, ChinaTalk’s chief biotech correspondent, will be at JPM next week and would love to meet up. Respond to this email to connect.
You could cover AI for ChinaTalk full time this year courtesy of The Tarbell Center for AI Journalism. Apply for the fellowship here — applications are due February 28th.
Lastly, next week we’re diving deep into China and IP coming up with Adam Mossoff, a professor of law at GMU. A sneak peek:
What Xi Believes
Lizzi C. Lee is a fellow on Chinese Economy at Asia Society Policy Institute’s Center for China Analysis and the host of Wall St TV.
Reading the full text of Xi Jinping’s February 2023 speech — published on the last day of 2024, almost two years later, in Qiushi 求是— isn’t how I expected to welcome the new year. It’s dense, grandiose, and filled with Marxist-Leninist jargon that demands a machete to cut through. This speech is vintage Xi — equal parts ideological lecture, historical justification, and political directive. It’s also the clearest articulation we’ve had of his vision for “Chinese-style modernization” (中国式现代化). And, as always with Xi, there’s no shortage of sharp critiques, bold ambitions, and ominous warnings.
Modernization, Xi-Style
Xi’s pitch is simple: modernization doesn’t have to follow the Western script. In fact, it shouldn’t. He blames Western modernization for prioritizing capital over people, resulting in runaway inequality, entrenched social divisions, and political instability. For Xi, this isn’t just a bug in the system — it’s the system’s defining feature.
Chinese modernization, he says, is different. It’s “people-centered,” aiming for a balance between material wealth and spiritual well-being. It’s a model designed to suit China’s history, culture, and governance system — not a one-size-fits-all approach borrowed from the West. And Xi doesn’t stop there: he frames it as a beacon for other developing nations, a way out of the “middle-income trap” that has snared so many others.
This might resonate with some, especially in a world where the Western model has lost some of its sheen. The problem lies in the art of the word balance. Xi’s model relies on the very market mechanisms he criticizes. For all his railing against capital, the Chinese economy still leans heavily on private enterprises, foreign investment, and global trade. Xi wants to avoid the pitfalls of capitalism while using its tools to fuel growth. If anything, his tendency to clamp down too much risks suffocating the economic dynamism he seeks to safeguard, embedding even more vulnerabilities into the system. That balancing act — rejecting Western modernity while borrowing liberally from it — is harder to achieve in practice.
The Party: Soulkeeper and Savior
If there’s one thing Xi makes clear, it’s that the Party is non-negotiable. Without the CCP, he says, Chinese modernization would “lose its soul” (丧失灵魂). That’s heavy stuff, and it underlines how deeply he ties the Party’s leadership to the nation’s success — or failure. Party leadership isn’t just important — it’s existential. Xi warns that losing the Party’s grip would spell doom for the entire project.
Interestingly, Xi seems clear-eyed about the challenges within his own ranks. He talks about the “deep-rooted problems” (深层次问题) that still plague the Party — issues that, if left unchecked, could stage a comeback. His warning is blunt: any lapse in discipline could let old problems “resurface like embers reigniting” (死灰复燃). This seems to reflect the constant tension within the Party to maintain control, enforce discipline, and keep its sprawling apparatus from falling into complacency.
Xi’s disdain for procrastination is striking. “See risks early, act quickly, make decisive calls,” (见事早、行动快，当断则断、当机立断) he demands. And the language here is truly explosive (pun intended!): “Act decisively when action is needed; make bold, swift decisions. Don’t let small problems grow into big ones, or big problems explode” (当断则断、当机立断，不能让小事拖大、大事拖炸). The message is clear: inaction is dangerous, and delay is unforgivable. It’s the kind of directive that might inspire action — or strike terror — depending on where you sit in the Party hierarchy.
Lean Into the Struggle
If there’s one word that defines Xi’s speech, it’s struggle (斗争). He frames it as the CCP’s defining trait, a “political gene” forged through a century of adversity.
Struggle isn’t just a strategy for Xi — it’s practically a moral imperative. He frames it as the CCP’s key to its past victories and future survival. “Weakness” and “retreat,” he argues, lead only to decline. It’s a stark, almost combative philosophy, underpinned by his conviction that China’s path is righteous and its rise inevitable.
When it comes to cultivating young leaders, Xi is equally unsentimental. He wants cadres forged in the fire of practice and struggle. His metaphor of choice? “Let cadres, especially young ones, learn to swim by swimming” (在游泳中学会游泳). The best way to spot capable leaders, he suggests, is to see who thrives in “severe and complex struggles” (严峻复杂的斗争).
But this obsession with struggle reveals as much about Xi’s insecurities as it does about his ambitions. [JS: he must see lazy cadres all the time whose hearts aren’t in the fight.] China’s rise, he acknowledges, is fraught with risks: economic pressures, geopolitical tensions, and internal dissent. Xi’s solution is to double down on the fight, whether it’s against foreign “containment,” domestic inefficiencies, or ideological wavering within the Party.
And here’s the irony: a system that constantly defines itself through struggle risks becoming trapped in a self-perpetuating cycle of conflict. Xi’s insistence on vigilance — his warning to cadres to “act decisively” and prevent “small risks from escalating into major crises” — sounds less like confidence and more like paranoia. It’s as if he’s bracing for a storm that never arrives but always looms on the horizon.
The West as the Convenient Villain
Xi’s critique of the West is one of the speech’s sharpest elements — and also one of its most revealing. He accuses Western modernization of being inherently exploitative, built on colonialism, inequality, and capital-driven greed. But Xi goes further, taking aim at what he calls the “myth” that modernization equals Westernization.
By positioning Chinese modernization as a viable and increasingly appealing alternative, he not only defends China’s path but also offers it as a model for other nations — though, as Xi claims, China won’t force it on others.
Of course, the focus of this messaging is other developing countries. In fact, Xi portrays China’s model not just as an alternative but as an improvement to the Western system, which he accuses of failing those who tried to copy it. It’s a not-so-subtle attempt to redefine modernization itself — and to shift the narrative away from Western dominance.
And while Xi rails against Western-style inequality, China’s own wealth gap remains an uncomfortable reality. The promise of “common prosperity” has been toned down in recent years, given economic malaise and the seemingly more urgent need to revive the animal spirits of the business community. Yet another reminder of how fast China can pivot (and how long the time delay of the Qiushi article is!).
Can Xi Deliver?
Xi’s speech is a declaration of ideological intent. It frames Chinese modernization as a project of historic significance, tied to the Party’s legitimacy, China’s rise, and even global civilization itself. It’s ambitious, audacious, and, yes, a little overwhelming. My main issue, though, is whether his vision is genuinely achievable. The message boils down to: “Do everything, perfectly, all at once.” It’s an impossible standard that Xi holds himself to. He wants to combine market efficiency with socialist equity, preserve environmental sustainability while driving industrial growth, and project confidence while guarding against constant threats. It’s a lot to ask of any system, let alone one as complex and unwieldy as China’s.
Some might say that the competing priorities and contradictions may not be a bug but a feature. Like many Party documents, it attempts to cover every conceivable issue with sweeping mandates, but it leaves room for both ideological purity and pragmatic flexibility to adapt to changing circumstances. But Xi goes a step further. What is clear to me is that Xi isn’t just trying to reshape China. He’s trying to reshape the very idea of modernity itself.
Back to the topic of ideology and practicality, I do believe, though, compared with Xi’s predecessors, he is less about paying lip service to ideological purity and is actually a man of deep conviction in such beliefs — but that’s a topic for another day.
ChinaTalk is a reader-supported publication. To receive new posts and support our work, consider becoming a free or paid subscriber.
Does America need a Manhattan Project for AI? Will espionage make export controls obsolete? How can the U.S. foster an open innovation ecosystem without bleeding too much intellectual property?
To discuss, ChinaTalk interviewed Lennart Heim, an information scientist at RAND, and Chris Miller, author of Chip War.
We get into…
The growing importance of inference scaling and what it means for export controls,
Regulatory oversights that have allowed China to narrow the gap in AI capabilities,
China’s best options for keeping up in a low-compute environment,
Methods to secure model weights and associated tradeoffs,
Partnerships in the Middle East and the tension between export controls and economies of scale,
Whether autocracies are better suited for facilitating AI diffusion.
Jordan Schneider: Let’s start with algorithms. Lennart, what’s your take on what DeepSeek’s models do and don’t mean for US-China AI competition, and maybe more broadly, what scaling on inference means for export controls?
Lennart Heim: To some degree, many were taken by surprise because, basically, what we’ve seen this year is that the gap between U.S. and Chinese AI models has been narrowing slightly. Of course, this always depends on how you measure it. Benchmarks are the best way to assess this, though whether they’re the most accurate success metric is a separate discussion we can have later.
It’s wrong to conclude that export controls aren’t working just because DeepSeek has developed a model that’s as good or nearly as good as OpenAI’s. That conclusion would be mistaken for two reasons.
First, export controls on AI chips were only implemented in 2022, and the initial parameters were flawed. Nvidia responded by creating the H800 and H800, chips that were just as effective as the restricted U.S. chips. This oversight cost us a year until the rules were updated in October 2023. DeepSeek, meanwhile, reportedly acquired 20,000 A100 chips before export controls were tightened and may have also obtained a number of H800s. These are still powerful chips, even if they’re not the latest. Because they have such a large stockpile, they’ll remain competitive for the foreseeable future.
Second, export controls don’t immediately stop the training of advanced AI systems. Instead, they influence the long-term availability of AI chips. For now, if someone needs sufficient chips to train a competitive AI system, they can likely still access them. However, as scaling continues, future systems may require millions of chips for pre-training, potentially widening the gap again.
For current systems, which use around 20,000 chips, we shouldn’t expect immediate impacts from export controls, especially given their short duration so far. The real question might be whether these controls affect the deployment phase. If Chinese users start extensively interacting with these systems, do they have enough compute resources to deploy them at scale? That remains to be seen.
Jordan Schneider: Let’s underline this. Inference scaling has the potential to exponentially increase compute demands. Can you explain why?
Lennart Heim: Yes. Once we have a well-trained system — developed with a significant number of chips — we can enhance its efficiency with techniques like reinforcement learning. This improves the system’s reasoning capabilities.
What do we mean by reasoning? Essentially, the model generates more outputs or tokens, which demands more compute power. For instance, if a model previously responded to queries in one second, it might now take 10 seconds. During that time, GPUs are processing data, and no other users can access those resources. This increased compute time per user significantly impacts overall resource requirements.
Not everyone has the necessary compute resources to handle these demands, and that’s a major factor. If DeepSeek or others open-source their models, we could gain better insights into the total compute resources required.
Chris Miller: Would you say this reflects a shift in the rationale for export controls? In our interview two years ago, we were thinking about AI progress primarily in terms of model training. Now, inference is an important driver of progress too. What does this imply for calibrating export controls going forward?
Lennart Heim: It’s a new paradigm, but not one that replaces the old approach — they coexist. We’ve seen trends like chain-of-thought reasoning, where models are asked to think step by step, and larger models tend to perform better at it.
I expect this pattern to continue. Bigger models may achieve better reasoning, though there could be a ceiling somewhere. In the semiconductor industry, transistors got smaller over time, but different techniques emerged to achieve that goal. We observe overarching trends with multiple enabling paradigms in AI also.
I don’t think this complexity fundamentally challenges export controls. As long as pre-training remains important and deployment depends on compute resources, export controls still matter.
If new architectures emerge that don’t rely heavily on compute, that would be a bigger challenge. Because if compute is no longer the main determinant of capabilities, current export controls become ineffective. But I think many, many parameters need to change for that to happen. Regulators can reasses over time.
Jordan Schneider: Let’s emphasize that point about compute. If models, after training, take significantly longer to produce answers — whether it’s three minutes, 10 minutes, or even an hour — that extended “thinking time” consumes compute resources. Compared to older systems that responded in seconds, this shift means nations will need far more compute capacity to achieve productivity, national defense, or other goals. For now, inference scaling makes the case for export controls even stronger. Compute prowess will be key for any government wanting to excel in AI technology.
Lennart Heim: Exactly. This also means that the distinction between training and deployment will become increasingly fuzzy over time. We already use existing trained models to create synthetic data and give feedback to train new systems. Early AI systems like AlphaGo and AlphaZero employed an element of self-play, where the model played against itself. That is training and deployment occurring simultaneously.
We’re likely to see similar trends with large language models and AI agents. This makes it harder to maintain clear-cut categories, and compute efficiency will play an even larger role.
As AI capabilities improve, they’ll require fewer resources to achieve the same benchmarks. A model might now need 100 GPU hours for a task that once took 500 GPU hours. This efficiency is part of the broader technological trend.
It’s hard to frame a national security conversation around specific capabilities, because any given capability becomes easier to access over time. That is the reality that policymakers need to deal with.
Chris Miller: Is it generally true that the contours of the national security argument around export controls are shifting, given the focus on inference infrastructure and test-time compute? If it was all about training, regulators could say “We don’t want them to train this type of AI application.” But if it’s actually about whether there’s infrastructure to run a model that can produce a million different use cases, it becomes more about productivity and less about discrete national security use cases.
Lennart Heim: It depends. The reasoning behind export controls has evolved. In 2022, export control discussions didn’t really mention frontier AI. By 2023, they began addressing it, and this year’s revised export controls took it a step further.
For example, the revised controls now include high-bandwidth memory units, which are key for AI chips. Why is this significant? HBM is especially important for deployment rather than training. Training workloads are generally less memory-intensive compared to deployment, where attention mechanisms and similar processes require more memory.
Banning HBM, and thereby limiting companies like Huawei from equipping AI systems with HBM, likely has a greater impact on deployment than training. However, I don’t think this motivation is explicitly stated in the documents.
Jordan Schneider: To draw a parallel from semiconductor history, there was a big debate about RISC versus CISC architectures back in the ’80s and ’90s. Pat Gelsinger pushed x86 as the dominant architecture, arguing that software wouldn’t need to be super efficient because hardware would continue improving exponentially. Essentially, Moore’s Law would clean up inefficiencies in code.
Fast forward to today, and it seems like there are enough AI engineers finding creative ways to use compute that algorithmic innovations will expand to match the available compute. Engineers at places like Anthropic, DeepMind, and OpenAI are the first to play with these resources. Would you agree this is the trend we should expect?
Chris Miller: Yes, that sounds about right. If compute is available, we’ll find ways to use it. An economist might ask, “What’s the marginal benefit of an additional unit of compute?” Ideally, we want the most algorithmic bang for our buck with each unit of compute.
In the last few years, we’ve seen GPU shortages in certain market segments, indicating strong economic output from every GPU deployed — or at least that’s the assumption behind the investments. It’s uncertain whether this trend will persists in the long-term.
The trajectory of Moore’s Law has historically been steady, but estimating improvements in algorithmic efficiency is much harder. There’s no reason to believe these improvements will follow a linear or predictable trend, so I’d remain agnostic about where we’re heading.
Lennart Heim: Even as compute efficiency improves, there’s still the question of how these breakthroughs are discovered. Are they serendipitous — like having a great idea in the shower — or do they emerge from systematically running hundreds of experiments?
Often, breakthroughs in compute efficiency come from large-scale experimental setups. Sometimes, these ideas are even inspired by the models themselves, like using a GPT model to develop the next iteration.
Leading AI companies have an ongoing internal competition for compute resources. With AI becoming commercialized, the competition intensifies because allocating more compute for research means less is available for deployment.
I’d be curious to see load graphs for these companies. Are experiments run at night when fewer people use ChatGPT? These are the types of strategies companies likely adopt when managing limited resources.
Jordan Schneider: To Chris’s point, as long as these systems are profitable and there’s value in increasing their intelligence, demand for them will persist. The smarter the systems, the more value they provide across sectors.
Looking at China, if export controls are working and TSMC can’t produce unlimited chips for Huawei, leaving the Chinese AI and cloud ecosystem with one-third of the capacity needed, what does that mean for research directions engineers in the PRC might take?
Chris Miller: Two things stood out to me this year.
First, rental prices for GPUs in China were reportedly lower than in the U.S., which is surprising in a supposedly GPU-constrained environment. This could suggest either that China isn’t actually GPU-constrained or that there’s lower demand than expected.
Second, Chinese big tech firms — ByteDance excluded — haven’t shown the same steady increase in capital expenditures on AI infrastructure as U.S. firms like Google or Microsoft. Charting capex trends from ChatGPT’s launch to now, Alibaba, Tencent, and Baidu don’t display the same commitment to scaling AI infrastructure.
Why might this be?
Fear of chatbots saying something politically sensitive about Xi Jinping.
Doubts about market demand for their products.
Lingering caution from the 2019–2020 regulatory crackdown, making massive investments seem unwise.
But there does seem to be a striking difference between how Chinese big tech firms are responding to AI relative to U.S. big tech firms. I wonder what that tells us more generally about compute demand in China going forward.
Lennart Heim: China’s venture capital ecosystem is quite different from the US. America’s sprawling VC system provides the risk capital needed to explore bold ideas, like building billion-dollar data centers or reactivating nuclear power plants.
Jordan Schneider: Exactly. In China, there’s less capital available for speculative investments. Investing tens of billions of dollars into cloud infrastructure for training AI models isn’t immediately profitable, so Chinese tech firms hesitate to do it. We recently translated twointerviews with DeepSeek’s CEO that explain this with more detail.
There have been large, loss-leading investments in hardware-heavy sectors of the economy, but not many software-focused investments.DeepSeek, by operating more like a research organization and less like an Alibaba-style traditional tech firm, has taken a longer-term approach. It’s unclear whether smaller incumbents with sufficient capital can continue innovating or if progress will depend on stolen algorithmic IP.
Lennart, what’s your perspective on securing model weights and algorithmic IP as we head into 2025?
Lennart Heim: A lack of compute usually means fewer algorithmic insights, which causes the ecosystem to slow down. But stealing model weights is a shortcut. I’m referencing RAND’s recent report, Securing Model Weights, on this question.
Training a system may require tens of thousands of GPUs, but the result is just a file, a few gigabytes or terabytes in size. If someone accesses that file, they reduce or even bypass the need for GPUs entirely.
New ideas are compute multipliers. Publication causes widespread diffusion of these multipliers, which we have seen with transformer architecture, for example.
But this changed about two years ago. In the name of security, OpenAI, DeepMind, and Anthropic no longer publish many detailed architecture papers. OpenAI hasn’t released their model architectures since GPT-4.
If you want to know what the architecture looks like, you have to go to the right parties in San Francisco and talk to the right people. Which is exactly the problem. You could walk out of these parties with huge compute efficiency multipliers.
These companies still mostly have normal corporate environments. But if we see AI as core to national security, these AI labs need to be secure against state actors. These companies will eventually need help from the U.S. government, but they also need to step up on their own. Because this IP leakage completely undermines American export controls.
Chris Miller: How do we know we’re striking the right balance between securing important IP and fostering the free exchanges of ideas that drive technological progress and economic growth? What’s the metric for assessing whether we’ve achieved that balance?
Lennart Heim: Right now, we’re living in the worst of both worlds. OpenAI and DeepMind aren’t going around sharing their research openly with other researchers, publishing on arXiv, or presenting at conferences like NeurIPS or ICML. They’re not diffusing knowledge widely to benefit everyone.
At the same time, their proprietary information is still vulnerable to hacking. So, instead of fostering diffusion within the U.S. ecosystem, we inadvertently enable adversaries or bad actors that are willing to use illicit measures to access this information. This is the worst-case scenario.
Clearly an open ecosystem is beneficial in many ways. That’s why some companies still open-source model weights and infrastructure — it helps push the entire U.S. ecosystem forward.
Assessing the ideal policy balance is hugely complex. There are many reports discussing the trade-offs of open-sourcing versus safeguarding. For now, though, it’s clear that we’re in a bad place — keeping knowledge from U.S. researchers while leaving it vulnerable to theft.
Jordan Schneider: Let me offer a counterargument. Developing algorithmic innovations for frontier AI models isn’t something that happens on an assembly line. The places that succeed most at this have cultivated a unique research culture and can attract top talent from around the world. That includes talent from China, which produces a huge share of advanced AI research and talent.
A highly classified, “skunkworks”-style approach could create two major downsides.
From a talent perspective, it becomes harder to attract people with diverse backgrounds if they need security clearances to access cutting-edge research.
Research in highly classified settings tends to be compartmentalized and siloed. In contrast, the open, collaborative environments in leading labs foster innovation by allowing researchers to share insights, compare experiments, and optimize resources.
Imposing rigid barriers could hinder internal collaboration within firms, making it harder for researchers to learn from each other or gain equitable access to resources like compute.
The Manhattan Project succeeded by isolating talent in the desert until they developed the atomic bomb. That’s not a model we can apply to OpenAI, Anthropic, or other AI labs. The internal openness that has allowed Western labs to thrive could be stifled by the kind of restrictions you’re suggesting.
Lennart Heim: Absolutely. I’m not arguing that security comes without costs. It’s important to consider where to put the walls. We already have some walls in the AI ecosystem — we call them companies. We could achieve a lot by strengthening those existing walls while maintaining openness within organizations.
If someone eventually decides that research must happen in fully classified environments, then of course that would slow down innovation.
For now, though, many measures could enhance security at relatively low cost while preserving research speed. The RAND report referenced earlier outlines the costs and methods of different security levels. Some security measures don’t come at any efficiency cost. Just starting with low-hanging fruit — measures that are inexpensive yet effective — could go a long way.
Jordan Schneider: I have two ideas on this front.
First, if we believe in the future of export controls and assume the U.S. and its allies will maintain significantly more compute capacity than China, it could be worthwhile for the labs or the National Science Foundation to incentivize research in areas where the U.S. is more likely to have sustainable advantages compared to China going forward.
Second, banking on these security measures seems like a poor long-term strategy for maintaining an edge in emerging tech. I mean, think about Salt Typhoon. The Chinese government has been able to the intercept phone calls of anyone they want at almost no cost. Yes, it’s possible to make eavesdropping harder, but I’m not sure any organization can secure all their secrets from China indefinitely.
Chris Miller: That raises the question of how to think about algorithmic improvements. Are they like recipes that can be easily copied? Or are they deeply embedded in tacit knowledge, making them hard to replicate even if you have the blueprints?
I’m not sure what the answer is, but replicability seems key to assessing how far to go with security measures.
