doshi: Thank you, Jordan. I’m really proud of “The Long Game.” It was unfortunate I couldn’t discuss it while in government, and I’m glad people found it useful. The book was based on an assessment of 5 million words of Chinese Communist Party material, as well as a rigorous social scientific approach to their behavior. The conclusion was that China has maintained, since the end of the Cold War, a grand strategy to displace American order — first regionally, then globally. My time in government reinforced my view that these conclusions were correct.

In the book, I argue that going back to the 1980s, the US and China were essentially quasi-allies against the Soviet Union. Everything changed after the Gulf War, the Soviet collapse, and the Tiananmen Square massacre. Suddenly China viewed America as the biggest threat and inaugurated a new policy — “hiding capabilities, biding time” (韬光养晦 tāoguāng-yǎnghuì) — what I call blunting American power quietly, not assertively, while benefiting from America’s system.

In military terms, they pursued anti-access/area denial approaches to keep us out. Economically, they sought Most Favored Nation (MFN) and Permanent Normal Trade Relations (PNTR) status to tie our hands. Politically, they wanted to stall American-led institutions in Asia lest they become platforms for challenging China. This approach worked quite well until the global financial crisis.

In Central Committee materials, we see China change its perception of American power, adopting a new strategy formulated by the Central Committee. China should “actively accomplish something.” This approach had military, economic, and political components focused on building order within Asia, not just blunting American order. Militarily, China invested in power projection capabilities to influence its neighbors. Economically, we saw the Belt and Road Initiative and efforts to use economic statecraft against others. Politically, China built international institutions meant to serve as the foundation for order-building within Asia.

I wrote that this worked until 2016, when China’s assessment changed again. They adopted a new phrase to guide Chinese policy: “great changes unseen in a century.” The idea was that China’s current opportunities and risks were unlike anything they’d faced in a hundred years. We witnessed the inauguration of a global Chinese grand strategy focused on global military bases and winning in Taiwan; dominating supply chains economically and making the world more dependent on China (what President Xi calls “dual circulation"); and technologically, leading the fourth industrial revolution to ensure China wins those technologies not just for prosperity but for power. Politically, it aimed to change global institutions to make them more conducive to autocracy.

All of this is motivated by a desire to “rejuvenate” China, which represents a 100-year goal. The Chinese Communist Party is fundamentally a nationalist party seeking rejuvenation. They’ve always been that way. They’re also a Leninist party wanting to centralize power in pursuit of that goal. That’s why I say China has a grand strategy — it has the concepts, capability, and conduct to pursue that vision. They don’t always execute perfectly and make mistakes, but there exists a strategic intention to create a partial hegemony over part of the world, reflected in those military, economic, and technological indices I mentioned.

The intelligence community had previously outlined these conclusions in threat assessments but hadn’t substantiated them using open sources. I’m more confident of these conclusions now than when I wrote the book. Looking at President Xi’s behavior and the Chinese Communist Party’s actions over the past five years sustains the argument that they are pursuing a global grand strategy — not always well or brilliantly executed, but nonetheless deadly serious.

Jordan Schneider: Let’s look toward the future. Rush, what’s the broader open source project you’re launching at CFR?

Rush Doshi: I’m continuing my research at Georgetown as a professor and running the China program at the Council on Foreign Relations. One of our most exciting initiatives builds on my approach in “The Long Game,” where I relied on Chinese texts and materials to make my arguments. Now we’re going to mass acquire, digitize, and translate those kinds of texts at scale, making them more available to the public so people can essentially read China in its own voice.

If you do that, China essentially tells you where it wants to go. The Chinese Communist Party must communicate with itself and its cadres, and we want to make those communications available to others. This is something the US government used to do from FDR in 1941 through 2013 under Barack Obama, when we had the Foreign Broadcast Information Service that translated foreign material. That resource largely stopped being available to scholars in 2013. Our hope is to rebuild it at scale with a far greater source base and with the help of artificial intelligence. We’ve made major investments in this effort, have access to a large cache of material, and will share more very soon.

Jordan Schneider: It has been pretty frustrating over the past few months to see China be so important rhetorically to every policy decision while genuine curiosity about understanding the country seems at a relative low point — both regarding net assessment and strategic intention. It’s on us to make this material accessible and interesting. We can’t just complain about it; we must do the work and present it effectively, even if we need to be persistent. We need to make a strong case that this is both important and engaging.

Rush Doshi: Well that’s what you do on this podcast!

Jordan: Thank you! And look, The Long Game sold well, so there’s an appetite for this content somewhere! I fully believe in the mission and have more faith in you than virtually anyone else to make Chinese strategic discussion something the world cares about and takes seriously. Godspeed, Rush!

Mood Music

Double-edged swords in the US-China Cold War

A guest post by the excellent Kyle Chan of the High Capacity substack.

The US and China are in a cold war, not a trade war. This is something much bigger than tariffs and trade deficits. It is much bigger than Taiwan or semiconductors. And it began long before Trump or Xi. The US and China are locked in a global contest of power that is playing out along every dimension: economic, technological, military, cyber, soft power, global prestige. Both sides are searching for any tool, any weapon, any piece of leverage they can use against the other—short of direct military action.

There is no such thing as escalation dominance. Trump thinks the US will win in a trade war because China sells more to the US than the other way around. A tit-for-tat escalation on tariffs means the US will always be able to tariff more Chinese goods than vice versa. Adam Posen has recently argued it’s actually China that has “escalation dominance” (a RAND concept in nuclear deterrence) because China has other ways of escalating beyond tariffs, including potentially denying Americans access to Chinese-made goods from smartphones to medicines. However, the reality is neither side has escalation dominance because both sides have already gone far beyond trade measures. If you’re looking at the full range of actions beyond trade tools, there’s virtually no limit to how far each side can go.

The US and China are posturing as if they have escalation dominance, which makes the problem worse. US Treasury Secretary Scott Bessent said on CNBC that China had made a “big mistake” in retaliating against Trump’s tariffs because China was “playing with a pair of twos.” China’s Ministry of Commerce has said that China would “fight to the end.” While there are already signs that Trump is backing down, the confidence that each side feels—or at least tries to project—only fuels a downward spiral of recklessness and emotion-driven bravado.