Lennart Heim: You can draw an interesting connection here to the semiconductor industry. We’re all familiar with cases of intellectual property theft from ASML, the Dutch company building the most advanced machines for chip manufacturing. However, it’s not enough to simply steal the blueprints.
There’s a lot of tacit knowledge involved. For instance, when someone joins a semiconductor company, they learn from experienced technicians who show them how to use the machines. They go through test runs, refining their processes over time. This knowledge transfer isn’t written down — it’s learned by doing.
While this principle applies strongly to the semiconductor industry, it may be less relevant to AI because the field is much younger.
Recently, I’ve been thinking about whether there are still low-hanging fruits and new paradigms to explore. Pre-training has been scaled up significantly over time, and it’s becoming harder to find new ideas in that area. However, test-time compute is an emerging paradigm, and it might be easier to uncover insights there.
I expect academics, DeepSeek, and others to explore this paradigm, finding new algorithmic insights over the next year that will allow us to squeeze more capabilities out of AI models. Over time, progress might slow, but we could sustain it by increasing investment. That’s still an open empirical question.
Jordan Schneider: On that note, Lennart, what does it really mean to be compute-constrained?
Lennart Heim: I’ve been thinking about this more from the perspective of export controls. There’s often an expectation that once export controls are imposed, the targeted country will immediately lose the ability to train advanced AI models. That’s not quite accurate.
To evaluate the impact of export controls, it’s useful to consider both quantity and quality.
The quality argument revolves around cutting off access to advanced manufacturing equipment, like ASML’s extreme ultraviolet lithography machines. Without them, a country can’t produce the most advanced AI chips. For instance, while TSMC is working with 3-nanometer chips, an adversary might be stuck at 7 nanometers.
This results in weaker chips with fewer FLOPS (floating-point operations per second). Due to the exponential nature of technological improvement, the performance gap is often 4x, 5x, or 6x, rather than a simple 10% difference. Export controls exacerbate this gap over time.
The quantity argument is equally significant. Chip smuggling still happens, but access to large volumes of chips is much harder due to export controls and restrictions on semiconductor manufacturing equipment.
Being compute-constrained impacts the entire ecosystem. With fewer chips, fewer experiments can be run, leading to fewer insights. It also means fewer users can be supported. For example, instead of deploying a model to 10 million users, you might only support 1 million.
This has a cascading effect. Fewer users mean less data for training and less revenue from deployment. Lower revenue reduces the ability to invest in chips for the next training cycle, perpetuating the constraint.
Additionally, AI models increasingly assist engineers in conducting AI research and development. If I have 10x more compute than my competitor, I essentially have 10x more AI agents — or “employees” — working for me. This underscores how compute constraints can hobble an entire ecosystem.
Chris Miller: That makes a lot of sense. My theory about the Chinese government’s response — and Jordan, let me know if this resonates — is that they seem less concerned with consumer applications of AI and more focused on using AI as a productive force.
Their strategy appears to prioritize robotics and industrial AI over consumer-facing applications. The hope is that limited compute resources, when deployed toward productive uses, will yield the desired returns.
The problem with this approach is that much of the learning from AI systems comes from consumer applications and enterprise solutions. Without a full ecosystem, their progress will likely be stunted. It’s like trying to balance on a one-legged stool.
Jordan Schneider: Chris, that’s an interesting observation. It’s a reasonable strategy for a country facing resource constraints, but it also highlights the limitations of being compute-constrained.
Chris Miller: Exactly. There’s also a political dimension to consider. In addition to being compute-constrained, China has spent the past five years cracking down on its leading tech companies. This has dampened their willingness to invest in consumer-facing AI products.
After all, a successful product could draw political scrutiny, which isn’t a safe place to be. That dynamic further limits the development of a robust AI ecosystem.
Lennart Heim: That’s a great point. The revenue you generate often determines the size of your next model. OpenAI’s success, for instance, has attracted venture capital and fueled further progress.
China’s state subsidies can offset some of this reliance on revenue. They can fund projects even without immediate returns, challenging the flywheel effect I described earlier.
Still, there are many less compute-intensive AI applications, like AI agents, that are being developed worldwide. These don’t require the same level of resources but still factor into national security concerns.
The key question is, what are we most worried about? For AGI or highly advanced AI agents, compute constraints will likely be a major factor. But China might already be leading in domains like computer vision.
The ideal balance between compute intensity and emerging risks remains an open empirical question. We’ll need to monitor how these dynamics evolve over time.
Middle East Expansions and Cloud Governance
Chris Miller: Another obvious implication of being compute-constrained is that you can’t export computing infrastructure. Perhaps that’s a good segue to discussing the Middle East.
Lennart Heim: Part of the compute constraint story, as you mentioned, is that if you need chips to meet domestic demand, you can’t export a surplus. If the PRC is barely able to meet its own internal demand — assuming that’s true, though we don’t have solid evidence yet — it’s clear that the U.S. and its allies are producing significantly more AI chips. This allows them to export chips, but there’s an ongoing debate about where and how these chips should be exported.
Existing export controls already cover countries like China, Russia, and others in the Country Group D5 category. However, there are also countries in Group D4, like the UAE and Saudi Arabia, which require export licenses for AI chips. These countries are increasingly ambitious in AI, and since early this year, the U.S. government has been grappling with whether and under what conditions to allow chips to be exported to them.
Export licenses offer flexibility. They can come with conditions — such as requiring buyers to adhere to specific guidelines — before granting access to AI chips. There’s clearly demand for these chips, and this debate will likely continue into the next year, as policies and the incoming administration determine where the line should be drawn.
Chris Miller: It’s been publicly reported that upcoming U.S. regulations might involve using American cloud providers or data center operators as gatekeepers for AI chip access in these countries. This approach would essentially make private companies the enforcers of usage guidelines.
Lennart Heim: That’s an intriguing approach. It creates a win-win scenario: these countries get access to AI chips, but under the supervision of U.S. cloud providers like Microsoft, which can monitor and safeguard their use.
It’s important to understand that export controls for AI chips differ from those for physical weapons. A weapon is controlled by whoever possesses it, but AI chips can be used remotely from anywhere. If a country faces compute constraints due to export controls, one solution is to use cloud computing services in other countries or build data centers abroad under shell companies.
Most AI engineers never see the physical clusters they use to train their systems. The data centers are simply wherever electricity and infrastructure are available. This makes it challenging to track chip usage.
There are three layers to this challenge:
Where are the chips? This is the most basic question.
Who is using the chips? Even if you know where they are, it’s hard to determine who is accessing them.
What are they doing with the chips? Even if you know who is using them, you can’t always control or monitor the models they train.
U.S. cloud providers can help address the second layer by verifying customers through “know your customer” regimes. I’ve written about this in a paper titled Governing Through the Cloud during my time at the Centre for the Governance of AI. Cloud providers can track large-scale chip usage and ensure compliance, making them far more reliable gatekeepers than companies in the Middle East.
Chris Miller: There’s broad agreement that no one should be allowed to build an AI cluster for nefarious purposes. But regulations seem to be taking this further, aiming to ensure long-term dominance of U.S. and allied infrastructure.
The idea is to not only set rules today but maintain the ability to enforce them in the future. This makes some people uncomfortable because it positions the U.S. and its allies as the long-term power brokers of AI infrastructure, potentially limiting the autonomy of other countries.
Lennart Heim: That’s a fair criticism, but I would frame it more positively. This is about ensuring responsible AI development. Depending on your geopolitical perspective, some might view it as the U.S. imposing its values, while others see it as necessary for safety and accountability.
Exporting chips isn’t the only option. Countries can be given access to cloud computing services instead. For example, if someone insists on acquiring physical chips, you could ask why they can’t simply use remote cloud services. But many countries want sovereign AI capabilities, with data center protection laws and other safeguards.
The ultimate goal should be to diffuse not only AI chips but also the infrastructure and practices for responsible AI development.
Jordan Schneider: This reminds me of a recent story that struck me. The American Battlefield Trust is opposing the construction of data centers near Manassas, a Civil War battlefield. It’s a tough dilemma — I want those data centers, but preserving historical sites is important too.
Intel’s Future and TSMC Troubles
Jordan Schneider: Speaking of sovereignty in AI, let’s discuss Intel. There’s been a lot of speculation about Pat Gelsinger’s departure and the board’s decision to prioritize product over foundry. Chris, what’s your take on this news?
Chris Miller: It’s a significant development and signals a major strategy shift for Intel, though the exact nature of that shift remains unclear.
There are several possible paths going forward:
Intel could sell some of its design businesses and double down on being a foundry.
It could do the opposite and focus on design while stepping back from foundry ambitions.
It might just try to muddle through, continuing its current strategy until its next-generation manufacturing process proves itself.
None of these options are ideal compared to where expectations were two years ago.
Intel will present a tough challenge for the incoming administration. The company has already received $6–8 billion through the CHIPS Act to build expensive manufacturing capacity, but there’s no guarantee it will succeed. Going forward, Intel will likely require significant capital from both the private and public sectors.
Jordan Schneider: This ties back to the fundamental pitch that Pat Gelsinger made during the CHIPS Act discussions — that America should have leading-edge foundry capacity within its borders.
This is a global industry, and the world would face severe consequences if Taiwan were invaded, regardless of U.S. manufacturing capacity. Taiwan is nominally an ally, and TSMC’s leadership should know better than to antagonize the U.S. government by selling leading-edge chips to Huawei, because Washington is the ultimate guarantor of Taiwan’s current status.
That said, it would certainly be preferable for the U.S. to have Intel emerge as a viable second supplier or even the best global foundry. But how much are you willing to pay for that? Even if you allocate another $50 billion or $100 billion, can you overcome the cultural and structural issues within Intel?
There’s no denying the enormous business challenges involved. Competing in this space has driven many companies out of the market over the past 20 years because it’s simply too hard. Chris, do you want to put a dollar figure on how much you’d be willing to raise taxes to fund a US-owned leading-edge foundry?
Chris Miller: You’re right — it’s not just about money. But it is partly about money because funding gives Intel the time it needs to demonstrate whether its processes will work.
Whether Intel succeeds or fails, it’s clear that if they only have 12 months, they’ll fail. They need 24 to 36 months to prove their capabilities. Money buys them that time.
The other variable is TSMC, which already has its first Arizona plant in early production. Public reports indicate that the yields at this plant are comparable to those in Taiwan, which is impressive given the negative publicity surrounding the Arizona plant in recent years.
TSMC is building a second plant in Arizona and has promised a third. If these efforts succeed, the need for an alternative US-based foundry diminishes because TSMC is effectively becoming that alternative.
The big question is how many GPUs for AI are being manufactured in Arizona. TSMC has publicly stated that 99% of the world’s AI accelerators are produced by them, and currently, that production is confined to Taiwan. Expanding this capability to Arizona would be a game-changer.
Lennart Heim: There has been public reporting that Nvidia plans to produce in the U.S. in the near future, which would be a positive development. But the broader question extends beyond logic dies. What about high-bandwidth memory? What about packaging? Where will those components be produced?
The strategic question is whether the U.S. should carve out a complete domestic supply chain for AI chips. Is it a strategic priority to have every part of the process — HBM, packaging, and more — onshore, or are we content with relying on external suppliers for certain elements?
Chris Miller: Intel has received commitments through the CHIPS Act, but those funds are contingent on them building new manufacturing capacity. The new leadership team at Intel might decide not to proceed with some of those plans.
This raises a critical question — if Intel doesn’t build those facilities, what happens to the allocated CHIPS Act funding? It’s important to note that Intel hasn’t received all the money; they’ve been promised financial assistance if they meet specific milestones.
This decision will likely land on the desk of the next administration, and they’ll need to assess whether additional private and public capital is necessary to ensure Intel’s competitiveness.
Jordan Schneider: Early on, policymakers should evaluate the trade-offs clearly. If we give $25 billion to America’s foundry, it might result in a 30% chance of competing with TSMC on a one-to-one basis by 2028. At $50 billion, maybe it’s a 40% chance. At $75 billion, perhaps it rises to 60%.
Even with massive investment, there’s a ceiling on how competitive Intel can become. The rationale for the initial $52 billion in CHIPS Act funding was compelling, but that was spread across many initiatives — not just frontier chip manufacturing.
For Intel to achieve parity with TSMC by 2028, you’d need to show how increased investment could meaningfully improve the odds. This is a challenge for Intel’s next CEO, the next commerce secretary, and whoever oversees the CHIPS Act moving forward.
Time is critical, and if Intel can’t make it work, we’re left relying on TSMC. That brings us back to the awkwardness of TSMC producing a significant number of chips for Huawei. We need to dive deeper into that story.
Lennart Heim: Great segue. The Huawei story broke a couple of months ago, and it highlights the challenges of enforcing export controls.
The basic premise of export controls is to prevent Chinese entities from producing advanced AI chips at foundries like TSMC. There are two main rules:
If you’re on the Entity List, like Huawei, you can’t access TSMC’s advanced nodes.
AI chips cannot be produced above certain performance thresholds.
TechInsights conducted a teardown of the Huawei Ascend 910B and found it was likely produced at TSMC’s 7-nanometer node. This violates both rules — the Entity List restriction and the AI chip performance threshold.
Shell companies and similar tactics make compliance tricky, but based on the available information, this should have been detected.
What’s even more concerning is TSMC’s role in this. If TSMC is producing an AI chip with a die size of 600 square millimeters — massive compared to smartphone chips — they should have raised red flags.
Any engineer can tell the difference here. There are probably structural issues at TSMC where the legal compliance team doesn't talk to the engineers.
But on the other side is the design teams. It's not like you send them something and then you stop talking. This is a co-design process. There was clearly ongoing communication on these kinds of things. But then they produced the logic dies for the Ascend 910B, although it’s still an open question whether all of these chips were produced at TSMC.
But TSMC’s involvement definitely undermines export controls. A good story you can spin is that this is a sign of some production issues happening at SMIC such that Huawei is still relying on TSMC. Definitely more insights are required here.
Intelligence Failures and Government Follow-on 本末倒置
Jordan Schneider: Speaking of shell companies, what was the U.S. intelligence community doing? The fact that this information had to come from TechInsights is mind-boggling. I can’t imagine there are many higher priorities than understanding where Huawei is manufacturing its chips. For this to break through TechInsights and Reuters feels like an absurd sequence of events. It highlights a glaring gap in what the U.S. is doing to understand this space.
Lennart Heim: We’ve seen this before, like when Huawei’s advanced phone surfaced during Raimondo’s visit to China. There’s clearly more that needs to be done. The intelligence community plays a role, but think tanks, nonprofits, and even private individuals can contribute to filling this gap.
For example, open-source research can be incredibly powerful. People can use Google Maps to identify fabs or check Chinese eBay for listings of H100 chips. There’s a lot you can do with the resources available, and nonprofits can play a critical role in providing this type of information.
The gap in knowledge here is larger than I initially expected, and there’s a lot of room for improvement.
Chris Miller: This also points to the broader challenges of collecting intelligence on economic and technological issues, which the U.S. has historically struggled with.
It’s also worth asking what information the Chinese Ministry of State Security (MSS) is gathering about technological advances in other countries? What conclusions are they drawing? If we’re struggling with this, I wonder what kind of semiconductor and AI-related briefings are landing on Xi Jinping’s desk. Do those briefings align with reality, or are they equally flawed?
Jordan Schneider: It sounds like the solution is just to fund ChinaTalk!
On the topic of MSS, the Center for Security and Emerging Technology (CSET) did some reports of China’s systems for monitoring foreign science (2021) and open-source intelligence operations (2022). But when you read Department of Justice indictments against people caught doing this work, it often seems amateurish and quota-driven.
I don’t have a clearance and can’t say for sure, but it makes me wonder — if the U.S. is struggling to figure out Huawei’s chips, maybe the Chinese are equally bad at uncovering OpenAI’s secrets. This might reflect bureaucratic challenges on both sides, such as bad incentives, underfunded talent pools, and difficulty competing with the private sector.
Subscribe to support ChinaTalk’s independent research.
Lennart Heim: That’s true. But there’s a broader issue here. I come from a technical background — electrical engineering — and transitioning into the policy world has been eye-opening.
One thing I’ve realized is that there are often no “adults in the room,” especially on highly technical issues. In domains like AI and semiconductors, there simply aren’t enough people who deeply understand the technology.
Getting these experts into government is a huge challenge because public-sector salaries can’t compete with private-sector ones. It’s not about hiring international experts — it’s about bringing in people who know these technologies inside and out. They need to be aware of the technical nuances to even track these developments properly.
For example, I’ve used this as an interview question — if you were China, how would you circumvent export controls? Surprisingly few people mention cloud computing. Most assume physical products and locations matter most, but it’s really about compute. It doesn’t matter if the H100 chip is in Europe — you care about how it’s used.
These types of insights require a technical mindset, and we need more of these brains in D.C. — in think tanks, government, and the IC. We’re still in the early days of implementing export controls, and the more technical expertise we bring in, the better we’ll get.
Many of the hiccups we’ve seen so far can be traced back to a lack of technical knowledge and capacity to address these issues effectively.
Jordan Schneider: We need to diffuse technical talent into government, but we also need to diffuse AI into the broader economy. Lennart, how should that happen?
Lennart Heim: Diffusion is a big topic. Earlier, we touched on where data centers should be built — Microsoft expanding abroad is one form of diffusion. Another aspect involves balancing protection, such as export controls, with promotion. These two strategies should go hand in hand.
Diffusing AI has several benefits. It can be good for the world and can also counter the development of alternative AI ecosystems, like those in the PRC. From a national security perspective, it’s better to have American-led AI chips, data centers, and technologies spread globally.
That raises an important question — as the gap between AI models narrows, with China catching up and smaller models improving, are models really the key differentiator anymore? From a diffusion standpoint, what should we focus on if models aren’t the most “sticky” element?
Take GitHub as an example. It previously used OpenAI’s Codex to help users write code but recently switched to Anthropic’s Claude. This shows how easily models can be replaced with a simple API switch. Even Microsoft acknowledges this flexibility, and it’s clear that models may not provide the long-term competitive edge we assume.
If models aren’t the differentiator, what is sticky? What should we aim to diffuse, and how should we go about it?
Chris Miller: The interesting question is which business models will prove to be sticky. Twenty years ago, I wouldn’t have guessed that we’d end up with just three global cloud providers dominant outside of China and parts of Southeast Asia.
Those business models have extraordinary stickiness due to economies of scale. The question now is — what will be the AI equivalent of that? Where will the deep moats and large economies of scale emerge?
These are the assets you want in your country, not in others. They provide enduring influence and advantages. While we don’t yet know how AI will be monetized, this is a space worth watching closely.
Lennart Heim: That’s a great point. It also ties into the idea of building on existing infrastructure. Take Microsoft Word — it’s incredibly sticky. Whether you love it or hate it, most organizations rarely switch away from it.
For example, the British government debated moving away from Microsoft Office for years. The fact that this debate even exists shows how difficult it is to dislodge these systems.
Maybe the stickiness lies in integrating AI into tools like Word, with features like Copilot calling different models. Or perhaps it’s in development infrastructure.
We’ve focused a lot on protecting AI technology, but we haven’t thought enough about promoting and diffusing it. This includes identifying sticky infrastructure and understanding how to win the AI ecosystem, not just by building the best-performing models but by embedding AI into tools and workflows.
Chris Miller: This brings us back to the Middle East and the tension between export controls and economies of scale. If economies of scale are crucial, you want your companies to expand globally as soon as possible.
That raises a question: does this mean relaxing export controls on infrastructure, or do you maintain strict control? Balancing the need for control with the benefits of scaling up globally is a delicate but important challenge.
Lennart Heim: What about smartphones? AI integration into smartphones seems like a big deal. For example, Apple has started using OpenAI models for some tasks but is also developing its own. At some point, I expect Apple to ditch external models entirely.
Interestingly, Apple is also moving away from Nvidia for certain AI tasks, developing its own AI systems instead. With millions of MacBooks and iPhones in users’ hands, Apple could quickly scale its AI.
This shift toward consumer applications — beyond chatbots — will define the next phase of AI. We’ll see if these applications prove genuinely useful. For now, feedback on Apple’s recent AI updates has been underwhelming, but that could change next year.
If Apple’s approach takes off, could it define who wins in AI?
Jordan Schneider: Let me take this from a different angle. AI matters because it drives productivity growth, and that’s what we should be optimizing for.
I trust that companies like Apple, Nvidia, and OpenAI will continue improving models and hardware. My concern is that regulatory barriers will block people from reaping the productivity benefits.
For example, teacher unions might resist AI in classrooms, or doctors might oppose AI in operating rooms. Every technological revolution has brought workplace displacement, but history shows that these changes leave humanity better off in the long run — more productive and satisfied.
The next few years will see political and economic fights between new entrants trying to deploy AI and labor forces pushing back, especially through regulation. These battles will determine how AI transforms industries.
Chris Miller: Agreed. Beyond the firm-versus-labor dynamic, there’s also a competition between incumbent firms and new entrants. This varies by industry but is equally important.
Then there’s the question of which political system — ours or China’s — is better suited to harness innovation rather than obstruct it. You could make arguments for either.
Jordan Schneider: Take Trump, for example. On one hand, he’s concerned about inflation and unemployment but also supports policies like opposing port automation.
Ultimately, I don’t think Trump himself will play a huge role in these decisions. Instead, it’s the diffuse network of organizations — standard-setting bodies, school boards, and others — that will shape the regulatory landscape. Culture also matters here. Discussions about AI’s risks — like safety concerns and job loss — have made it seem more frightening than it should.
These risks are real, but they need to be balanced against the benefits of technological progress. Right now, the negative cultural conversation about AI could influence these diffusion debates.
Xi Jinping might be even more worried about unemployment than Trump. But some parts of China’s non-state-owned economy are probably more willing to experiment and adapt new workflows.
The U.S. may be too comfortable to navigate the disruptions needed to fully harness AI’s potential. This complacency could slow progress compared to China’s willingness to experiment aggressively.
Chris, what do you think about a Manhattan Project for AI?
Chris Miller: The term “Manhattan Project” for AI isn’t quite right. The Manhattan Project was secretive, time-limited, and narrowly focused. What we need for AI is long-term diffusion across society.
The better analogy is the decades-long technological race with the Soviet Union, marked by broad R&D investments, aligned incentives, and breaking barriers to innovation. This kind of sustained effort is what we need for AI.
Lennart Heim: That requires projects that focus on onshoring more fabs and data centers — like CHIPS Act 2.0. It also requires energy and permitting reform.
Compute is key, and building more data centers is a good starting point, but we also need to secure what we build. This includes data centers, model weights, and algorithmic insights. If we’re investing in these capabilities, we can’t let them be easily stolen. Innovation and security must go hand in hand.
Jordan Schneider: One thing I’d add is the importance of immigration reform. The Manhattan Project had over 40% foreign-born scientists. If we want to replicate that success, we need to attract the world’s best talent.
This is a low-cost, high-impact solution to drive growth, smarter models, better data centers, and more productivity. It’s crucial to have the best minds working in the U.S. for American companies.
Lennart Heim: Absolutely. Many of the top researchers in existing AI labs are foreign-born. Speaking personally as a recent immigrant, I’d love to contribute to this effort. If we’re doing this, let’s do it right.
Reading Recommendations
Jordan Schneider: Let’s close with some holiday reading recommendations. Lennart, what was your favorite report of the year?
Lennart Heim: Sam Winter-Levy at Carnegie just published a report calledThe AI Export Dilemma: Three Competing Visions for U.S. Strategy. It touches on many of the topics we discussed, like how we should approach diffusion, export controls, and swing countries. It has some good ideas.
Jordan Schneider: I’d like to recommend The Gunpowder Age: China, Military Innovation, and the Rise of the West in World History by Tonio Andrade. We’ll be doing a show on it in Q1 2025. It’s an incredibly fun book and addresses a real deficit in Chinese military history. The author dives deep into Chinese sources and frames the Great Divergence through the lens of gunpowder, cannons, and guns.
He uses fascinating case studies, like battles between the Ming and Qing against the Portuguese, British, and Russians, to benchmark China’s scientific innovation during the Industrial Revolution.
The book argues — similar to Yasheng Huang’s perspective from our epictwo-hour summer podcast — that the divergence between China and the West happened much later than commonly believed. Into the 1500s and 1600s, China was still on par with the West in military innovation, including boat-building, cannon-making, and gun-making.
The writing is full of flair, which is rare in historical works. It’s military history, technology, and China vs. the rest of the world — all my sweet spots in one book.
What’s your recommendation, Chris?
Chris Miller: For some more deep history, I recommend A Brief History of Intelligence: Evolution, AI, and the Five Breakthroughs That Made Our Brains by Max Bennett.
It’s a history of brains and how they’ve evolved over millions of years, starting with the first neurons. The author is an AI expert who became fascinated by the evolution of intelligence and ended up becoming a neuroscience expert in the process.
The book is extraordinary — more fun than I expected — and thought-provoking in how it explores the history of thinking across all kinds of beings, including humans.
ChinaTalk is a reader-supported publication. To receive new posts and support our work, consider becoming a free or paid subscriber.
I spent ten days in Norway this summer. What follows are reflections from my time there on Oslo, the Vikings, and WWII.
Oslo Vibes
“This place isn’t perfect Jordan,” a civil servant told me, “please tell me you won’t make that your angle.” I then asked him what the worst neighborhood in Oslo is, walked there, and felt it was nicer than half of Manhattan.
The first few days of 19 hours of sunlight in 72-degree weather were an unparalleled endorphin rush, but by day six I felt a little strung out.
Servicepeople regardless of your race start conversations in Norwegian so as to not make immigrants feel unwelcome.
I played some pickup sand volleyball in one of the thousand Oslo parks with a Kurdish culture affinity club. No-one on my team could tell me how to say “nice serve” in Kurdish but when some Kendrick came on their speakers, they all sang along to “certified Loverboy, certified pedophile.”
Chinese EV showrooms dotted Oslo, with Nio taking plum position on the main street right outside parliament. The salesman there said vibes are mostly good, though every few weeks someone walks in just to say “we don’t like Chinese cars here.” The XPENG 小鹏 saleswoman unprompted told me, “we are Chinese but a private company not owned by the government like BYD. Also, Volkswagen owns 5 percent and Norwegian oil fund owns some of us too.”
Norway up until the 70s was one of the biggest Israel supporters. Their two Labor parties both ran their countries for decades, and living on a kibbutz was a thing Norwegian lefties did. But Norwegian soldiers saw some shit as peacekeepers in Lebanon in the 80s, everyone got really invested in the Oslo Peace Process and felt burned by the Israelis in the subsequent decades. “We were a colonized country too, you know. First the Danes then the Swedes…”
Thanks presumably to oil wealth guilt, Norway might be the country most into ESG. The government in early June officially recognized Palestine but Parliament decisively voted down a push to make the Oil Fund divest from all companies with ties to Israel. They did recently sell $70m of Caterpillar stock…? The ratio of pride to Palestinian flags was maybe 5:1.
Haaretz recently ran a feature on rising antisemitism in Norway which convinced me I didn’t want to move there. For an illustrative excerpt on what happened when a group of Jews tried to join an International Women’s Day protest to raise awareness of Hamas. They got approval to join, and on parade day this happened:
The hostile reaction manifested almost immediately. Initially, the group was refused entry to the event and had to prove that they had the organizers' authorization to participate. "One of the organizers went on shouting and cursing, and then took one of our signs and threw it on the ground," Nilsen recalls. "After the police made sure he couldn't get close to us, more and more organizers told us that our message conflicted with the messages of the event.
"They looked at us with hatred and disgust and started to shout that we were Zionists and fascists. Then the crowd joined in with slogans and rhythmic chanting that we were already used to, like 'Murderers,' 'No to Zionists in our streets' and 'From the river to the sea, Palestine shall be free.'"
They avoided getting into a direct confrontation, Nilsen relates, "and we instructed our group not to scatter and not to respond. But when the atmosphere heated up, some of the other demonstrators – Norwegian men and women of my age – approached the members of the group very closely and whispered into their ear things like 'child murderer' and skadedyr' ['parasites' in Norwegian].
"I've had anti-Israeli calls shouted at me in the past," Nilsen continues. "But this time it was very different. The hatred came from people I thought we shared basic values with. The feeling was that we were being canceled as human beings. We weren't women and men – we were the embodiment of evil."
Parks midday on a Monday were packed. There’s an abundance of minigolf. Workdays in winter start very early so people can get some sunlight outside the office in the afternoon.
Norwegian youth wear the most boring clothes I’ve ever seen in a city. The one signature that stood out were these rainbow-tinted athletic glasses. A few years ago, a comedian made a hit song about the top brand which features a yodel.
Norway had the highest ratio of American to local music I’ve ever seen in a Spotify Top 50. The vast majority of what modern Norwegian hip hop, pop, and indie I came across was flat.
At first I thought there was some adverse selection going on where the best artists try to make it in English, but an arts and culture newspaper editor told me that actually that the cool thing nowadays is to sing in the local language. The Swedes have figured this out…what gives, Norway?
The closest to okay top Norwegian act I came across was Karpe, a rap duo of a Hindu and Muslim second generation immigrants. Electronic music was much stronger. I quite liked this mix and was told they do jazz well too.
Vikings
After flipping through a handful of intro to Vikings books, Children of Ash and Elmstood out for its writing and breadth. It an excellent portrait of the Vikings which brought the terror as well as the humanity to the culture. For instance, I quite liked this discursion into Viking bread.
Some more good writing:
And this:
This list of sea-king names was amazing:
The sagas were also surprisingly accessible and make for great audio books. The Poetic Edda would be my bet for an entry point.
But let’s not forget, the Vikings were actually horrible. This account of a king’s burial by a travelling Arab diplomat in the 900s is one of the most terrifying primary sources I’ve ever come across.
Sexual violence trigger warning.
Modern Norwegian History
Aside from non-fiction on Vikings and Hitler in Norway, the only book-length title I came across telling the history of modern Norway was The Norwegian Exception: Norway’s Liberal Democracy since 1814. I found its thesis hysterical: it’s been incredibly lucky. Its neighbors Sweden, Denmark, and Russia never invaded. The touchiest moment came in 1905 with Sweden…I’m sorry but I can’t help at laughing at the nationalist chest-puffing in Scandanavia.
But ultimately, good call by Norway conceding on the great reindeer dispute of 1905.
Other lucky turns: Norway’s time under Nazi Germany was the easiest ride of any country that got conquered in WWII (good book the occupation here). The country should get some credit for not having a civil war, fumbling the bag when it comes to exploiting the boom in global trade in the late 19th century, successfully leveraging water power to industrialize in the early 20th, and of course making the most out of its oil riches.
But by the 70s, they somehow they became the party of no fun.
WWII
Aside from Vikings, you also have a number of incredibly detailed but not particularly engaging books on Hitler’s invasion. Here’s the case for caring:
The most interesting bits I found were on the strategic level, where before Germany made its move the UK was also dancing around a pre-emptive invasion primarily to secure iron ore. At one point, France pitched the UK to come into the Winter War on the side of the Finns, doing the enormously idiotic move of putting them directly in conflict with the USSR.
Can’t pass on another opportunity to clown on Chamberlain.
Photos
Oslo is big on public art and every other statue was naked. City Hall had some particularly suggestive murals.
Since we launched, the team at @DefenseAnalyses has been hearing more and more from current and former defense thinktankers who are straightjacketed by the stifling bureaucracy and deep risk aversion endemic among @RANDCorporation, @CSIS, @CFR_org, @BrookingsInst and elsewhere. This is an unforced error of gigantic proportions. In this critical time, the United States must ensure that the dynamism of its strategic thinking keeps up with the pace of global change.
We must foster a new generation of defense intellectuals that follow in the best traditions of Andy Marshall, Herman Kahn, and Edward Luttwak. Instead, a sclerotic establishment continues to pile on its limp everything bagel statecraft in the pages of rags like @ForeignAffairs: saying nothing, proposing nothing, committing to nothing.
DARC seeks to create a coalition of those unwilling to wait for the retirement party to bring about change. We believe there would be much progress in defense thinking if analysts simply did not fear the career impact of saying what needed to be said.
To that end, DARC will be publishing an ongoing series of working papers as part of its new Senior Fellows program. These papers will be published pseudonymously, allowing for a candid expression of real views. We are seeking work on the following topics:
Defense Strategy: What are the sacred cows of the defense strategy and foreign policy world when it comes to the war and conflict? Why are they wrong?
Procurement and Supply Chain: What must be done to revitalize armaments innovation and production in the United States? What are authorities that could be used to accelerate progress rapidly?
Future of Conflict: Where is war and conflict going? How does conflict in other domains such as politics, culture, and gaming inform our forecast?
Thinktanks: What has gone wrong with the defense thinktank ecosystem? What can be done to make it better?
To support this work, Senior Fellows will receive between $5,000 and $10,000, depending on the complexity and depth of the work.
The biggest show right now in China (and by big I mean national phenomenon blocking out the sun on weibo and wechat) is a reality show where three celebrity couples all married for at least ten years and on the verge of divorces take an 18-day road trip together. It is some of the most gripping content I've ever seen.
I’ve assembled a starter pack of clips on YouTube below to get you all hooked. There are English subs which are subpar but enough to give nonspeakers the gist. We’ll be recording a special episode to discuss with Emily of the excellent Substack next week.
For some biographical context because you'll be jumping around…From left to right in the image:
couple 1:
Maimai, housewife with no hobbies who doesn’t care about music. Li Xingliang, singer who's moderately but not super successful. They have two kids.
couple 2: Jessica Alba and an even more awful Tony Robbins
Huang Shengyi, the most famous person on the show, an actress whose biggest role was 20 years ago in the classic Kung Fu Hustle. ChatGPT says her American celebrity analogue is Jessica Alba. I’d suggest Hillary Duff. Yangzi, her husband (who started out as her manager...) is a former actor now a dilettante who does antiques and livestreaming and random things. They have two kids, live mostly separately, but she really wants him to still be a part of their kids' lives and he comes in thinking there's nothing wrong with their relationship.
Scene-setting argument featuring Yangzi mansplaining why he stays up till 3am every night with his friends and doesn't on ski vacations with his kids because he doesn't want to support 'western' as opposed to Chinese hobbies (the other couples make fun of him for 'mansplaining' and he doesn't know the word and thinks its a compliment!)
Next up, another legendary argument about him not supporting her professionally where you start to see her push back! Finally, there’s a friend lunch where he talks about his childhood and argues "my parents weren’t around and I turned out fine so why should I be around for my kids..." In the latest episode Yangzi was tolerable for a day and Huang Shengyi said she didn’t want to divorce him but the whole country is hoping she comes to her senses…
couple 3: Scott Disick and someone he doesn’t deserve
Ge Xi, housewife who's now more of an influencer, can support herself and sells things online. Her husband is Liu Shuang, who used to be a very big personality on weibo but is less famous now. They have no kids.
We're gunna skip them for now as they're a little less engaging, but basically he's depressed and not nice to her and she is flowering as a person and realizing she doesn't need him.
If you're intrigued by the clips, I'd next watch the first episode as it gives broader context to the relationships.
Liu Shuang just stays in his room all day, so Ge Xi got herself a one seat sofa for their living room…
Immigration Reform…ish
This is a thing that happened.
Divyansh continues: “The J-1 Skills List required international students from certain countries to return home for 2 years after studying in the U.S. For Chinese STEM scholars, this meant being forced back to China to share cutting-edge knowledge with CCP-controlled industries and participate in its civil-military fusion strategy. The State Department deserves credit for recognizing and addressing this glaring vulnerability. However, the fact that it took nearly 16 years to fix this speaks volumes.”
It is absurd for the Bureau of Consular Affairs to take this long to emerge from their bureaucratic coma to be 2% less terrible to America’s most talented immigrants (not that they should be terrible to any immigrants, which of course they are to thousands on a daily basis…).
This was the most obvious fix Biden’s politicals clearly have been hammering on for years. The fact that they deep stated this change of all things until December 2024 underscores the rot in CA. Here’s to hoping against hope that Trump sticks to what he said about green cards and the Rubio team takes a hammer to CA to make sure it happens.
America spends $700m a year on “Consular Systems and Technology” In 2023 OIG found practically no progress on IT modernization, uncovering that in the 2010s the “CA’s original procurement package” was so bad that “the acquisition process had to be started over, delaying implementation of the CSM program by approximately 58 months.” That five-year clock started in 2012. As of Jan 2023, of the 90 things that IT effort was supposed to modernize, only one was half-fixed. The rot runs back decades.
State shells out another $300m a year on visa adjudication that could be done for $1m of o1 tokens, and a cool billion dollars on passport services. Dear DOGE: we can shave billions off the CA budget while cutting lines to process visas and passports by letting AI do first pass adjudication!
ChinaTalk is a reader-supported publication. To receive new posts and support our work, consider becoming a free or paid subscriber.
Outro music brought to you by the Defense Analyses and Research Corporation
ChinaTalk is hiring for a dedicated China AI lab analyst. Chinese fluency and a technical background are strongly preferred. Apply here!
We’ve got a new show up on the podcast feed with of the Interconnects Substack talking through the biggest AI stories of this year and next. Listen in on Apple Podcasts or Spotify.
Today we’re running a guest piece from Ray Wang, a Washington-based analyst.
On December 2, the Department of Commerce released new export control packages targeting Chinese access to high-bandwidth memory (HBM) with semiconductor manufacturing equipment, including tools essential for HBM manufacturing and packaging, along with the addition of over 140 Chinese chipmakers and chip toolmakers on the Entity List. The new control on HBM — an essential component for AI chips used to train complex AI models and support the AI data center, will constrain China’s AI development which is already hampered by earlier rounds of export controls, including those announced in October 2022, October 2023, and April 2024
.Why HBM Matters
The proliferation of large language models (LLMs) has prompted substantial demand for high-performance computing (HPC) and AI data center infrastructure. HBM, or High-Bandwidth Memory, a type of dynamic random-access memory (DRAM), has become a key component of AI chips — specifically Graphics Processing Units (GPUs) and application-specific integrated circuits (ASICs) that train AI models and power data centers.
Integrating HBM with GPUs or ASICs effectively addresses the so-called “memory wall” bottleneck — a performance constraint caused by the gap between processor speeds and memory access rates. By enabling rapid access to data with lower energy consumption, HBM improves the efficiency of data-intensive AI workloads. This is why most GPUs and ASICs need to incorporate HBM to optimize performance in AI training and inference tasks.
Even from a cost structure perspective, HBM is vital as well, as it accounts for 50% or more of the total cost of an AI chip. Nvidia H100 GPU for example, HBM accounts for 50% of the total cost, followed by 40% of advanced packaging and advanced manufacturing of logic chips, which in Nvidia’s case, are both done by the global foundry leader TSMC. The rest of the materials like printed circuit boards (PCBs) share the last 10% of the total cost.
The global demand for HBM has soared over the past two years, driven by increasing demand for GPUs and ASICs to support AI model training and data center buildout. Morgan Stanley’s December report forecasts the global demand for HBM in 2025 will double that of the 2024 levels. The HBM market size is previously projected to reach up to $33 billion by 2027 — an eightfold increase from $4 billion in 2023. Indeed, there are early signs that prove such a bullish outlook. For instance, HBM suppliers like SK Hynix and Micron have already sold out their HBM production until late 2025.
HBM’s unique function has made it an indispensable component for AI accelerators, as well as the broader AI chip supply chain. Today, almost all leading GPUs and ASICs, including those from Nvidia, AMD, Intel, Google, Amazon, Tesla, Microsoft, and Huawei — integrate HBM to enhance their chips’ performance (see Figure 1). Its essential role has elevated HBM’s strategic value, positioning it as a linchpin in the AI chip supply chain — one of the key reasons prompting the Biden administration’s decision on HBM restriction.
Asian Chipmakers Run the Game
According to Goldman Sachs, SK Hynix and Samsung Electronics dominate the market with more than 90% of the global HBM market (see Figure 2). Notably, SK Hynix and Micron are leading the race in the most advanced HBM, outpacing Samsung, which is struggling to qualify for Nvidia’s standard to supply the most advanced HBM.
SK Hynix, in particular, has emerged as the world’s leading HBM manufacturer, securing the bulk of orders from Nvidia’s advanced GPU — the top HBM buyer in the market. SK Hynix’s success in high-margin HBM has even led to its financial performance outperforming its long-time rival Samsung’s chip sector, which has struggled in both the foundry and HBM sectors (Figure 3).
In addition to memory makers, TSMC is another critical player in this equation. Apart from its renowned capability in advanced logic chip manufacturing — another key component for AI chips, TSMC also controls approximately 90% of the annual global capacity for Chip-on-Wafer-on-Substrate (CoWoS) — an advanced packaging technology required for integrating HBM and logic dies on a silicon interposer and then positions on top of the packaging substrate.
TSMC’s CoWoS advanced packaging capabilities are indispensable because nearly all of the integration of existing GPUs or ASICs with HBMs relies on its advanced packaging in Taiwan. This includes companies such as Nvidia, AMD, Marvell, Broadcom, and AWS. While TSMC’s leadership in advanced logic chip manufacturing already positions itself as one of the most important actors in the AI chip supply chain, its global dominance in CoWoS packaging further consolidates its central role. Interestingly, AI chip packaging is a bottleneck that has yet to be treated with enough attention.
Is China Falling Behind?
China has been lagging behind in both HBM and AI chip packaging — more because of underinvestment as opposed to export controls. HBM has only begun attracting significant attention within the memory industry in the past two years. Before that, it remained largely overlooked. Since 2013, SK Hynix has been developing HBM, initially in partnership with AMD for HBM1. Despite its industry-leading start, it did not translate to instant success for either SK Hynix or AMD due to minimal demand for HBM, generating negligible revenue for its overall DRAM sector. The same dilemma confronted other memory giants as well. Samsung, for example, even dissolved its HBM team in 2019, citing the segment’s limited market potential.
Similarly, while Chinese biggest DRAM makers like CXMT have narrowed the technology gap with competitors in traditional DRAM, they have skipped on HBM development — likely because of its perceived limited market potential. These years of insufficient investment in HBM have left the Chinese memory industry behind the market leader. The same logic applies to the domestic packaging for AI chips.
This gap becomes even clearer when closely examining the product roadmap of the four major DRAM manufacturers closely (see Figure 5). Samsung commenced mass production of HBM2 (2nd generation of HBM) in 2016, followed by SK Hynix in 2018. Chinese memory maker CXMT however, only recently began its massive production of HBM2, suggesting that China is roughly 6 to 8 years or three generations behind the front-running manufacturers. This gap is evident in earlier reports of Huawei and Baidu stockpiling Samsung’s HBM2E (3rd generation of HBM) and Chinese domestic firms still in the process of developing HBM2.
Based on the product roadmap, CXMT should be able to catch up with existing advanced HBM in roughly six to eight years. Yet, the existing and recent restrictions on semiconductor manufacturing equipment (SME), including manufacturing and packaging tools for HBM could push out that timeline. Many SMEs have overlapping functions (e.g. etching, lithography) for HBM and logic chip manufacturing, as well as advanced packaging processes. As a result, these restrictions, whether directly targeting logic chipmaking, HBM manufacturing, or packaging, are likely to hamper firm’s progress in HBM and the advanced packaging it requires. These challenges are further exacerbated by existing curbs on advanced lithography tools critical for cutting-edge HBM production.
It is also worth considering how the previous restrictions on advanced memory chips might affect China’s HBM development. Since HBM is essentially a memory technology that stacks several DRAM dies, limitations on advanced DRAM chips could continue to be a roadblock to China’s HBM advancement.
More importantly, taking the pace of development into account is pivotal. If Chinese memory makers continue to advance slower than market leaders in the coming years, the technological gap will be hard to narrow. In 2024, Chinese GPUs and ASICs are estimated to account for merely 1% of global HBM consumption. The rest is comprised of consumption from U.S. firms like Nvidia, Google, AMD, AWS, Intel, Microsoft, and Tesla — all reliant on the HBM from SK Hynix, Samsung, and Micron. The 1% share of HBM consumption by Chinese GPUs and ASIC, is mainly supplied by Samsung, instead of Chinese memory makers.
To that end, SK Hynix, Samsung, and Micron can generate much more revenue than Chinese memory makers from global GPUs/ASICs firms in coming years and reinvest it in R&D for the next generation of HBM or other areas essential for the company’s development. HBM’s strong market growth also makes it easier for these firms to compel their leadership and investors to allocate more resources to HBM development to maintain or even expand its edge — a trend already evident in companies like SK Hynix and Samsung. These business rationales, in contrast, will not necessarily apply to the Chinese memory firms given the limited demand for now.
Samsung is also a big loser for BIS’ new rule. 20% of its HBM revenue in 2024 was to China, and those sales are now banned. This impact should be soon shown in Samsung’s earnings in the coming quarters. On the other hand, the new rule should have a relatively small impact on SK Hynix and Micron, which both supply their HBM mostly to Nvidia and other non-Chinese firms.
Lastly, China's advanced packaging technologies and capacity remain limited. Compounding this challenge, AI chip packaging leader TSMC is unlikely to provide services to leading Chinese AI firms due to existing restrictions. With that in mind, even if China makes advancements in HBM technology in the coming years, its ability to close the gap with TSMC in advanced packaging remains uncertain under enhanced SME restrictions. Without advanced packaging capability, Chinese HBM will struggle to optimally incorporate it with GPUs or ASICs, which will ultimately affect their AI chips’ performance. Admittedly, emerging Chinese packagers like JCET and Tongfu Microelectronics have “CoWoS-like” packaging capability, it is yet unclear how successfully these firms can package the domestic HBM and GPU given the limited information.
That said, one should not underestimate the Chinese capability to close the gap with the market leader. Leading memory makers like YMTC and CXMT have proved their ability to rapidly narrow the gap in NAND Flash and DRAM with the ability to rapidly ramp up capacity to disrupt the market. Given the optimistic outlook for domestic HBM demand for GPUs and ASICs, increasing R&D investment, and continued government support, Chinese memory and packaging firms are poised to accelerate technological advancements. This is likely true at a time when both government and industry have heightened urgency to develop domestic HBM and AI chip supply chains amid increasing U.S. restrictions. Moreover, Chinese President Xi’s pursuit of “High-Quality Productive Forces” and “Self-Sufficiency” is likely to bring more government support for domestic HBM and AI chip supply chains.
These factors are likely to compel domestic GPU and ASIC providers to adopt homegrown HBM, stimulating the memory industry’s growth and spurring more public and private investment in this area. Chinese AI chip companies are also expected to enlarge their collaboration with domestic HBM maker and advanced packaging firms given their limited access to foreign products and the imperative to strengthen the local AI chip supply chain.
In fact, there are already some signs indicating these trends. Following Beijing's call earlier this year to prioritize domestic chip adoption, several Chinese industry groups issued statements on Monday, warning domestic firms that "U.S. chips are unreliable" in response to the BIS's new restrictions on Monday. Recent reports also suggested that Huawei, for example, alongside the government, is supporting local HBM and advanced packaging capabilities. Additionally, domestic foundries like XMC are reportedly ramping up efforts to produce HBM, signaling early efforts in building a Chinese ecosystem for HBM.
China’s AI development may not face immediate setbacks given that much of the advanced hardware supporting its AI industry is still mostly foreign made. For instance, most leading AI firms — such as Alibaba, Baidu, and Tencent still train their models with Nvidia GPUs procured before restrictions. Similarly, Huawei’s latest Ascend GPUs still use SK Hynix and Samsung’s HBM2 and HBM2E, also sourced before the restrictions took effect. China's semiconductor industry is likely to feel the impact in late 2025 or 2026, considering many Chinese firms have been preparing for this restriction by purchasing additional equipment over the past year. Nevertheless, China’s AI and semiconductor industry are ultimately on track to encounter a substantial “hardware bottleneck.” They will increasingly feel the impact of restrictions on high-end logic and memory chips (including HBM), as well as SMEs. Huawie’s chipmaking partner, SMIC, for example, is already struggling with producing logic chips below 7nm with commercially viable yield rates, despite earlier progress. The memory leader CXMT, is likely to face a similar struggle as the SME restriction disrupts its HBM product development and production. .
In short, the forthcoming restrictions on advanced HBM access will impede the performance of future Chinese AI chips, including those from major players like Huawei, and startups like Biren and Moore Thread. The broader SME export control will undermine China’s ability to develop and enhance its HBM and AI chips.
Despite export controls significantly impacting the industry, they cannot entirely block Chinese firms from advancing in critical technologies but instead force progress through costlier, slower, and more challenging paths.
ChinaTalk is a reader-supported publication. To receive new posts and support our work, consider becoming a free or paid subscriber.
Today we’re running a guest piece on lessons from the 2010 China-Japan critical minerals kerfuffle to the recent export controls China dropped in response to the revised Biden semiconductor controls. Authors Seaver Wang, Peter Cook, and Lauren Teixeira are analysts at The Breakthrough Institute, an environmental think tank based in Berkeley.
ChinaTalk is heading to NeurIPS next week! Respond to this email to connect we’d love to meet up in person.
Unpacking the Cautionary Tale of the China-Japan Rare Earths Incident
Following the Biden administration’s recent expansion of restrictions on the sale of advanced semiconductor chip and manufacturing technologies to China, Chinese policymakers have responded rapidly by restricting exports of the critical minerals germanium, gallium, and antimony to the United States. Other new provisions, whose specifics remain unclear at the time of writing, may target additional materials like tungsten and graphite. This announcement comes in the wake of recent moves by Beijing to establish stricter export permit frameworks for a range of critical commodities including tungsten, graphite, magnesium, and aluminum alloys.
With U.S.-China trade tensions only likely to intensify in coming months, such raw material supply chain risks are becoming increasingly relevant for energy transition efforts as a whole. Other critical mineral export restrictions relevant for clean energy technologies could conceivably soon follow, with effects that may not be limited to the United States.
In Myanmar, for instance, resistance fighters from the Kachin Independence Organisation recently captured much of Pang War township, shutting down Myanmar’s largest rare earth mining hub and prompting Chinese authorities to close the nearby border crossing. Virtually unregulated mining in Myanmar contributes an unknown but large fraction of global rare earth element (REE) mining, with all of Myanmar’s production shipped to neighboring China, the world’s dominant refiner of rare earth materials. This latest development in a civil war turning increasingly in favor of anti-junta resistance groups has begun to send shivers through critical mineral markets, with industry observers speculating about a future rare earth shortage and price spikes.
If REEs become scarce, Chinese policymakers may clamp down on REE exports next to reserve more of these valuable critical minerals for domestic high-tech industries. Such restrictions could impose constraints upon the rest of the world, including overseas efforts to nurture clean technology sectors like wind power and electric vehicles that rely upon REEs for permanent magnet drives.
In response to critiques that clean energy sectors depend overly on key commodities imported from China, climate advocates and climate policy hawks have oftenarguedthat a country only needs to import battery minerals, solar panels, or electric cars once, at which point they escape the control of the exporting trade partner while operating for decades. This contrasts with continuous flows of fossil fuels whose interruption can immediately catapult energy supplies into a crisis. This reasoning is correct in principle, but supply chain disruptions would nevertheless stall acquisition—or manufacturing—of subsequent batches of low-carbon technologies. Such supply vulnerabilities can certainly freeze energy transition efforts in their tracks and force factories producing energy technologies to go idle.
Map of rare earth mining activity in Myanmar in the vicinity of Pang War township (upper right) and Momauk township (lower right). Orange circles in left-hand map and yellow dots in right-hand map indicate surface collection ponds at rare earth mine sites identified from satellite imagery by Myanmar Witness, Mizzima, and Global Witness, from whose work this figure is adapted. Junta-aligned forces previously controlled mines near Pang War, with the region changing hands recently following capture by the Kachin Independence Organisation. Mining near Momauk township has proceeded under the administration of the Kachin Independence Organisation, which has canceled some proposed new projects and is reviewing existing mining practices amidst pressure from local communities.
The preeminent and most-invoked example of supply chain coercion remains China’s disruption of rare earth exports to Japan during the 2010 Senkaku Islands incident. A critical reexamination of this incident reveals a few useful insights that may help observers better understand these newest limits on critical mineral exports from China. The de-facto embargo in 2010 was likely opportunistic rather than planned—an impromptu exploitation of an issue prominent in preceding China-Japan negotiations rather than the culmination of some grand industrial geostrategic conspiracy. Yet this incident certainly cemented the perception that Chinese policymakers have and will wield control of strategic supply chains for geopolitical leverage—a concern that Beijing itself has reinforced since. These latest developments emphasize that policymakers and clean technology companies should be taking immediate steps to reduce supply chain vulnerabilities, if necessary even at what may seem like uneconomically high cost.
When Geopolitical Tensions Spill over into Supply Chains
According to conventional retellings, a number of Japanese companies reported a halt to expected rare earth ore shipments from China beginning Tuesday, September 21, 2010. This coincided with more overt political pressure tactics that had begun days prior, including a cessation of ministerial and provincial exchanges, a popular campaign to limit tourist visits to Japan, the detainment of four Japanese nationals in China, and an exclusion of Japanese companies from bidding on Chinese public projects. Many of these actions followed a Japanese government decision on September 19 to extend the detention of a Chinese fishing boat captain involved in collisions with two Japanese Coast Guard vessels near the Senkaku Islands on September 7.
Japan released the captain on September 25, and Chinese customs offices partially resumed clearing some REE shipments for export several days later. However, international traders, companies, and government officials continued to report systematic interruptions and delays for shipments to Japan as well as some U.S. and Europe-bound exports throughout October and the first half of November.
Asahi reporting from September 24 already described the situation as “appearing to be an effective embargo”, quoting a China-based REE manufacturing executive as having received instructions from Chinese customs officials to “stop exports until the 29th”. International customers including representatives of Australian, Canadian, Chinese, and Japanese firms all confirmed the suspension of shipments. Japanese government surveys of industry stakeholders in late September 2010 reported a clear consensus that export problems had increased after 21 September, driven by numerous sudden changes in Chinese customs enforcement. Out of 31 responding firms that confirmed their involvement in rare earths trade, all 31 reported encountering export obstacles. By Tuesday 28 September, Japan’s Minister for Economic and Fiscal Policy Banri Kaieda described the situation accordingly at a press conference: “Right now, the de-facto export prohibition that China has adopted is causing profoundly great impacts on Japan’s economy.”.
Notably, contemporary commentary showed a clear understanding that China had already slashed their REE export quotas a few months earlier, observing that the disruptions beginning in late September seemed separate and distinct from this earlier policy change. Articles by Toyo Keizai and Mitsubishi’s think tank MUFC published just days before the export halt very matter-of-factly articulated that China was reducing export quotas to nurture domestic industries, regulate foreign investment in the sector, and limit expansion of new mining for environmental reasons. Reporting from September 25 highlighted arguments by industry observers that they wouldn’t expect producers to exhaust their export quotas until late October at the earliest, with traders noting that even Chinese producers with ample spare export quotas had “been dissuaded” from exporting. In subsequent interviews with researchers, Japanese officials confirmed an internal understanding at the time that the central Chinese government had issued an order, and that the Japanese government interpreted the incident as an economic sanction.
These contemporaneous official government statements, media reporting, comments from industry, and policy responses across English, Japanese, and Chinese-language sources shared a clear understanding that Chinese officials had implemented a de-facto export ban, prompting many governments worldwide to lodge protests while urgently pursuing supply chain alternatives and countermeasures. Even a Chinese People’s Daily article from late 2012 more or less stated: “Even though China did not publicly admit to employing economic sanctions, China did in reality halt export shipments, subjecting Japan to some difficulties at the time”.
How Real were the Impacts on Rare Earth Element Trade?
Some work in late 2010 and sincehaschallenged this prevailing storyline, arguing that this period of scarce supply and price spikes beginning in late 2010 did stem mostly from China’s stricter export quotas imposed in July—a policy action well predating the diplomatic dispute. Such commentators argue China’s export policies sought only favorable domestic economic outcomes—more stringent environmental regulation of the REE sector and better capture of value-added downstream industries. Revisionist retellings at times go even further, arguing that any resulting supply shortages in late 2010 did not explicitly target Japan and warning against invented narratives of Chinese mineral supply chain coercion. However, such interpretations stray from the historical record and often overstate their case.
Commentators arguing that China’s undeclared interference in rare earth trade in late 2010 is exaggerated “folklore” have often cited a few articles that analyzed monthly Japanese customs or UN data on value of trade in various goods, claiming that overall, broad categories of rare earths traded with Japan do not seem to exhibit quantitative disruptions during this period. However, such low-resolution, indirect data does not distinguish between far more valuable heavy rare earth elements important for high-tech applications versus more abundant light rare earths like cerium that primarily see use in more mundane industrial processes.
Overall, the cited data don’t contain enough detail to draw a clear conclusion that no significant disruptions occurred. Summary data on total rare earth shipment arrivals in Japan might conceal a decline in imports from China compensated by urgently redirected materials sourced from Southeast Asia, Europe, or North America. Similarly, indirect metrics like the monthly value of rare earth shipments from China to Japan may not accurately capture simultaneously evolving variables, like a decline in import tonnage offset by a corresponding spike in rare earth prices. Moreover, monthly-scale data may not confidently capture shorter-term disruptions and delays that began towards the end of September 2010 and varied from week to week thereafter. Finally, such sterile retrospective analyses of trade data in isolation ignores a vast weight of contemporaneous, corroborating testimony and reporting, such as the official surveys of affected industries.
One should also recall that broader Chinese economic coercion aimed at Japan in late September 2010 was not seeking to damage Japan materially so much as to accomplish a specific goal: the successful release of the detained fishing boat captain.
It is true however that China did dramatically alter export and industrial policies for the rare earths sector earlier in 2010. These changes indeed stemmed in part from domestic environmental considerations and aspirations to further develop downstream value-added industries like rare earth permanent magnet manufacturing. And while the particularly sharp reduction in export quotas in July 2010 generated significant international attention and discussion for months predating the Senkaku islands dispute, Chinese national policy had long treated REEs as a strategic commodity and regularly revised regulations and export practices over the years.
While rare earth elements are widely distributed globally, southern China hosts a notable concentration of ion adsorption clay (IAC) deposits, located at relatively shallow depths and mineable using simpler methods in small-scale operations. These deposits tend to form in temperate or tropical climates with higher temperatures and rainfall that can leach REEs from bedrock and concentrate them in clay soils. IAC deposits typically contain higher grades of heavy REEs relative to hardrock REE deposits, may not require onsite milling, and allow for initial processing onsite using pit leaching, often using ammonium sulfates. Such low-cost IAC mining operations have driven much of the growth in China’s rare earth sector over recent decades, albeit with considerable environmental impacts that have prompted stricter regulations since the mid-2010s.
Starting in 1985, the Chinese government began offering an export rebate to REE enterprises to encourage rare earth exports, refunding the value-added tax that producers paid on exported products. Following China’s overtaking of the United States as the world’s largest REE producer in the late 1980s, Chinese policymakers designated REEs as strategic minerals as early as 1990, with national production ramping up dramatically through the 1990s thanks largely to growth in small-scale projects targeting IAC deposits. By 2000, in light of increasing domestic industry demand for rare earths, the central government reduced export rebates before eliminating them altogether in 2005. With the subsequent introduction of export duties for rare earths in 2007, China’s export strategy had entirely reversed. Policymakers had already implemented export quotas years earlier in 1999 to control total national production and curb smuggling, and would progressively reduce quotas every year between 2005 and 2010.
Illustrative timeline of Chinese rare earth element sector trends and industrial policy actions.
The dramatically lowered export quota announced in mid-2010 may very well have contributed to the continuing customs delays and export disruptions throughout October and November that year. But the intensity and timing of events in late September coincided far too closely with the China-Japan diplomatic crisis to discount as a bureaucratic coincidence. Even Chinese commerce minister Chen Deming drew a link between the two issues in a televised interview on 26 September, suggesting that Chinese businesses might be acting on their own patriotic initiative to pause shipments.
International governments certainly interpreted this rare earths shock as an undeclared set of sanctions. In retrospect perhaps these trade disruptions appear short-lived to observers today, but in the moment the affected actors saw these measures as indefinite and reacted with alarm. By October 3rd, Japanese officials were negotiating with the Mongolian government to develop new rare earth mining projects in Mongolia. By late October, Japan and Korea had announced a partnership in which Japan would help Korea survey potential deposits. Within a couple months, the United States and Japan were exploring projects in California, Australia, and Indonesia.
Implications of China’s Recent Export Restriction
Chinese policymakers likely weaponized rare earth exports opportunistically in the moment. The framework for export limits did genuinely originate out of industrial policy crafted with China’s national interest in mind, and well predated the tensions that prompted their temporary weaponization. Japan and China had been negotiating specifically over rare earth export quotas earlier in the year—including just weeks beforehand. With the issue fresh in recent memory, Beijing policymakers understood full well that this was a powerful lever in China-Japan relations.
The 2010 rare earths disruption and the current situation now unfolding between China and the United States share some similarities but also exhibit notable differences. As in 2010, the new Chinese export restrictions on gallium, germanium, and other critical mineral shipments to the U.S. clearly form part of a planned tit-for-tat response to the latest U.S. restrictions targeting the Chinese semiconductor industry manufacturing chain. In contrast to the 2010 incident, however, Beijing’s leverage of critical mineral supply chains is now overt and explicit, rather than ambiguous. Furthermore, whereas disruption of rare earth shipments to Japan may have served a narrow, temporary geopolitical purpose, U.S.-China cooperation for advanced technology leadership clearly spans a far broader scope, with no simple or near-term resolution in sight.
Meanwhile, with China now operating export control frameworks for everything from tungsten to magnesium to rare earth concentrating equipment to solar manufacturing machinery, the possibility of further escalation looms large—with significant implications for clean technology supply chains and trade.
The lesson from the 2010 rare earths shock and its origins emphasizes that Chinese export controls on critical minerals likely originated out of narrow national economic self-interest, rather than serving as part of some grand strategic conspiracy. But fundamentally speaking, the combination of overwhelming market share control and absolute authority over export policies means that Beijing can control market supply and international prices for a wide host of critical commodities with the stroke of a pen, an ability whose geopolitical utility is clearly now obvious to Chinese leaders. Should the right situation arise, the tools for bottlenecking trade already exist, including substantial latitude for subtle, undeclared, and plausibly deniable economic coercion in addition to the overt measures now enjoying the spotlight.
The only solution to this dynamic of self-perpetuating Chinese critical mineral market overconcentration is a forceful strategy of supply chain expansion and diversification. Such a strategy must dispense with the futile practice of tepidly ushering new entrants into unforgiving markets built upon lopsided terms of competition. Rather, governments must stubbornly and persistently ensure that alternative producers survive and multiply—in and of itself a necessary prerequisite for fostering competition and breaking monopoly power.
Ultimately, the world’s access to crucial advanced energy technologies cannot depend upon some People’s Liberation Army Air Force pilot’s ability to execute a reckless aerial maneuver around a Taiwanese patrol aircraft. The ease with which even the most optimistic clean energy commentator can imagine such a contingency should stress that both the geopolitical and decarbonization stakes of such efforts are high.
Back to Jordan writing: For some contemporary context, Bloomberg’s Gerard Dipippo echoes my take. As long as China is still selling outside the country, the US firms can play the trade diversion game just as well as Huawei and SMIC can.
Deepseek is a Chinese AI startup whose latest R1 model beat OpenAI’s o1 on multiple reasoning benchmarks. Despite its low profile, Deepseek is the Chinese AI lab to watch.
Before Deepseek, CEO Liang Wenfeng’s main venture was High-Flyer (幻方), a top 4 Chinese quantitative hedge fund last valued at $8 billion. Deepseek is fully funded by High-Flyer and has no plans to fundraise. It focuses on building foundational technology rather than commercial applications and has committed to open sourcing all of its models. It has also singlehandedly kicked off price wars in China by charging very affordable API rates. Despite this, Deepseek can afford to stay in the scaling game: with access to High-Flyer’s compute clusters, Dylan Patel’s best guess is they have upwards of “50k Hopper GPUs,” orders of magnitude more compute power than the 10k A100s they cop to publicly.
Deepseek’s strategy is grounded in their ambition to build AGI. Unlike previous spins on the theme, Deepseek’s mission statement does not mention safety, competition, or stakes for humanity, but only “unraveling the mystery of AGI with curiosity”. Accordingly, the lab has been laser-focused on research into potentially game-changing architectural and algorithmic innovations.
Deepseek has delivered a series of impressive technical breakthroughs. Before R1-Lite-Preview, there had been a longer track record of wins: architectural improvements like multi-head latent attention (MLA) and sparse mixture-of-experts (DeepseekMoE) had reduced inference costs so much as to trigger a price war among Chinese developers. Meanwhile, Deepseek’s coding model trained on these architectures outperformed open weights rivals like July’s GPT4-Turbo.
As a first step to understanding what’s in the water at Deepseek, we’ve translated a rare, in-depth interview with CEO Liang Wenfeng, originally published this past July on a 36Kr sub-brand. It contains some deep insights into:
How DeepSeek’s ambitions for AGI flow through their research strategy
Why it views open source as the dominant strategy and why it ignited a price war
How he hires and organizes researchers to leverage young domestic talent far better than other labs that have splurged on returnees
Why Chinese firms settle for copying and commercialization instead of “hardcore innovation” and how Liang hopes Deepseek will ignite more “hardcore innovation” across the Chinese economy.
Uncovering DeepSeek: The Ultimate Tale of Chinese Tech Idealism
Of China’s seven large-model startups, DeepSeek has been the most discreet — yet it consistently manages to be memorable in unexpected ways.
A year ago, this unexpectedness came from its backing by High-Flyer 幻方, a quantitative hedge fund powerhouse, making it the only non-big tech giant with a reserve of 10,000 A100 chips. A year later, it became known as the catalyst for China’s AI model price war. A year later, it became known as the catalyst for China’s AI model price war.
In May, amid continuous AI developments, DeepSeek suddenly rose to prominence. The reason was that they released an open-source model called DeepSeek V2, which offered an unprecedented price/performance ratio: inference costs were reduced to only 1 RMB per million tokens, which is about one-seventh of the cost of Llama3 70B and one-seventieth of the cost of GPT-4 Turbo.
DeepSeek was quickly dubbed the “Pinduoduo of AI,” and other major tech giants such as ByteDance, Tencent, Baidu, and Alibaba couldn’t hold back, cutting their prices one after another. A price war for large models in China was imminent.
This diffuse smoke of war actually concealed one fact: unlike many big companies burning money on subsidies, DeepSeek is profitable.
This success stems from DeepSeek’s comprehensive innovation in model architecture. They proposed a novel MLA (multi-head latent attention) architecture that reduces memory usage to 5-13% of the commonly used MHA architecture. Additionally, their original DeepSeekMoESparse structure minimized computational costs, ultimately leading to reduced overall costs.
In Silicon Valley, DeepSeek is known as “the mysterious force from the East” 来自东方的神秘力量. SemiAnalysis’s chief analyst believes the DeepSeek V2 paper “may be the best one of the year.” Former OpenAI employee Andrew Carr found the paper “full of amazing wisdom” 充满惊人智慧, and applied its training setup to his own models. And Jack Clark, former policy head at OpenAI and co-founder of Anthropic, believes DeepSeek “hired a group of unfathomable geniuses” 雇佣了一批高深莫测的奇才, adding that large models made in China “will be as much of a force to be reckoned with as drones and electric cars” 将和无人机、电动汽车一样，成为不容忽视的力量.
In the AI wave — where the story is largely driven by Silicon Valley — this is a rare occurrence. Several industry insiders told us that this strong response stems from innovation at the architectural level, a rare attempt by domestic large model companies and even global open-source large-scale models. One AI researcher said that the Attention architecture has hardly been successfully modified, let alone validated on a large scale, in the years since it was proposed. “It’s an idea that would be shut down at the decision-making stage because most people lack confidence” 这甚至是一个做决策时就会被掐断的念头，因为大部分人都缺乏信心.
On the other hand, large domestic models have rarely dabbled in innovation at the architectural level before, partly due to a prevailing belief that Americans excel at 0-to-1 technical innovation, while Chinese excel at 1-to-10 application innovation. Moreover, this kind of behavior is very unprofitable — after all, a new generation of models will inevitably emerge after a few months, so Chinese companies need only follow along and focus on downstream applications. Innovating the model architecture means that there is no path to follow, meaning multiple failures and substantial time and economic costs.
DeepSeek is clearly going against the grain. Amid the clamor that large-model technology is bound to converge and following is a smarter shortcut, DeepSeek values the learning accumulated through “detours” 弯路, and believes that Chinese large-model entrepreneurs can join the global technological innovation stream beyond just application innovation.
Many of DeepSeek’s choices differ from the norm. Until now, among the seven major Chinese large-model startups, it’s the only one that has given up the “want it all” 既要又要 approach, so far focusing on only research and technology, without the toC applications. It’s also the only one that hasn’t fully considered commercialization, firmly choosing the open-source route without even raising capital. While these choices often leave it in obscurity, DeepSeek frequently gains organic user promotion within the community.
How did DeepSeek achieve this all? We interviewed DeepSeek’s seldom-seen founder, Liang Wenfeng 梁文锋, to find out.
The post-80s founder, who has been working behind the scenes on technology since the High-Flyer era, continues his low-key style in the DeepSeek era — “reading papers, writing code, and participating in group discussions” 看论文，写代码，参与小组讨论 every day, just like every other researcher does.
And unlike many quant fund founders — who have overseas hedge-fund experience and physics or mathematics degrees — Liang Wenfeng has always maintained a local background: in his early years, he studied artificial intelligence at Zhejiang University’s Department of Electrical Engineering.
Multiple industry insiders and DeepSeek researchers told us that Liang Wenfeng is a very rare person in China’s AI industry — someone who has “both strong infra engineering and modeling capabilities, as well as the ability to mobilize resources” he “can make accurate, high-level judgments, while also remaining stronger than first-line researchers in the details”. He has a “terrifying ability to learn”, and at the same time, he is “not at all like a boss and much more like a geek.”
This is a particularly rare interview. Here, this technological idealist provides a voice that is especially scarce in China’s tech world: he is one of the few who puts “right and wrong” before “profits and losses” 把“是非观”置于“利害观”之前, who reminds us to see the inertia of the times, and who puts “original innovation” 原创式创新 at the top of the agenda.
A year ago, when DeepSeek first came off the market, we interviewed Liang Wenfeng: “Crazy High-Flyer: A Stealth AI Giant’s Road to Large Models” 疯狂的幻方：一家隐形AI巨头的大模型之路. If the phrase “be insanely ambitious and insanely sincere” 务必要疯狂地怀抱雄心，且还要疯狂地真诚 was merely a beautiful slogan back then, a year later, it has become action.
Part 1: How was the first shot of the price war fired?
Waves: After DeepSeek V2’s release, it quickly triggered a fierce price war in the large-model market. Some say you’ve become the industry’s catfish.
Liang Wenfeng: We didn’t mean to become a catfish — we just accidentally became a catfish. [Translator’s note: This is likely a reference to Wong Kar-wai’s new tv show 王家卫“Blossoms Shanghai” 繁花, where catfish are symbolic of market disruptors due to their cannibalistic nature.]
Waves: Was this outcome a surprise to you?
Liang Wenfeng: Very surprising. We didn’t expect pricing to be so sensitive to everyone. We were just doing things at our own pace and then accounted for and set the price. Our principle is that we don’t subsidize nor make exorbitant profits. This price point gives us just a small profit margin above costs.
Waves: Zhipu AI 智谱AI followed suit five days later, followed by ByteDance, Alibaba, Baidu, Tencent, and other big players.
Liang Wenfeng: Zhipu AI reduced the price of an entry-level product, while their models comparable to ours remained expensive. ByteDance was truly the first to follow, reducing its flagship model to match our price, which then triggered other tech giants to cut prices. Since big companies’ model costs are much higher than ours, we never expected anyone would do this at a loss, but it eventually turned into the familiar subsidy-burning logic of the internet era.
Waves: From the outside, price cuts look a lot like bids for users, which is usually the case in internet-era price wars.
Liang Wenfeng: Poaching users is not our main purpose. We cut prices because, on the one hand, our costs decreased while exploring next-generation model architectures, and on the other hand, we also feel that both APIs and AI should be accessible and affordable to everyone.
Waves: Before this, most Chinese companies would directly copy the current generation’s Llama architecture for applications. Why did you start from the model structure?
Liang Wenfeng: If the goal is to make applications, using the Llama structure for quick product deployment is reasonable. But our destination is AGI, which means we need to study new model structures to realize stronger model capability with limited resources. This is one of the fundamental research areas needed for scaling up to larger models. And beyond model structure, we’ve done extensive research in other areas, including data construction and making models more human-like — which are all reflected in the models we released. In addition, Llama’s structure, in terms of training efficiency and inference cost, is estimated to have a two-generation gap behind international frontier levels in training efficiency and inference costs.
Waves: Where does this generation gap mainly come from?
Liang Wenfeng: First of all, there’s a training efficiency gap. We estimate that compared to the best international levels, China’s best capabilities might have a twofold gap in model structure and training dynamics — meaning we have to consume twice the computing power to achieve the same results. In addition, there may also be a twofold gap in data efficiency, that is, we have to consume twice the training data and computing power to achieve the same results. Combined, that’s four times more computing power needed. What we’re trying to do is to keep closing these gaps.
Waves: Most Chinese companies choose to have both models and applications. Why has DeepSeek chosen to focus on only research and exploration?
Liang Wenfeng: Because we believe the most important thing now is to participate in the global innovation wave. For many years, Chinese companies are used to others doing technological innovation, while we focused on application monetization — but this isn’t inevitable. In this wave, our starting point is not to take advantage of the opportunity to make a quick profit, but rather to reach the technical frontier and drive the development of the entire ecosystem.
Waves: The Internet and mobile Internet eras left most people with the belief that the United States excels at technological innovation, while China excels at making applications.
Liang Wenfeng: We believe that as the economy develops, China should gradually become a contributor instead of freeriding. In the past 30+ years of the IT wave, we basically didn’t participate in real technological innovation. We’re used to Moore’s Law falling out of the sky, lying at home waiting 18 months for better hardware and software to emerge. That’s how the Scaling Law is being treated.
But in fact, this is something that hasbeen created through the tireless efforts of generations of Western-led tech communities. It’s just because we weren’t previously involved in this process that we’ve ignored its existence.
Part 2: The Real Gap Isn’t One or Two Years. It’s Between Original Innovation and Imitation.
Waves: Why did DeepSeek V2 surprise so many people in Silicon Valley?
Liang Wenfeng: Among the numerous innovations happening daily in the United States, this is quite ordinary. They were surprised because it was a Chinese company joining their game as an innovation contributor. After all, most Chinese companies are used to following, not innovating.
Waves: But choosing to innovate in the Chinese context is a very extravagant decision. Large models are a heavy investment game, and not all companies have the capital to solely research and innovate instead of thinking about commercialization first.
Liang Wenfeng: The cost of innovation is definitely not low, and past tendencies toward indiscriminate borrowing were also related to China’s previous conditions. But now you see, whether it’s China’s economic scale, or the profits of giants like ByteDance and Tencent — none of it is low by global standards. What we lack in innovation is definitely not capital, but a lack of confidence and knowledge of how to organize high-density talent for effective innovation.
Waves: Why do Chinese companies — including the huge tech giants — default to rapid commercialization as their #1 priority?
Liang Wenfeng: In the past 30 years, we’ve emphasized only making money while neglecting innovation. Innovation isn’t entirely business-driven; it also requires curiosity and a desire to create. We’re just constrained by old habits, but this is tied to a particular economic phase.
Waves: But you’re ultimately a business organization, not a public-interest research institution — so where do you build your moat when you choose to innovate and then open source your innovations? Won’t the MLA architecture you released in May be quickly copied by others?
Liang Wenfeng: In the face of disruptive technologies, moats created by closed source are temporary. Even OpenAI’s closed source approach can’t prevent others from catching up. So we anchor our value in our team — our colleagues grow through this process, accumulate know-how, and form an organization and culture capable of innovation. That’s our moat.
Open source, publishing papers, in fact, do not cost us anything. For technical talent, having others follow your innovation gives a great sense of accomplishment. In fact, open source is more of a cultural behavior than a commercial one, and contributing to it earns us respect. There is also a cultural attraction for a company to do this.
Waves: What do you think of those who believe in the market, like [GSR Ventures’[ Zhu Xiaohu 朱啸虎?
Liang Wenfeng: Zhu Xiaohu is logically consistent, but his style of play is more suitable for fast money-making companies. And if you look at America’s most profitable companies, they’re all high-tech companies that accumulated deep technical foundations before making major breakthroughs.
Waves: But when it comes to large models, pure technical leadership rarely forms an absolute advantage. What bigger thing are you betting on?
Liang Wenfeng: What we see is that Chinese AI can’t be in the position of following forever. We often say that there is a gap of one or two years between Chinese AI and the United States, but the real gap is the difference between originality and imitation. If this doesn’t change, China will always be only a follower — so some exploration is inescapable.
Nvidia’s leadership isn’t just the effort of one company, but the result of the entire Western technical community and industry working together. They see the next generation of technology trends and have a roadmap in hand. Chinese AI development needs such an ecosystem. Many domestic chip developments struggle because they lack supporting technical communities and have only second-hand information. China inevitably needs people to stand at the technical frontier.
Part 3: More Investments Do Not Equal More Innovation
Waves: DeepSeek, right now, has a kind of idealistic aura reminiscent of the early days of OpenAI, and it’s open source. Will you change to closed source later on? Both OpenAI and Mistral moved from open-source to closed-source.
Liang Wenfeng: We will not change to closed source. We believe having a strong technical ecosystem first is more important.
Waves: Do you have a financing plan? I’ve seen media reports saying that High-Flyer plans to spin off DeepSeek for an IPO. AI startups in Silicon Valley inevitably end up binding themselves to major firms.
Liang Wenfeng: We do not have financing plans in the short term. Money has never been the problem for us; bans on shipments of advanced chips are the problem.
Waves: Many people believe that developing AGI and quantitative finance are completely different endeavors. Quantitative finance can be pursued quietly, but AGI may require a high-profile and bold approach, forming alliances to amplify your investments.
Liang Wenfeng: More investments do not equal more innovation. Otherwise, big firms would’ve monopolized all innovation already.
Waves: Are you not focusing on applications right now because you lack the operational expertise?
Liang Wenfeng: We believe the current stage is a period of explosive growth in technological innovation, not in applications. In the long run, we hope to create an ecosystem where the industry directly utilizes our technology and outputs. Our focus will remain on foundational models and cutting-edge innovation, while other companies can build B2B and B2C businesses based on DeepSeek’s foundation. If a complete industry value chain can be established, there’s no need for us to develop applications ourselves. Of course, if needed, nothing stops us from working on applications, but research and technological innovation will always be our top priority.
Waves: But when customers are choosing APIs, why should they choose DeepSeek over offerings from bigger firms?
Liang Wenfeng: The future world is likely to be one of specialized division of labor. Foundational large models require continuous innovation, and large companies have limits on their capabilities, which may not necessarily make them the best fit.
Waves: But can technology itself really create a significant gap? You’ve also mentioned that there are no absolute technological secrets.
Liang Wenfeng: There are no secrets in technology, but replication requires time and cost. Nvidia’s graphics cards, theoretically, have no technological secrets and are easy to replicate. However, building a team from scratch and catching up with the next generation of technology takes time, so the actual moat remains quite wide.
Waves: Once DeepSeek lowered its prices, ByteDance followed suit, which shows that they feel a certain level of threat. How do you view new approaches to competition between startups and big firms?
Liang Wenfeng: Honestly, we don’t really care, because it was just something we did along the way. Providing cloud services isn’t our main goal. Our ultimate goal is still to achieve AGI.
Right now I don’t see any new approaches, but big firms do not have a clear upper hand. Big firms have existing customers, but their cash-flow businesses are also their burden, and this makes them vulnerable to disruption at any time.
Waves: What do you see as the end game of the six other large-model startups?
Liang Wenfeng: Two or three may survive. All of them are in the “burning-money” phase right now, so those with a clear self-positioning and better refinement of operations have a higher chance of making it. Other companies might undergo significant transformations. Things of value won’t simply disappear but will instead take on a different form.
Waves: High-Flyer’s approach to competition has been described as “impervious,” as it pays little attention to horizontal competition. What’s your starting point when it comes to thinking about competition?
Liang Wenfeng: What I often think about is whether something can improve the efficiency of society’s operations, and whether you can find a point of strength within its industrial chain. As long as the ultimate goal is to make society more efficient, it’s valid. Many things in between are just temporary phases, and overly focusing on them can lead to confusion.
Part 4: A group of young people doing “inscrutable” work
Waves: Jack Clark, former policy director at OpenAI and co-founder of Anthropic, said that DeepSeek hired “inscrutable wizards.” What kind of people are behind DeepSeek V2?
Liang Wenfeng: There are no wizards. We are mostly fresh graduates from top universities, PhD candidates in their fourth or fifth year, and some young people who graduated just a few years ago.
Waves: Many LLM companies are obsessed with recruiting talents from overseas, and it’s often said that the top 50 talents in this field might not even be working for Chinese companies. Where are your team members from?
Liang Wenfeng: The team behind the V2 model doesn’t include anyone returning to China from overseas — they are all local. The top 50 experts might not be in China, but perhaps we can train such talents ourselves.
Waves: How did this MLA innovation come about? I heard the idea originated from the personal interest of a young researcher?
Liang Wenfeng: After summarizing some mainstream evolutionary trends of the attention mechanism, he just thought to design an alternative. However, turning the idea into reality was a lengthy process. We formed a team specifically for this and spent months getting it to work. [Jordan: really reminiscent of how Alec Radford’s early contribution to the GPT series and speaks to the broader thesis we’ve argued in the past on ChinaTalk that algorithmic innovation is fundamentally different from pushing the technological frontier in something like semiconductor fabrication. Instead of needing a PhD and years of industry experience to really be useful, you can push the frontier by being a really sharp and hungry 20something (of which China has many!). Dwarkesh’s interview with OpenAI Sholto Douglass and Anthropic’s Trenton Bricken illustrates this dynamic well. Dwarkesh opens with the ine “Noam Brown, who wrote the Diplomacy paper, said this about Sholto: “he's only been in the field for 1.5 years, but people in AI know that he was one of the most important people behind Gemini's success.”]
Waves: The emergence of such divergent thinking seems closely related to your innovation-driven organizational structure. Back in the High-Flyer era, your team rarely assigned goals or tasks from the top down. But AGI involves frontier exploration with much uncertainty — has that led to more management intervention?
Liang Wenfeng: DeepSeek is still entirely bottom-up. We generally don’t predefine roles; instead, the division of labor occurs naturally. Everyone has their own unique journey, and they bring ideas with them, so there’s no need to push anyone. While we explore, if someone sees a problem, they will naturally discuss it with someone else. However, if an idea shows potential, we do allocate resources top-down.
Waves: I heard that DeepSeek is very flexible in mobilizing resources like GPUs and people.
Liang Wenfeng: Anyone on the team can access GPUs or people at any time. If someone has an idea, they can access the training cluster cards anytime without approval. Similarly, since we don’t have hierarchies or separate departments, people can collaborate across teams, as long as there’s mutual interest.
Waves: Such a loose management style relies on having highly self-driven people. I heard you excel at identifying exceptional talent through non-traditional evaluation criteria.
Liang Wenfeng: Our hiring standard has always been passion and curiosity. Many of our team members have unusual experiences, and that is very interesting. Their desire to do research often comes before making money.
Waves: Transformers was born at Google’s AI Lab, and ChatGPT at OpenAI. How do you compare the value of innovations at big companies’ AI labs versus startups?
Liang Wenfeng: Google’s AI Lab, OpenAI, and even Chinese tech companies’ AI labs are all immensely valuable. The fact that OpenAI succeeded was partly due to a few historical coincidences.
Waves: So, is innovation largely a matter of luck? I noticed that the middle row of meeting rooms in your office has doors on both sides that anyone can open. Your colleagues said that this design leaves room for serendipity. The creation of transformers involved someone overhearing a discussion and joining, ultimately turning it into a general framework.
Liang Wenfeng: I believe innovation starts with believing. Why is Silicon Valley so innovative? Because they dare to do things. When ChatGPT came out, the tech community in China lacked confidence in frontier innovation. From investors to big tech, they all thought that the gap was too big and opted to focus on applications instead. But innovation starts with confidence, which we often see more from young people.
Waves: But you don’t fundraise or even speak to the public, so your visibility is lower than those companies actively fundraising. How do you ensure DeepSeek remains the top choice for those working on LLMs?
Liang Wenfeng: Because we’re tackling the hardest problems. Top talents are most drawn to solving the world’s toughest challenges. In fact, top talents in China are underestimated because there’s so little hardcore innovation happening at the societal level, leaving them unrecognized. We’re addressing the hardest problems, which makes us inherently attractive to them.
Waves: When OpenAI’s latest release didn’t bring us GPT5, many people feel that this indicates technological progress is slowing and are starting to question the Scaling Law. What do you think?
Liang Wenfeng: We’re relatively optimistic. Our industry as a whole seems to be meeting expectations. OpenAI is not a god (OpenAI不是神), they won’t necessarily always be at the forefront.
Waves: How long until AGI is realized? Before releasing DeepSeek V2, you had models for math and code generation and also switched from dense models to Mixture of Experts. What are the key points on your AGI roadmap?
Liang Wenfeng: It could be two, five, or ten years–in any case, it will happen in our lifetimes. There’s no unified opinion on a roadmap even within our company. That said, we’ve taken real bets on three directions. First is mathematics and code, second multimodality, and third natural language itself.
Mathematics and code are natural AGI testing grounds, somewhat like Go. They’re closed, verifiable systems where high levels of intelligence can be self-taught. Multimodality and engagement with the real human world, on the other hand, might also be a requirement for AGI. We remain open to different possibilities.
Waves: What do you think is the end game for large models?
Liang Wenfeng: There will be specialized companies providing foundation models and services, achieving extensive specialization in every node of the supply chain. More people will build on top of all of this to meet society’s diverse needs.
Part 5: All the methods are products of a previous generation
Waves: Over the past year, there have been many changes in China's large model startups. For example, Wang Huiwen [co-founder of RenRen, a facebook clone, and Meituan, a food delivery company], who was very active at the beginning of last year, withdrew midway, and companies that joined later began to show differentiation.
Liang Wenfeng: Wang Huiwen bore all the losses himself, allowing others to withdraw unscathed. He made a choice that was worst for himself but good for everyone else, so he's very decent in his conduct - this is something I really admire. [Wang Huiyuan founded foundation model company 光年之外 Lightyear only to quickly fold it back into Meituan. For more on Meituan and AI, see this recent 36Kr feature].
Waves: Where are you focusing most of your energy now?
Liang Wenfeng: My main energy is focused on researching the next generation of large models. There are still many unsolved problems.
Waves: Other large model startups are insisting on pursuing both [technology and commercialization], after all, technology won't bring permanent leadership as it's also important to capitalize on a window of opportunity to translate technological advantages into products. Is DeepSeek daring to focus on model research because its model capabilities aren't sufficient yet?
Liang Wenfeng: All these business patterns are products of the previous generation and may not hold true in the future. Using Internet business logic to discuss future AI profit models is like discussing General Electric and Coca-Cola when Pony Ma was starting his business. It’s a pointless exercise (刻舟求剑).
Waves: In the past, your quant fund High-Flyer had a strong foundation in technology and innovation, and its growth was relatively smooth. Is this the reason for your optimism?
Liang Wenfeng: In some ways, High-Flyer strengthened our confidence in technology-driven innovation, but it wasn't all smooth sailing. We went through a long accumulation process. What outsiders see is the part of High-Flyer after 2015, but in fact, we've been at it for 16 years.
Waves: Returning to the topic of innovation. Now that the economy is starting to decline and capital is no longer as loose as it was, will this suppress basic research?
Liang Wenfeng: I don't necessarily think so. The adjustment of China's industrial structure will necessarily rely more on hardcore technological innovation. When people realize that making quick money in the past was likely due to lucky windows, they'll be more willing to humble themselves and engage in genuine innovation.
An Yong: So you're optimistic about this as well?
Liang Wenfeng: I grew up in the 1980s in a fifth-tier city in Guangdong. My father was a primary school teacher. In the 1990s, there were many opportunities to make money in Guangdong. At that time, many parents came to my home; basically, they thought studying was useless. But looking back now, they’ve all changed their views. Because making money isn't easy anymore—even the opportunity to drive a taxi might be gone soon. It’s only taken one generation.
In the future, hardcore innovation will become increasingly common. It’s not easy to understand right now, because society as a whole needs to be educated on this point. Once society allows people dedicated to hardcore innovation to achieve fame and fortune, then our collective mindset will adapt. We just need some examples and a process
ChinaTalk is a reader-supported publication. To receive new posts and support our work, consider becoming a free or paid subscriber.
Our article was originally published in Asterisk Magazine. Today,ChinaTalk is rereleasing it alongside exclusive commentary from Jason Matheny, CEO of RAND at the end of the post.
RAND’s halcyon days lasted two decades, during which the corporation produced some of the most influential developments in science and American foreign policy.
Today, RAND remains a successful think tank — by some metrics, among the world’s best.1 In 2022, it brought in over $350 million in revenue, and large proportions still come from contracts with the US military. Its graduate school is among the largest for public policy in America.
But RAND’s modern achievements don’t capture the same fundamental policy mindshare as they once did. Its military reports may remain influential, but they hold much less of their early sway, as when they forced the U.S. Air Force to rethink several crucial assumptions in defense policy. And RAND’s fundamental research programs in science and technology have mostly stopped. Gone are the days when one could look to U.S. foreign policy or fundamental scientific breakthroughs and trace their development directly back to RAND.
How was magic made in Santa Monica? And why did it stop?
The Roots of RAND
Economists, physicists, and statisticians — civilian scientists to that point not traditionally valued by the military — first proved their utility in the late stages of World War II operational planning. American bomber units needed to improve their efficiency over long distances in the Pacific theater. The scientists hired by the Army Air Force proposed what at the time seemed a radical solution: removing the B-29 bomber’s armor to reduce weight and increase speed. This ran counter to USAAF doctrine, which assumed that an unprotected plane would be vulnerable to Japanese air attacks. The doctrine proved incorrect. The increased speed not only led to greater efficiency, it also led to more U.S. planes returning safely from missions, as Japanese planes and air defense systems were unable to keep up.2 Civilian scientists were suddenly in demand. By the end of the war, all USAAF units had built out their own operations research departments to optimize battle strategy. When the war ended, the question turned to how to retain the scientific brain trust it had helped to assemble.
General Henry “Hap” Arnold, who had led the Army Air Force’s expansion into the most formidable air force in the world, had started to consider this question long before the war had ended. He found an answer in September 1945, when Franklin Collbohm, a former test pilot and executive at Douglas Aircraft, walked into Arnold’s office with a plan: a military-focused think tank staffed by the sharpest civilian scientists. Collbohm did not have to finish describing his idea before Arnold jumped and agreed. Project RAND was born.
Arnold, along with General Curtis LeMay — famous for his “strategic bombing” of Japan, which killed hundreds of thousands of civilians — scrounged up $10 million from unspent war funds to provide the project’s seed money, which was soon supplemented with a grant from the Ford Foundation. This put RAND into a privileged position for a research organization: stably funded.
On top of that financial stability, RAND built what would become one of its greatest organizational strengths: a legendarily effective culture, and a workforce to match it.
In an internal memo, Bruno Augestein, a mathematician and physicist whose research on ballistic missiles helped usher in the missile age, highlighted a set of factors that catalyzed RAND’s early success. In short: RAND had the best and brightest people working with the best computing resources in an environment that celebrated excellence, welcomed individual quirks, and dispensed with micromanagement and red tape.
Early RAND leadership was, above all else, committed to bringing in top talent and jealously guarded the sort of intellectual independence to which their academic hires were accustomed. Taking the mathematics department as an example, RAND hired John Williams, Ted Harris, and Ed Quade to run it. While these were accomplished mathematicians in their own right, these three were also able to attract superlative talents to work under and around them. As Alex Abella writes in Soldiers of Reason, his history of RAND, “No test for ideological correctness was given to join, but then none was needed. The nation’s best and brightest joining RAND knew what they were signing on for, and readily accepted the vision of a rational world — America and its Western allies — engaged in a life-and-death struggle with the forces of darkness: the USSR.”
As the Cold War intensified, the mission became the sell. The aim of RAND, as the historian David Hounshell has it, “was nothing short of the salvation of the human race.”3 The researchers attracted to that project believed that the only environment in which that aim could be realized was independent of the Air Force, its conventional wisdom, and — in particular — its conventional disciplinary boundaries
RAND’s earliest research aligned with the USAF’s (the Army Air Force had become its own service branch in 1947) initial vision: research in the hard sciences to attack problems like satellite launches and nuclear-powered jets.4 However, the mathematician John Davis Williams, Collbohm’s fifth hire, was convinced that RAND needed a wider breadth of disciplines to support the Air Force’s strategic thinking. He made the case to General LeMay, who supervised RAND, that the project needed “every facet of human knowledge to apply to problems.”5 To that end, he argued for recruiting economists, political scientists, and every other kind of social scientist. LeMay, once convinced, implored Williams to hire whoever it took to get the analysis right.
And so they did. RAND’s leadership invested heavily in recruiting the best established and emerging talent in academia. An invitation-only conference organized by Williams in New York in 1947 brought together top political scientists (Bernard Brodie), anthropologists (Margaret Mead), economists (Charles Hitch), sociologists (Hans Speier), and even a screenwriter (Leo Rosten). The promise of influence, exciting interdisciplinary research, and complete intellectual freedom drew many of the attendees to sign up.
Within two years, RAND had assembled 200 of America’s leading academics. The top end of RAND talent was (and would become) full of past (and future) Nobel winners, and Williams worked around many constraints — and eccentricities — to bring them on. For instance, RAND signed a contract with John von Neumann to produce a general theory of war, to be completed during a small slice of his time: that spent shaving. For his shaving thoughts, von Neumann received $200 a month, an average salary at the time.
Beyond the biggest names, RAND was “deliberate, vigorous, and proactive” in recruiting the “first-rate and youthful staff” that made up most of its workforce. The average age of staff in 1950 was under 30.6 Competition between them helped drive the culture of excellence. Essays and working papers were passed around for comments, which were copious — and combative. New ideas had to pass “murder boards.” And the competition spilled into recreational life: Employees held tennis tournaments and boating competitions. James Drake, an aeronautical engineer, invented the sport of windsurfing. The wives of RAND employees — who were, with a few notable exceptions, almost all male — even competed through a cooking club where they tried to make the most "exotic" recipes.
After bringing in such extraordinary talent, RAND’s leadership trusted them to largely self-organize.Department heads were given a budget and were free to spend it as they felt fit. They had control over personnel decisions, which allowed them the flexibility to attract and afford top talent. As a self-styled “university without students,” RAND researchers were affiliated with departments with clear disciplinary boundaries, which facilitated the movement of researchers between RAND and academia. But in practice, both departments and projects were organized along interdisciplinary lines.
The mathematics department brought on an anthropologist. The aeronautics department hired an MD. This hiring strategy paid off in surprising ways. For instance, while modeling the flow of drugs in the bloodstream, a group of mathematicians stumbled upon a technique to solve a certain class of differential equations that came to be used in understanding the trajectory of intercontinental ballistic missiles.
The caption of this image, from the May 11, 1959, issue of Life magazine, reads: ‘After-hours workers from RAND meet in home of Albert Wohlstetter (foreground), leader of RAND’s general war studies. They are economists gathered to discuss study involving economic recovery of U.S. after an all-out war.’ Leonard McCombe / The LIFE Picture Collection via Getty Images.
Finding an Institutional Footing
RAND was at the forefront of a postwar explosion in federal funding for science. Hundreds of millions of dollars poured into universities, think tanks, and industrial R&D labs. Almost all of it was directed toward one purpose: maintaining military superiority over the Soviet Union. In 1950, over 90% of the federal research budget came from just two agencies: the Atomic Energy Commission and the Department of Defense.7 Significant portions of this funding went toward basic research with no immediate military applications.8 Vannevar Bush, the influential head of the war-era Office of Scientific Research and Development, had argued for this approach in his 1945 book Science, the Endless Frontier: Freeing up scientists to follow their own research interests would inevitably lead to more innovation and ensure American technological dominance. Bush’s was not the only, or even the dominant, view of how postwar science should be organized — most science funding still went toward applied research — but his views helped inform the organization of a growing number of research institutions.9 No organization embodied this model more than RAND. Air Force contracts were the financial backbone of the organization. They provided the money required to run RAND, while profits were used to fund basic research. In the 1950s, USAF contracts comprised 56% of RAND’s work, while other sponsors made up just 7%.10 That left more than a third of RAND’s capacity open to pursue its own agenda in basic research. Many of the developments made there would be used in their applied research, making it stronger — and more profitable — in the process. This flywheel would become critical to RAND’s success.
Not all of these developments were successful, especially at first. RAND’s early research efforts in systems analysis — an ambitious pursuit in applying mathematical modeling that RANDites were optimistic could produce a holistic “science of warfare” — were flops. The first project, which aimed to optimize a strategic bombing plan on the Soviet Union, used linear programming, state-of-the-art computing, and featured no fewer than 400,000 different configurations of bombs and bombers. It proved of little use to war planners. Its assumptions fell prey to the “specification problem:” trying to optimize one thing, in this case, calculating the most damage for the least cost led to misleading and simplistic conclusions.11
But RAND would soon find its footing, and a follow up to this work became a classic of the age. The 1954 paper Selection and Use of Strategic Air Bases proved the value of RAND’s interdisciplinary approach — though its conclusions were at first controversial. Up to the 1950s, there had been little analysis of how the Strategic Air Command, responsible for the United States’s long range bomber and nuclear deterrent forces, should use its Air Force bases. At the time, the SAC had 32 bases across Europe and Asia. The study, led by political scientist Albert Wohlstetter, found that the SAC was dangerously vulnerable to a surprise Soviet attack. The SAC’s radar defenses wouldn’t be able to detect low-flying Soviet bombers, which could reduce American bombers to ash — and thereby neutralize any threat of retaliation — before the Americans had a chance to react. Wohlstetter’s study recommended that the SAC keep its bombers in the U.S., dispersed at several locations to avoid concentration at any place.
LeMay, RAND’s original benefactor and commander of the SAC, resisted Wohlstetter’s conclusions. He worried the plan would reduce his control over the country’s nuclear fleet: With the SAC based in the U.S., LeMay would have to cede some authority to the rest of the U.S. Air Force. He pushed against it many times, proposing several alternatives in which the SAC kept control over the bombers, but no plan fully addressed the vulnerabilities identified by the report.
Undaunted — and sure of his logic — Wohlstetter pushed his conclusions even further. He proposed a fail-safe mechanism, where nuclear bombers would have to receive confirmation of their attack from multiple checkpoints along the way, to prevent rogue or mistaken orders from being followed. Wohlstetter went around LeMay, to Defense Secretary Charles Wilson and General Nathan Twining, chairman of the Joint Chiefs of Staff, who ultimately accepted the study’s recommendations in full. It took over two decades, but they proved their value in 1980 when a faulty chip erroneously warned of an impending Soviet strike. While no order for a retaliatory attack was issued, had there been one, the fail-safe mechanism would have prevented the bombers from actually attacking the USSR. Selection and Use of Strategic Air Bases was a triumph for RAND. Not only had they provided correct advice to the USAF, they had also proved their independence from the institution’s internal politics.
And the flywheel would prove its value many times over. RAND’s basic research helped drive the development and strategy of ICBMs, the launch of the first meteorological satellite, and, later, on cost reductions in ICBM launch systems.
Diversification and Decline
RAND’s conclusions ran counter to USAF doctrine several times — and each time RAND fought to maintain its independence. When the USAF commissioned RAND to study the Navy’s Polaris program — in order to show that it was inferior to the Air Force’s bombers for nuclear weapon delivery — RAND found that the Polaris missiles were, in fact, superior. The same happened with another study, which challenged the effectiveness of the B-70 bomber in 1959.
Over time, however, these tensions added friction to the relationship. To make matters worse, between 1955 and 1960, the USAF’s budget declined in both absolute terms, and relative to the rest of the defense community. In 1959, the Air Force froze RAND’s budget, presumably due to the budget cuts — and their disputes with RAND.
This situation was not unique to the USAF, or to RAND. As the 1950s rolled into the ’60s, scientists at civilian institutions increasingly moved to disentangle themselves from their military benefactors. Throughout the decade, DOD funding for basic research would only continue to decline.12
RAND weathered the transition by successfully seeking out new customers — the AEC, ARPA, the Office of the Comptroller, the Office of the Assistant Secretary of Defense for International Security Affairs (ISA), NASA, the NSF, the NIH, and the Ford Foundation, to name a few. The percent of the outside funding coming from the USAF dropped from 95% when RAND started to 68% in 1959.13 But their success came at a cost: This diversification is what led to RAND losing its edge in producing the cutting edge of policy and applied science.
Funding diversification reshaped both RAND’s culture and output. The increased number of clients made scheduling researchers’ work harder. Each client expected a different standard of work, and the tolerance levels for RAND’s previously freewheeling style varied. The transaction costs of starting a new contract were much higher and the flexible staffing protocols that had worked for the USAF in the 1950s needed to be systematized. The larger organization led to ballooning internal administration expenses.
Along with all of this, RAND’s increased size attracted more political detractors. In 1958, a RAND paper called Strategic Surrender, which examined the historical conditions for surrender, had generated a political firestorm. Politicians were furious with RAND for exploring conditions under which it would be strategic for the U.S. to surrender. Senators weren’t particularly interested in the study itself, but those who wanted to run for president (like Stuart Symington of Missouri) used it as evidence that the Eisenhower administration was weak on defense.
The Senate even passed a resolution (with an 88–2 margin) prohibiting the use of federal funds for studying U.S. surrender. RAND’s management, realizing that an intentional misinterpretation of their work potentially threatened future funding streams, now had to consider the wider domestic political context of their work. All of these factors changed RAND’s culture from one that encouraged innovation and individuality to one that sapped creativity.
But the biggest change was yet to come. In 1961, Robert McNamara took over the Department of Defense and brought with him a group of RAND scholars, commonly called the “Whiz Kids.” Their most important long-term contribution to U.S. governance was the Planning-Programming-Budgeting System. PPBS took a Randian approach to resource allocation, namely, modeling the most cost-effective ways to achieve desired outcomes. In 1965, after President Johnson faced criticism for poor targeting of his Great Society spending, he required nearly all executive agencies to adopt PPBS. Many RAND alumni were hired by McNamara and his team to help with the Great Society’s budgeting process.
In 1965, Henry Loomis, the deputy commissioner on education, approached RAND about conducting research on teaching techniques. Franklin Collbohm, RAND’s founder and then president, declined. He preferred that RAND stay within the realm of military analysis. RAND’s board disagreed and would eventually push Collbohm out of RAND in 1967. The board thought it was time for a change in leadership — and to RAND’s nonmilitary portfolio.
The entry of a new president, Henry S. Rowen, an economist who had started his career at RAND, cemented this change. By 1972, the last year of Rowen’s tenure, almost half of all RAND projects were related to social science. For better or worse, this eroded RAND’s ability to take on cutting-edge scientific research and development.
RAND entered domestic policy research with a splash — or, rather, a belly flop. The politics of social policy research were markedly different from working with the DOD. For one, there were substantially more stakeholders — and they were more vocal about voicing their disagreements. One crucial example is when RAND proposed police reforms in New York City, but pressure from the police unions forced them to retract.
John Lindsay, the Republican mayor of New York, had tasked RAND with improving the New York Police Department, which had recently been implicated in narcotics scams, corruption, and police brutality. The report showed that in less than 5% of the cases in which an officer was charged with a crime or abusing a citizen did the officers receive anything more than a reprimand. The findings were leaked to The New York Times, which added to the impression among the police that RAND was the mayor’s mouthpiece.
RAND, for the first time, had to face the reality of local politics: a sometimes hostile environment, multiple stakeholders who sometimes acted in bad faith, and none of the free reign that characterized their first decades. RAND’s experience with the police report, and the controversy over the study of surrender, led RAND to be more conservative about the research it put out. And additionally, the focus on policy research crowded out the scientific research.
For example, beginning in the 1970s, RAND’s applied mathematics research output slowed to a trickle, before stopping altogether in the 1990s. It was replaced by mathematics education policy. The same is true for physics, chemistry, and astronomy. Another emblematic development in the dilution of RAND’s focus was the founding in 1970 of the Pardee RAND Graduate School, the nation’s first Ph.D.-granting program in policy analysis. While the idea of training the next generation in RAND techniques is admirable, RAND in the early years explicitly defined itself as a “university without students.”
RAND is still an impressive organization. It continues to produce successful policy research, which commands the eyes of policymakers in over 82 federal organizations and across dozens of local and even foreign governments. Still, their work today is inarguably less groundbreaking and innovative than it was in the ’50s. This relative decline was partially caused by internal policy choices, and partially by the eventual loss of their initial team of leading scientists. But part of it was also inevitable: We no longer live in an era when branches of the U.S. military can cut massive blank checks to think tanks in the interest of beating the Soviets. The successes of 1950s RAND do come with lessons for modern research organizations — about the importance of talent, the relevance of institutional culture, and the possibilities of intellectual freedom — but the particular conditions that created them can’t be replicated. It is remarkable that they existed at all.
Modern Magic at RAND
The following commentary comes directly from RAND’s CEO, Jason Matheny.
RAND CEO Jason Matheny here. Your readers may recall from my appearance on your podcast last year that I, too, am a RAND history nerd. There are many great details in your Asterisk article about RAND’s early contributions in the 1950s and ‘60s. Thanks for bringing them to life.
RAND’s contributions in the last five decades have been no less consequential. The world’s challenges are certainly different from the ones RAND researchers confronted in the early years. But it is RAND’s ability to reorient itself toward the biggest challenges that has been our “magic.” We shouldn’t expect or want RAND to look the same as it did during the Cold War.
I thought your readers would be interested to pick up where your story stops. And since your article focuses on national security, I’ll concentrate my comments there. (That said, there have been just as many breakthroughs in RAND’s social and economic policy analyses over the years.)
RAND’s security research in the modern era has been forward-looking, has challenged long-held wisdom, and has anticipated once-unthinkable threats. And I’m not saying this only as RAND’s CEO. Before I joined RAND two years ago, I was one of countless people at the White House and elsewhere in government who relied on RAND analysis to make critical decisions.
Many recent RAND studies will remain classified for years. While their full impact will be assessed with time — much as was the case with RAND’s work in the 1950s and 1960s – they have been among RAND’s most influential. Below are some examples of projects that we can describe here:
Russia: RAND was among the first organizations to identify Russia’s growing military capabilities following its 2008 war in Georgia and the threat these posed to new NATO members in the Baltic states. This work prompted important planning and infrastructure changes that are being used today to support Ukraine.
U.S. military power: RAND’s series of overmatchstudies transformed policymakers’ understanding the loss of U.S. military superiority in key areas over time.
Operating in the Pacific theater: RAND was among the first to highlight the vulnerability of the U.S. military’s forward infrastructure in the Pacific and ways to overcome that vulnerability.
Nuclear strategy: RAND’s recent work on nuclear deterrence, including wargames analyzing nuclear-armed regional adversaries, brought about a resurgence of deterrence thinking within the government.
B-21: RAND analysis of penetrating versus standoff bomber capabilities led directly to the decision to establish the B-21 program.
Military forces: RAND‘s work on military personnel, the ability to develop and sustain the all-volunteer force over time, appropriate pay and benefits for the force, and the vulnerabilities to service members and their families, has been the primary source of analysis for decisionmakers within the Department of Defense and Congress.
Drones: RAND’s analysis of small UAVs and swarming options was the first to analyze how a sensor grid can substantially strengthen deterrence in the Asia-Pacific region. Current DoD programs can be traced directly to this pathbreaking analysis.
PTSD and TBI: RAND’s work on the invisible wounds of war, PTSD, and traumatic brain injury, was the first careful documentation of psychological and cognitive injuries from modern combat. This work launched a society-wide effort to detect and treat such injuries.
Logistics: RAND analysis prompted the revolution in combat logistics in both the Air Force and the Army, emphasizing wartime flexibility and resilience as the organizing principles for supply and maintenance.
AI: RAND was early in systematically evaluating how defense organizations could integrate contemporary AI methods based on deep learning, in evaluating large language models, and in assessing threats to model security.
With rapid developments in emerging technology and an increasingly confrontational PRC government, the world needs RAND’s analysis more than ever. I know that your readers care deeply about these challenges. Those who want to work toward solutions should consider working at RAND or applying to our new master's degree program in national security policy.
To hear more from Jason, check out the two-hour interview we did last year on ChinaTalk, which was my favorite episode of 2023.
Similar stories of outsiders applying quantitative thinking improving performance also played out in other branches. Such experiences were also seen in the Navy, where better usage of anti-submarine depth charges led to higher efficiency to the extent that German naval planners thought that the U.S. had invented a new type of depth charge.
D. Hounshell, “The Cold War, RAND, and the Generation of Knowledge, 1946–1962,” Historical Studies in the Physical and Biological Sciences 27, no. 2(1997): 237–267.
Dan Kevles, “Cold War and Hot Physics: Science, Security, and the American State, 1945–1956,” Historical Studies in the Physical and Biological Sciences 20, no. 2 (1990): 244.
Audra J. Wolfe, Competing with the Soviets: Science, Technology, and the State in Cold War America (Baltimore: The Johns Hopkins University Press, 2013), 36.
D. Hounshell, “The Cold War, RAND, and the Generation of Knowledge, 1946–1962,” Historical Studies in the Physical and Biological Sciences 27, no. 2(1997): 237–267.
Mark Witzke is a China analyst and nonresident scholar at the UC San Diego 21st Century China Center. See more on US-China relations and unexpected connections between the countries like the UFC on his Twitter or Bluesky @mkwitzke.
As Trump’s return to the White House draws nearer, China watchers are paying close attention to his cabinet picks. One underrated wild card in the relationship might actually be the interests of a sports league once condemned by Republicans as “human cockfighting”.
This past weekend, President-elect Trump walked out in MSG in New York in a moment of triumph to attend a mixed martial arts (MMA) event. It was his first public appearance since the election and served as a sort of coronation ceremony where popular MMA fighters like Jon Jones and Michael Chandler gave tribute to the new President.
Formerly a foe of the Republican party due to John McCain’s staunch opposition to the league, the UFC now appears to be a MAGA PR arm. Trump has frequently appeared at UFC events in times of intense political pressure. After January 6th, Trump made one of his first major public appearances at UFC 264. Days after his arrest, he appeared defiantly at UFC 287 with Mike Tyson. Trump used the UFC 302 event to launch his TikTok account, just two days after receiving a guilty verdict in his hush money trial.
Trump’s relationship to the UFC goes all the way back to 2001, when the UFC was relegated to backwater venues and struggled to dispel the notion that it was too dangerous to be legal. At that time, according to sports journalist Karim Zidan, “Donald Trump took a chance on the UFC in and allowed the organization to host two consecutive events at his Atlantic City casino, the Trump Taj Mahal. And the UFC has really been loyal to him ever since then.” At all three of his nominations, Dana White spoke for Trump at the RNC and has helped him make inroads with new media and popular fighters.
What might Dana White expect in return? For starters, easing visa restrictions on fighters with links to controversial figures like Ramzan Kadyrov. But White, with his global operations, surely has his eye on larger payouts. Prior to the pandemic, the UFC had big plans for China. They held events in Macau in 2012 and 2014 and revved up activity in the mainland a few years later with events held in Beijing, Shanghai, and Shenzhen in 2017, 2018, and 2019.
While the earlier events focused more on foreign fighters, the 2019 event in Shenzhen was headlined by Hebei native, Zhang Weili, where she became the UFC’s first female Asian champion. Upon her victory she took the mic and spoke with her limited English, "My name is Zhang Weili!...I'm from China. Remember me!" She has since gone on to lose and then regain her champion status, most recently defending her belt at UFC 300 against her countryman, Yan Xiaonan. Chinese state media spoke approvingly of the event and noted the growing interest towards the sport in China.
Other moves taken by the UFC to grow their market in China included establishing a training center in Shanghai in 2019, inking streaming content distribution deals with Migu (a part of China Mobile), and cultivating talent through its Road to UFC program that gives local fighters the chance to make it to the UFC. The UFC even signed a deal with the Chinese Olympic Committee to help train athletes. In an interview, Kevin Chang, head of UFC Asia, said that China was a priority for the UFC, that they thought carefully about the differences in promoting on Chinese social media platforms versus in the US and they had already gained millions of fans.
But since the pandemic, there have been no major UFC events in China — an attempt to hold an event in Shanghai at the end of last year was abruptly canceled for no apparent reason. This Saturday, however, the UFC will return to China (Macau) and hold their first event there since 2019, marking its return to a market with immense growth potential. While other companies are trying to figure out how to pull out and decouple, Trump’s favorite sport league will have a continued interest in smooth relations between the US and China.
This sports exchange may bring to mind the old “ping pong diplomacy” where in the early 1970s, an international exchange of table tennis players helped open the door for a renewal of US-China relations. In a new era for the US and China where the relationship will almost certainly get stormy, could a sports league where people punch each other in the face serve as an unexpected circuit-breaker?
It may seem silly, but so was ping pong diplomacy. Perhaps Chinese officials make an appearance at a US event or push for more mainland events in an attempt to appeal to Trump. Xi Jinping himself has long had a personal interest in sports, with plans to make China a sporting superpower and a desire to make Chinese men more “manly.” See ChinaTalk’s feature on the CCP masculinity crisis.
Dana White and the UFC might not have a particular interest in all the nuts and bolts of Trump’s China policy with regards to tariffs or investment screening, but he certainly would be loath to see total decoupling of US business interests from China. Dana, along with Elon Musk, Jeff Yass, and Howard Lutnick may well serve as the second term’s version of Gary Cohn and Steve Mnuchin.
Tweets of the Week
ChinaTalk is a reader-supported publication. To receive new posts and support our work, consider becoming a free or paid subscriber.
Few books have influenced me as much as the Makers of Modern Strategy series. The three volumes (published in 1942, 1986, and 2023) are indispensable to understanding statecraft, leadership, and the evolution of warfare across millennia.
The man behind this behemoth collection is Hal Brands, a professor at the Johns Hopkins School of Advanced International Studies and a returning ChinaTalk guest.
In our conversation, we discuss:
The process for compiling such an ambitious collection of essays;
Unique insights and new topics covered in the 2023 edition, including Tecumseh, Kabila in the Congo, and Strategies of Equilibrium in 17th Century France;
Advice for reading the book effectively;
Revolutions in military affairs, from the atom bomb to quantum computers.
For reference, you can compare the content of the three volumes with this spreadsheet, courtesy of Nicholas Welch.
Jordan Schneider: Hal, thanks so much for producing this book.
Looking back at the process, which essays were you the most excited to publish?
Hal Brands: That’s a little bit like asking me which of my children I’d prefer to keep. They are all beautiful and they are all my favorites. There were maybe a handful that are worth mentioning just because, from the beginning of the project, I thought they were going to be really cool.
The first substantive chapter is an essay by Sir Lawrence Freedman, the great British scholar [Ed. Coming soon to ChinaTalk!]. He is the only scholar who wrote an essay in the 1986 version and in the 2023 version. His essay is called, “Strategy: A History of an Idea.”
It illustrates how definitions of the terms “strategy” and “strategist” have changed over time. I had Freedman in mind when brainstorming ideal authors to write that essay, and I was just delighted that he could do it.
Another interesting angle on a classic subject is Hew Strachan’s essay on Clausewitz. Carl von Clausewitz has been a recurring character in both the 1943 and 1986 editions of the book. He looms over the field of strategic studies.
But Strachan’s interpretation is basically that everybody gets Clausewitz wrong. Michael Howard’s translation of Clausewitz — which all of us professional nerds have read and relied on — is actually a distortion of Clausewitz’s argument about the relationship between war and politics.
When you get Hew on board to do an essay like this, you know he’s going to say something profound and you know he’s going to say something original. I was even a little surprised by just how jarring that reinterpretation was. It’s really going to make a splash.
Jordan Schneider: I’ll shout out two more essays that I really enjoyed — one was the Tecumseh essay. For our non-American listeners, Tecumseh was a Native American leader and war hero who banded during the War of 1812 between the US and the UK. He came pretty close to beating the US and shutting down Western expansion.
Tecumseh pulled together a larger fighting force than any other American Indian chief in history, creating a twelve-hundred-mile barricade to limit westward expansion of the United States…
[H]e propagated centripetal religious beliefs that advanced political power within tribes and encouraged accession to the Confederacy; he used social suasion to reduce reliance on colonial-produced goods; he won foreign economic support that freed up fighters for military campaigns; he secured consequential European military involvement; and he produced an organized military force capable of defeating the US militarily.
The Shawnee Confederacy threat precipitated the doubling of the size of the US military, and the Confederacy imposed the largest combat losses the US had known to that point.
The United States government defeated this elegant strategy not on the battlefield, but economically.
[The New Makers of Modern Strategy, pp. 369-370]
The other essay that completely blew my mind was Jason Stearns’ “Strategies of Persistent Conflict,” which looked at the Congo wars. The logics of persistent and brutal conflict is different from the military strategies described by Clausewitz and Jomini.
Having those modern wrinkles added to the canon was really interesting.
Hal Brands: Those were two of the most original essays in the volume, both in terms of the subjects and also how they compel us to rethink strategy. The thesis of Kori Schake’s essay is that Tecumseh was really a practitioner of what we would consider an all-of-society strategy. As you mentioned, he came close to succeeding.
Depiction of Shawnee chief Tecumseh facing off with William Henry Harrison (the governor of the Indiana Territory) during negotiations over the sale of tribal lands in 1811. Source.
Jason Stearns wrote the essay on wars in Africa. That one is so interesting because it turns the traditional Clausewitzian paradigm on its head, pointing out that protracting the war can be a form of strategy. Not in the Fabian sense of trying to exhaust your enemy and then defeat him, but in the sense that the war may actually be profitable for the groups undertaking it — continuation of the war can itself be a strategy.
[I]n the Democratic Republic of the Congo (DRC), as well as in other weak states…waging war becomes both a lifestyle and a fundamental tool of political survival, providing a means of managing dissent and doling out patronage…
It was during this stalemate of [The Second Congo War] that… [t]he assorted belligerents became deeply invested in various forms of economic activity, a blend of racketeering, extortion, and taxation. …
Following the blueprint for United Nations peace processes at the time, diplomats pushed for peace talks, which they hoped would be followed by a power-sharing agreement and the reunification of the country. …
Some scholars go so far to argue that the penchant for power-sharing agreements by Western donors has inadvertently incentivized rebellions by making them an acceptable path to power and lowering the cost of insurgency. While this finding is contested, it is clear that international norms against protracted military conflict have made it more difficult to achieve military victories. …
The war made large-scale agricultural production almost impossible, cutting off trade routes to the rest of the country, pillaging livestock, and preventing investment. The economy became increasingly focused on the mining sector, which in turn became extremely militarized. Meanwhile, employment opportunities shrank for the youth, making armed insurgency more attractive.
[The New Makers of Modern Strategy, pp. 1048-1050]
Jordan Schneider: There’s something that’s so dark about this entire book. I catch myself getting excited as I flip through this book trying to choose which essay to read. “Maybe it’s time for Napoleon. Maybe it’s time for nuclear war.”
These are exciting topics, but it’s also tragic that as a species we have spent so many thousands of years innovating new ways to kill our fellow humans. Do you have any thoughts on that?
Laurent-Désiré Kabila at the Kigoma base in Tanzania, mid 1960s. The Kigoma location allowed Kabila’s rebels to import arms from China and letters from Che Guevara through the Port of Dar es Salaam. Kabila is in the centre in the middle row, right above the little boy. Source: The Individual in African History,p. 282
Hal Brands: You put your finger on an important point, which is that the content of strategy changes in different eras as different technologies and different challenges emerge. The basic practice — the nature of strategy — doesn’t change that much over time. It’s really about trying to use the means at your disposal to achieve whatever aims you seek, in the face of all the resistance and chaos of the world.
Even though it’s something that exists in peacetime as well as wartime, we most frequently pay attention to strategy when the stakes are high. That’s typically when violent conflict is either happening or threatening to happen. There is an inherently dark nature to the subject material.
As I often point out in my writings or when I’m talking to students — strategy is itself a very optimistic endeavor, because the idea of it is that you can impose a certain purpose on events rather than simply being tossed around by them. You can use power in purposeful and coherent ways. That’s the enduring challenge of strategy, and that’s the thing that pretty much everybody featured in this volume was wrestling with in one way or another.
The History of Security Studies
Jordan Schneider: Let’s take a step back then and look at this concept of security studies. Do you want to talk about the origins of this as a pseudo-discipline and how the first book ended up coming together in 1943?
Hal Brands: Absolutely. I’ll make a point here that’s a little deep into the academic weeds, which is — there’s often said to be a difference between strategic studies and security studies.
Strategic studies, to put it bluntly, is the study of political-military issues. It’s got a somewhat narrower thrust. Security studies can encompass all sorts of things. There are various different types of security. It can deal with climate, it can deal with human security, it can deal with a whole range of issues. It’s typically thought of as a somewhat more capacious discipline.
In my mind, they’re very closely related. People who are involved in one camp or the other will say they are different things. That distinction is just worth mentioning for CYA on my part.
The development of Makers of Modern Strategy is inseparable from the emergence of strategic studies and security studies as related fields in the United States. The first volume of Makers, as you mentioned, was published in 1943, it was started a couple of years before that. The editor was a guy named Edward Mead Earle, who was at the Institute for Advanced Study in Princeton, New Jersey. He had really been involved in a rethinking of the requirements of national security as the world fell apart in the 1930s.
He was a big proponent of the idea that the United States needed a more coherent approach to grand strategy, bringing together all the different forms of national power to deal with all of the threats — economic, military, ideological — that emerged as fascist regimes gained the ascendancy during the 1930s.
He was motivated to pull the book together by the idea that the United States was henceforth going to be far more deeply and far more consistently involved in global affairs than it had been in the past. The American people — not just national security elites but just educated men or women on the street — needed a deeper understanding of military affairs and strategic affairs more broadly if the United States was going to have the educated citizenry it needed to be effective in this era.
That was the goal of the first volume. It developed in parallel to the emergence of strategic studies research and teaching programs. It was part of the development of the intellectual sinews of the American superpower during the late 1930s and 1940s. It was a smash hit.
Jordan Schneider: The idea for this book was great from the start, but it took money to fund the professorships and create the conferences to entice more academics into grappling with those questions of national power and grand strategy.
The book and the broader thinking that this field generated, which ended up informing a lot of how America has engaged with the world for the past 75 years, wouldn’t have happened without that initial academic seed funding sorts to allow people to research and write along the lines that he initially laid out.
Hal Brands: That’s exactly right. Ideas may be cheap, but good ideas aren’t cheap. Developing a cadre of intellectuals who are going to work on major research projects — that takes money. The emergence of strategic studies as a field was led by Carnegie and a couple of other foundations and philanthropic entities.
Then, of course, the field really develops in the context of World War II and the Cold War, when also the US government is throwing more money at these areas than ever before — including by funding the RAND Corporation.
You wouldn’t have gotten security studies or strategic studies as fields in the United States without the collapse of the international system in the 1930s, the interest that spurs in these sorts of issues and then the investments that philanthropic entities in the US government make in it over the decades to follow.
Jordan Schneider: Let’s stay on this 1943 book. It is a fascinating document, because it’s literally in the middle of the war. You have essays by Earle talking about Hitler’s strategy and he’s like, “Yeah, we think they’re going to lose, but we’re not sure.”
There’s another essay about Stalin where the author is like, “Yeah, we’ll see about this spring offensive.”
There seems to be a lot of personality in the authors where they can show their prejudices on their sleeve. Everyone’s just making fun of Erich Ludendorff for being an idiot. Looking at that book, what stuck out to you about those essays?
Hal Brands: As you mentioned, a lot of the essays were written really without knowing how the war was going to end. The essay on Hitler makes the point (which in retrospect was true) that Hitler was a better strategist before the war began than he was after the war began.
The book was published in 1943, so it was probably completed in 1942. This was at a time when the outcome of the war was very much in doubt. For long stretches of 1942, it seemed plausible that the Axis might be able to, if not win the war, at least push their conquest to the point where winning it would be extremely difficult for the Allies. It was history written in real time, which is hard. That’s one thing.
The second thing is that the composition of the contributors is very much a product of the moment. If you go through and you look at the biographies of the people involved, a number of them were essentially refugees from Hitler’s Europe. They were European academics who’d been pushed off the continent by Nazi conquests and then ended up in the United States where, of course, they enriched the intellectual life of this country as well.
Then, the third point is that the contributors were very much aware that this was not a value-free exercise. They were not necessarily taking a god’s eye view of the international system.
Of course, they were trying to be objective and dispassionate in their analysis of history, but the point of the book was to help democratic societies do strategy better. This was not disinterested history. This was history with a commitment to helping democratic societies survive and flourish in a very dangerous world. That ethos has survived in the succeeding volumes.
Jordan Schneider: This idea of new history as a stimulus to action, with this book aimed at everyday concerned citizens, and not necessarily scholars of Jomini — that’s what makes this book so fun for nonprofessionals or students to flip through. The best essays make you want to buy a book about the topic because you’re interested in learning more.
There are so many little gems in these sentences and paragraphs that both try to teach you a lesson about the essentials of these stories, but also really end up enticing you to want to learn more.
One example from the essay on Delbrück, the military historian. He was this German guy who was the first one to actually try to count up how many people would have been at a Roman legion — for example, was Caesar exaggerating when he said he was fighting against 500,000 guys.
Hal Brands: It helps that the essays, particularly in the first volume, are about people. There’s something relatable about essays that are about people. That’s the thing that will draw in the folks who may not be academic experts on Jomini, but are just interested in military affairs and interested in strategy and interested in reading interesting things. That was part of what made the first book such a success. I know it’s part of the appeal of the book still.
The other nice thing about the first volume, by the way, is that the essays are all relatively short. They’re punchy. They get to the point. When I was putting together this volume, that was one of my goals, to make sure that the essays were meaty but didn’t go on forever and ever.
Jordan Schneider: There’s an essay on Hitler that essentially says, “We shouldn’t forget that Hitler is a genius.” It argues that the way he was able to pull off the 1930s is something that deserves praiseful discussion in the context of a grand strategist. That really stuck out to me, and it must have made quite the splash in 1942.
Let’s turn to the 1986 version of this book. What stuck out to you about that one?
Hal Brands: It’s an interesting book, in part because it took so long to do. The first discussion about updating Makers really started to happen in the 1950s. There were a bunch of different attempts to get a second volume, and various false starts involving various historians. It took 40+ years for the thing ultimately to come together.
What’s interesting to me about the second volume is that it’s written in light of the dangers of war in the nuclear age. Nuclear weapons create an element that were not there when the first Makers was published. That is reflected in the Lawrence Freedman essay in that volume that’s about the nuclear revolution and the schools of strategic thought that are associated with it.
It also hangs over a bunch of the essays in other ways. It’s there in terms of discussions of Clausewitz. It’s there in terms of just thinking about how high the stakes of war in particular have gotten and how important it is for people to understand what goes into good strategy in war.
In some ways, what’s also interesting about that book is that the definition of strategy changes from volume to volume. The definition of strategy in the first volume is very broad — it’s essentially what we think of as grand strategy, all elements of national power to achieve some important objective.
The definition of strategy in the second volume is narrower. It’s more closely related to military affairs and political-military affairs than it is to the larger conception of strategy. The nuclear revolution and the shadow it cast over all war and all statecraft in the second half of the 20th century has something to do with that.
Jordan Schneider: It’s weird that WWII ended two years after the first edition was published, and then the Cold War wraps up three years after the second one was published. I don’t know if there’s some leading indicator here.
Hal Brands: Maybe we’ll win it all in 2025 or 2026. Look out, Xi Jinping.
Project Management and Long Haul History Research
Jordan Schneider: Let’s discuss the newest edition. How does a project like this come together? Does Princeton University Press just call you up? Was there an interview process?
Hal Brands: There is a long story of how this volume came together that will be of interest only to me and my immediate family members. The short version is that Princeton had been thinking about doing a third edition because it had been 30+ years since the second volume.
It was clear that we were entering what would have been referred to in 2017 and 2018 as “the new era of great power competition.” A lot of the questions about nuclear strategy and long-term rivalry that had gone into abeyance with the end of the Cold War were coming back in a very serious way.
The editor of Princeton, Eric Crahan, came down to Washington and had a conversation with me and also with a couple of friends who were involved with the project. Then, for a variety of reasons, mostly pertaining to other personal commitments, couldn’t follow it all the way to the end.
We put together — in conversation with Princeton — a proposal for how to structure the book. The final product looks something like that initial proposal. The idea behind it was to do a book that would be richly historical like the other two volumes, but where the choice of topics would be relevant and would be recognizable to people grappling with challenges of US-China rivalry, nuclear deterrence, and the other issues of today.
Jordan Schneider: As you were going through that back catalog, what were the ones that you thought you couldn’t do without, and how did you decide to cut other subjects?
Hal Brands: Well, certain things are just so fundamental to an understanding of strategy that you really can’t do without them, especially if the idea is for this volume to stand on its own. You can read this volume without having already read volumes 1 and 2.
There’s a fair amount of overlap. Although all of the essays are new and original, when the book covers what’s called foundations and founders, basically, these aren’t the greatest hits of strategy, going back to Thucydides and the Peloponnesian War, Machiavelli, Clausewitz, and so on and so forth.
In each of those cases, the people who wrote on those subjects put really interesting new twists on the subject. I’ll call out Matt Kroenig’s essay on Machiavelli, which is actually quite original and quite interesting.
One of the real goals of the volume was to bring stuff up to date. Even though the 1986 version was written under the nuclear shadow, there were only four, maybe five essays that really dealt in detail with post-1945 issues. By the time we did version 3, obviously, we knew how the Cold War had ended. There was an entire generation of great scholarship on the Cold War. There is a whole section of 9 or 10 essays on Cold War-era stuff. Then there’s a whole section on post-Cold War content, because the post-Cold War era was 30 years old by the time the book was in gestation.
There’s much more of an effort to renew our understanding of strategy, not just through the greatest hits again, but also by looking at newer subjects that hadn’t been covered by earlier volumes.
Jordan Schneider: My favorite direct take on China is actually a riff-off of an old essay. It’s called “Economic Foundations of Strategy” by Jonathan Kirshner and Eric Helleiner. Instead of doing just Smith, Hamilton, and List, it was beyond Smith, Hamilton, and List. There was this really fun comparison between Chinese thinking and Western thinking in the late 19th and early 20th centuries about what kind of economy you needed in order to be a great power.
Hal Brands: I’ve got to give a shoutout to the two authors of that essay. Jonathan and Eric can take credit for that twist on the original. I went to them with a more conventional idea of an essay on the economic foundations of strategy. They asked if they could do something totally different, and it ended up being much better than what I had in mind.
Jordan Schneider: Did you start with a topic and then find an author, or did you start with the authors and then find topics? How did that matching process end up working for you?
Hal Brands: It’s a mix of both. There are some people who are so brilliant and so established in the field that you know you just have to have them in the volume. Basically, I would have let them publish their shopping list if they had offered to do that. I was going to have Lawrence Freedman in this volume no matter what he wanted to write. I was going to have John Gaddis in this volume no matter what he wanted to write.
Then, there are some people who you know are experts on a certain topic. You go to them and you ask, “Could you write on thing X?” Liz Economy has written — for my money — the best book on Xi Jinping’s China. I approached her and asked if she would write on that, and she very graciously agreed.
Then sometimes, you’ll take a proposal to someone and say, “Could you write on subject A?” They will say, as was the case with Jonathan Kirshner and Eric Helleiner, will say, “Well, why don’t I write on this other thing instead?” That happened in a few cases and it invariably made the volume better.
Jordan Schneider: That essay was excellent, bringing the legalist Sun Yat-sen and Albert Hirschman together into one argument. I can see how that wasn’t just an idea you pitched to them right out of the gate. It’s interesting how that editorial give-and-take works.
Hal Brands: There were a bunch of essays where that was the case. There’s an essay on the origins of the laws of war in the 19th century where I went to Wayne and asked him to do something more conventionally, came back and pushed back.
That creative tension or the give and take is actually one of the most interesting parts of an edited project because the people who you are recruiting to write these essays know far more about the subjects than I do. They’re typically a better judge of what’s interesting and what’s new.
Jordan Schneider: I’m curious, did they feel they had to bring their A-game? This is The Makers of Modern Strategy. This isn’t any essay collection.
Hal Brands: I will say this — I had far less trouble rounding up writers for this than you often do with edited collections. Let’s be honest, there’s not a huge professional payoff for writing essays for edited collections, in general.
But this is a special volume. It has been the authoritative text in strategic studies for 80 years, as you’ve pointed out. It’s a compendium of some of the greatest scholars in the field over a few different generations. I was hoping that authors would be excited about signing up for it for that reason — because I certainly couldn’t pay them enough to make it rewarding in a pecuniary sense.
I was just delighted that the vast majority of the people that I approached were willing to do it. The vast majority were excited about doing it. This is the thing that was really, really amazing. The vast majority turned in their essays in good shape and on time. I don’t say that because I wasn’t expecting good work from these people — they’re all stars. But, man, that’s usually hard when it comes to edited collections.
Jordan Schneider: There is a very cool intergenerational dialogue that is going on here. You’ve got contributors in their 80s and you’ve got contributors in their 30s as well. Aside from topic diversity, were there other diversities you were trying to build into this collection?
Hal Brands: There are a lot of different dimensions of diversity here. You mentioned one of them, which is that within this volume, we have two, maybe three different generations of scholars. At the more senior end, you have somebody like John Gaddis who’s been writing about strategy literally for half a century and does it as well as anybody else and just has an unparalleled knowledge of the field.
You have folks who are in the middle. Frank Gavin, my colleague at Johns Hopkins, who is certainly one of this generation’s preeminent extroverts on nuclear strategy, has written a couple of books about it and lent his expertise to this endeavor.
Then, you have younger folks as well. That includes Carter Malkasian, author of the best book on the US war in Afghanistan and one of the top scholars of the post-9/11 wars more broadly. Charlie Edel, who’s roughly of my vintage, wrote about John Quincy Adams.
There are some folks who I think view strategy in the more traditional sense in this volume, as essentially a political-military issue. Then, there’s somebody like Jason Stearns — I don’t know if he thought of himself as a scholar of strategy before he wrote this essay, but he brought a really interesting perspective on how strategy works in modern wars in Africa.
Jordan Schneider: How do you recommend people read the book?
Hal Brands: I recommend that people start by reading the opening essay, which is by me. This isn’t just self-flattery — the opening essay helps contextualize everything that’s going to come and try to piece together some of the common threads that you can pull across 45 different essays. I’d highly recommend that they read Lawrence Freedman’s essay which explains how our understanding of strategy has evolved over time.
Then, I’d say they should read about the things that most interest them. This isn’t a book where you have to read all 1,168 pages. You can get something out of it by reading the six or seven essays on the subjects that most concern you.
Then, I’d also recommend reading an essay or two that you wouldn’t normally read, that’s outside that six or seven. That’s actually where you’ll get new insights about strategy. If you read about Mao Zedong as a strategist, or if you read about the post-Meiji generation in Japan, or you read about Soleimani and Gerasimov or whatever the case may be, even if that wasn’t what got you interested in the book in the first place, there’s a payoff there because it’ll push you to think about strategy and how it’s practiced in different ways.
Jordan Schneider: For me, the least interesting essays were the China ones, which may be the same for a lot of the listeners of ChinaTalk. Thinking about China in the context of all these other essays and historical case studies was more rewarding in my opinion.
Hal Brands: All of the essays were chosen because they had something to inform our understanding of problems in the present. It could be that if you read a strategy about long-term competition — as seen by Jackie Fisher or Andy Marshal — it gives you some leverage on thinking about the US-China relationship today even though that’s not what the essay is really about.
It could be that an essay on the dynamics of multipolar rivalry in the early modern European system gives you some purchase on the dynamics of diplomacy in our current era. This is meant to be a book where you can find the relevance in pretty much any essay you read, even if the parallels aren’t directly drawn. You want the thing to stand on its own. You want people to be able to profitably read it 10 years from now, but it should also speak to the problems that people have in mind when they dip into a book like this.
Jordan Schneider: That’s an interesting way to read it — read the essays you’re interested in, but also read the essay that seems least interesting to you. For me, I gotta say — the title of the essay “French Strategies of Equilibrium in the 17th Century” didn’t necessarily do it for me. But there is some cool stuff in there! These are total weirdos. It’s a really different context, but also not 100% different, because it’s still people, it’s still states, and they’re still subject to their own constraints and opportunities.
Hal Brands: There’s that one, which is a great example of something where the relevance may not be immediately significant when you read the title. As you get into it, there is deep significance for understanding the challenges we face today.
You already mentioned the essay on Tecumseh where that’s the case. Mike Morgan, who was a professor at UNC Chapel Hill (and also happened to be my grad school roommate) has an essay on ideal politics or strategies of liberal transformation, how people have thought about the role of liberal ideas in taming and transforming international competition over time, that you can’t help but see echoes of that in post-Cold War American statecraft.
Jordan Schneider: You mentioned that you were trying to write for something that will still be impactful in 20 years. There are not a lot of incentives in contemporary academia pushing people toward projects like that.
Hal Brands: Well, in history, it’s different. The nice thing about writing history is that in most cases you’re not shooting at a moving target.
We know how World War II ended. You should be able to write something about World War II that stands the test of time if you do it well, and that people can profitably pick up 10, 20, or 30 years down the road. In fact, I’m working on a project that has a chapter about World War II. One of the best books that I’ve read on the subject was published in 1968. That’s definitely possible.
It obviously becomes harder the closer you get to the present. That is an unavoidable dilemma. You can see it, by the way, in all of the volumes in this franchise. We talked about the Hitler essay, an essay on Japanese strategy in the first volume, which cuts off in the middle of the war as just things are getting really interesting. The essay by Condi Rice on soviet strategy in the second volume that leaves you hanging, as you mentioned, five years before the Soviet Union itself comes to an end. There are essays in this volume. We mentioned the essay on Xi Jinping, the essay on the Kim Dynasty in North Korea.
I have no doubt that people are going to be able to read those profitably a number of years from now. Stuff’s going to happen, and they will become dated over time. At some point, somebody will feel it necessary to do a fourth volume of Makers of Modern Strategy. I’m sure that’ll be an interesting one as well.
Jordan Schneider: One of the most surreal essays was “Dilemmas of Dominance: American Strategy from George H.W. Bush to Barack Obama” by Chris Griffin. I lived through most of that history. I don’t want to think that I’m that old, but I vividly remember the start of Bush II’s Iraq war. As someone who’s been reading news obsessively ever since then, it is so surreal to see 20-30 years of history that I personally experienced just slimmed down into just 25 pages.
Unipolarity was, as identified by Krauthammer, a matter of fact. It was the product of the wave of events that left the United States a solitary superpower, bolstered by the resilience of its Cold War-era alliances, an increasingly liberalized world economy, expanding democratization, and the implausibility of any near-term peer competitors. The fact of unipolarity presented Bush and his successors with a fundamental, unexpected question: how should the United States exercise its newfound dominance in the international system?
[The New Makers of Modern Strategy, p. 870]
That seemed to be a particularly difficult one, especially, as you said, how archives are going to be open, and people are going to reevaluate all of the judgments that have been made, particularly over something that’s so recent.
Hal Brands: I will give a special word of thanks to the author of that piece, Chris Griffin, who now basically plays the role in strategic studies that Carnegie played for it in the 1930s and 1940s. His day job is with the Smith Richardson Foundation, which has funded just amazing work on a variety of strategic style-to-use topics over the years. There’s no conflict of interest. They did not fund this project. Chris was chosen entirely on the merits, but they deserve recognition for the work that they have done in this field.
What’s interesting about Chris’ essay is that it helps us understand the degree to which primacy, as much as it was a strategy, was a condition that gave rise to various habits in American foreign policy. Some of those habits were good. Some of those habits were bad. A lot of those habits persisted across multiple American presidential administrations.
The point that Chris makes, which I agree with, is that there was more continuity across post-Cold War American statecraft than we often think, in terms of what international system the United States was trying to bring about, in terms of what it thought threatened that international system and in terms of a shared commitment over a period of at least 25 years to trying to lock in as much of the good stuff, the spread of democracy, the US military advantage over any rivals, the promotion of globalization that followed the end of the Cold War.
We at least have enough perspective on this period. We can look at it over a generation plus to be able to see some of the continuities between administrations and evaluate the period as a whole.
Jordan Schneider: It’s a particularly tricky one to write, because everyone who was in those positions or writing books about every topic at a time is going to have a take that isn’t necessarily the one that they want to be enshrined in Makers of Modern Strategy lore for the next 30 years. Anyways, brave effort by him to try to synthesize all those presidents.
After you get the drafts, to what extent did you try to have them have a coherent tone, and put them in dialogue with one another? What was the back and forth between you and the writers?
Hal Brands: Well, I was downright fanatical on the length of the essays, because pretty early in the process, my editor at Princeton told me that if we got beyond 1,199 pages, and the book is pretty close to that, the spine would quite literally crack, and you wouldn’t have a book anymore. You’d have two separate unbound books at that point. That was one area where I definitely weighed in.
Look, all of the people who contributed to this volume are really distinguished thinkers, writers, and scholars. You don’t want to have a heavy hand in dealing with the stuff they produce. I tried not to mess with the tone of the essays. I certainly tried not to mess with the conclusions of the essays.
There were a couple of cases where I suggested, “Hey, you might consider this dimension of the problem,” simply because I had a degree of familiarity with the thing that people were writing. There are areas where I suggested trims to try to get it down length. There were a few areas where we went back and forth a little bit on not so much directly going into conversation with other essays. The John Gaddis’ essay at the end of the book is really the only one that does that explicitly. Just teasing out key dynamics that I knew were going to be present through chunks of the volume, because I had read all of the pieces in a way that nobody else had.
Again, when you’re dealing with a group of contributors this prominent and this good, the less meddling you do, the better.
Jordan Schneider: There’s a famous story with Robert Caro and Bob Gottlieb in the first edition — or the first book he wrote on Robert Moses, where the first draft was so long that they ran up against the spine problem. In the latest movie, they talk about how cutting down chapters to make it into one volume is one of the biggest regrets of their life.
What’s wrong with doing two volumes? How did you land at the page limit and the amount of topics?
Hal Brands: I’m really a stickler for brevity. I’ve rarely read a 17,000-word essay that wouldn’t have been better as a 14,000-word essay or an 11,000-word essay. I say that as somebody who’s written some 17,000-word essays.
My view was that you could cover most of these subjects with adequate nuance and with adequate depth at 10,000 words, and that readers would get more out of that because they’d be more likely to actually read all of it than they would be if the essays were 20,000 words long.
I love the first volume. I love the second volume. If I have one critique of the second volume in particular, it’s that some of the essays are really, really long and become a little bit difficult for non-expert readers to get through. That was a problem I was determined to avoid. Virtually all of the contributors were on board with that in one way or another. It really turned out not to be a huge issue. I actually think the book is better for it.
Jordan Schneider: Why’d you do this alone? This couldn’t have been the plan from the beginning, was it?
Hal Brands: No, this was not the plan from the beginning. I was initially going to have two co-conspirators in this project. One, is Frank Gavin, my very good friend and colleague at Johns Hopkins, who runs the Henry Kissinger Center there. The other, Eliot Cohen, also of Johns Hopkins, SAIS, was the dean of the school at the time that we were putting those together.
There’s no real story behind why neither of them ended up doing it. They both just ended up with a variety of other commitments that made this hard to do.
Fortunately, Frank was able to contribute to the volume. He wrote a remarkable, idiosyncratic, deeply insightful essay on the perplexities of nuclear strategy, which I think people are going to be getting a lot out of for many years to come.
Jordan Schneider: Let’s talk a little bit about the Gaddis essay. What was the origin? Why do you think it was so cool?
Hal Brands: John Gaddis was the hardest to get of all of the contributors, despite the fact that he was my dissertation advisor, or maybe because of the fact that he was my dissertation advisor. The calculation was he had already done his part for me and didn’t need to help with this one. In all seriousness, I did finally get him to agree to do it.
What he was really interested in doing was writing a reflection on all of the essays in the volume and explaining them in the context of the larger craft of strategy.
John was an eminently good sport in all of this. As soon as I got the first drafts of the essays, I would read them, mark them up, and send them to John. John would read them and, as we were rushing to get the volume ready, wrote his own essay on this.
His essay covers 2,500 years of history — everybody from Pericles to Putin is in the essay. It does the quintessential John Gaddis thing, where there are four amazing insights per page, and you feel you want to stop and think about the first one, but you’re already on to the second one and the third one and so forth.
It’s maybe 7,000 words. It’s not a particularly long essay, but it’s a really fitting summation of a lot of the insight and wisdom that the other contributors brought to the volume. It’s also a summation of what John has learned and taught us about strategy over a 50-plus-year career studying it. I felt very privileged to get him involved with the project, because I just couldn’t think of anybody better to bring the volume to a conclusion.
Jordan Schneider: Hal, do you have any thoughts or observations of the upcoming generation of scholars, and where their interests are? Where the field is going and what might be different in the 2040 version?
Hal Brands: One thing that might be different is that none of the people in the 2040 version are going to work in history departments. The reason for that is that the discipline of history as it’s practiced in academia has just changed a lot over the past 40 or 50 years.
I would guess that the percentage of contributors to this volume who work in history departments is lower than it was in the 1986 volume, for instance, because the people who study decision making, statecraft, war, and peace — are now as likely to be found in professional military education institutions, political science departments, policy schools, and think tanks as they are likely to be found in traditional history departments.
That trend will continue. I don’t know that it’s necessarily a bad thing. Diplomatic and military history, while they haven’t exactly flourished within history departments in the last 40 years, have flourished in these other spaces.
The makeup of the next generation may be a little bit different. Of course, the issues that they’re preoccupied with and the experiences that they bring to the task will be different as well. If you were writing for the first volume, you were drafting your essay at the end of 1941, you would live through some serious history over the past five years. That shaped almost everyone’s approach to the task.
Same thing. There’s a different set of histories that the people who contributed to this volume lived through. That’d be the same with the next volume as well.
Jordan Schneider: Maybe this gets to the one critique I’d have of the essay collection. One thing that I think really weighed heavily on the 1943 one was the weight of the technological machine age transformation that allowed a world war to happen in the first place.
In the second edition, you had the invention of the nuclear bomb as something that hovered over everything.
My expectation is the 2050 edition will have a number of essays about cyber attacks, AI, quantum computing, or technological changes that aren’t even on our radar yet.
Hal Brands: Haha, the next volume would just be written by different versions of ChatGPT. The revolution will come in a different way.
We do have an essay in this volume, which is one of the more provocative ones by Josh Rovner at American University, which basically says, “None of this stuff is as revolutionary as you think. New technologies come, new technologies go. We always think they’re going to revolutionize warfare and grand strategy. They typically revolutionize it less than we think. Then, the next set of technologies comes on.”
Now, it’s provocative, and people will argue with that thesis. What Josh is doing is exactly in the spirit of the volume, which is trying to historicize the debates that we’re having about cyber and AI and quantum today by looking at how previous technological step-changes have and haven’t changed the practice of strategy.
Jordan Schneider: Do you worry that the field of history, as it aspires to be timeless and everlasting, creates a biased preference for researchers who don’t internalize big technological changes?
Hal Brands: Josh isn’t a historian, he’s a political scientist. We let him in anyway. He did a great job.
Jordan Schneider: Last question — how did you reconcile the goal of decentering the US if Americans were also the target audience of the book?
Hal Brands: I’m not sure that America is the target audience, actually. I try to be transparent about my motives and what excites me about doing this volume, which is to help citizens of the democratic world be better at doing strategy because it matters for our future.
The Taiwanese edition of The New Makers of Modern Strategy. Source.
It matters in the present moment, as the democratic hegemony that we became accustomed to after the end of the Cold War is by no means guaranteed. I want the book to help the democracies of the world understand strategy better.
Strategists in Russia or Iran could probably learn something from reading this book. There is something universal about the challenge of strategy, even though every strategic dilemma has its own characteristics.
I would also say that the choice of chapters in the book is deliberate in the sense that it’s meant to get away from the transatlantic focus of the first volume, less so in the second volume. There’s an essay about Tecumseh. There’s an essay about Russian and German strategies under Hitler and Stalin. There are multiple essays about China, essays about Japan, and the Middle East, and strategies of nonviolent resistance India.
You’re right that the US is more at the center of the story than probably any other country. In that respect, the focus of the book is just an artifact of the part of history that it looks at.
Jordan Schneider: Folks, this was not a sponsored episode. I just think this book and this collection is really fantastic. It’s hard for me to imagine a listener to ChinaTalk that’s interested in the sorts of topics that I cover week to week that wouldn’t really enjoy and find this volume valuable. Really encourage you all to check it out and let me know what you think about it.
It might be cool to do some follow-ups if there’s particular audience feedback on a handful of the essays to maybe get a little panel of contributors together.
I do want to close, Hal, with a line that you had in your introductory essay. You say that, “If history is an imperfect teacher, it’s still the best we have. History is the only place we can go to study what virtues have made for good strategies and what vices have produced bad ones. The study of history lets us expand our knowledge beyond what we have personally experienced, thereby making even the most unprecedented problems feel a bit less foreign.
Indeed, the fact that strategy cannot be reduced to mathematical formulas makes such vicarious experience all the more essential. History, then, is the least costly way of sharpening the judgment and fostering the intellectual balance that successful statecraft demands. Above all, studying the past reminds us of the stakes that the fate of the world can hinge on getting strategy right.”
As we enter a scary world in 2025, think everyone would benefit from taking a moment to breathe and read some essays on Jomini, and Clausewitz, and Tecumseh and John Quincy Adams. It’s a good way to spend Sunday mornings.
Thank you so much, Hal. Thanks to all the contributors for putting together such a remarkable edition.
Hal Brands: Thanks, Jordan. It’s great to have this option to talk with you.
Last week, the Wall Street Journal editorial board asked Donald Trump why China would not invade Taiwan on his watch. Trump told the Journal that the Chinese would not dare to invade. As Trump put it: “[Xi Jinping] knows that I am f—ing crazy.”
One must pity the Chinese analyst asked to predict what a second Trump administration will mean for U.S.-Chinese relations. Like Richard Nixon before him, Trump is ready to play the lunatic; he clearly believes that the less predictable he is to the Chinese, the better off America will be. Though China occupies a central place in Trump’s campaign rhetoric, his campaign has not published or endorsed any detailed China policy proposals. The actions of the last Trump administration do not provide a better guide. Divided by infighting, its China policy was not consistent. At times, Trump’s foreign policy swung wildly as specific individuals rose or fell from his favor. Things do not get much easier if one looks at the views of the politicians and policy wonks that Trump would call on in a second administration. Their views are varied. Among Trump’s closest allies, we find fundamental disagreements on the proper ends and proper means of American strategy toward China.
Given these hurdles. I will not try to predict the path a second Trump administration might tread. It seems more useful to lay out a few observations on the different schools of thought now contending for leadership of that policy. My observations are shaped by the dozens of interviews I have conducted over the last two months with Republican staffers, think tankers, and former officials. A longer and more thorough report of my findings will be published by FPRI later this year. This is a pre-election preview.
The questions that divided Republicans in 2017 are not the questions that will divide them in 2025. Trump’s election shattered a policy consensus shared by the leaders of both parties for the better part of four decades. Many of the architects of this consensus were still influential during Trump’s first years in office. On the other hand, many who rejected “engagement” with China had spent years exiled from power. Others were completely new to service in the executive branch. This was a diverse group who did not all reject engagement for the same reasons. These differences were not initially apparent, as their objections were too marginal to the pre-Trump policy debates for much scrutiny to be given to them. Nor was it immediately apparent to these officials where the new bounds of public opinion or presidential approval lay. Thrust into power quite suddenly, they were forced to improvise as they went—and improvise again as the Chinese reaction to Trump’s trade war changed the context in which they worked. All of these factors gave China policy under Trump 1.0 an unusually chaotic flavor.
None of these conditions hold this election season. The architects of engagement are no longer relevant. A tough line on China is now taken as a starting point for all factions involved. Over the last eight years, a new ecosystem of conservative think tanks, policy journals, and Congressional offices has sprouted up to provide Trumpism with the intellectual coherence it lacked in 2017. Policy proposals are now numerous and detailed. Out of power, former Trump officials have had the time to carefully lay out their vision for American strategy in Asia. They have done this in speeches, policy reports, and full-length books. Disagreements between their different schools of thought are formally debated on both panels and podcasts.
Points of Consensus and Conflict in Trump World
Amid these debates, one finds several points of consensus. The disputing intellectuals, wonks, and politicians all agree that China is the most significant foreign policy problem the United States now faces. They describe China as a challenge that must be met in many dimensions: military, economic, and technological (some would add “ideological” to this list, but that is a point of debate, not consensus). Republicans agree that the U.S. armed forces are poorly structured and lack the resources needed to counter the military challenge posed by the People’s Liberation Army (PLA). They agree that America’s commercial and financial relationship with China underwrote the rise of a powerful rival while undermining America’s own industrial base. They believe that China has taken advantage of the traditional American commitment to globalization and free markets, and that doubling down on this commitment is foolish. To level the playing field, some mix of tariffs, export controls, capital controls, and industrial policy is necessary. They agree that the Biden administration’s China policy—while an improvement on that of the Obama administration—has nonetheless been feckless. They believe that the Biden administration articulates geopolitical goals that it has not resourced, cares too much about perceptions of amity, cares too little about perceptions of strength, and has not sold the American people on its foreign policy priorities.
But behind this consensus lie many fundamental disagreements.
The debates about China policy can be largely sifted into two buckets: economics and geopolitics. It is common for individuals to be closely allied in the economic sphere but not in the geopolitical sphere, or vice versa. For example, senators Marco Rubio and J.D. Vance are close allies on the economic front; there are few meaningful distinctions between the economic strategy each endorses. Their respective takes on the geopolitical problem posed by China are much harder to reconcile.
In theory, one’s position on the CHIPS Act or tariff rates might influence one’s position on military commitments to Taiwan or military aid to Ukraine. In practice, this is rarely so. The economic and geopolitical debates occur on different planes.
One way to represent the core principles at play in the geopolitical debate is with a classic two-by-two matrix (popularized on the internet as a “political compass”).
Optimism vs. Pessimism
On the x axis I place the single most important difference between the various schools of thought: assessments of American power and state capacity. Where one falls in many of the most prominent debates—such as “Can the United States can afford to support both Ukraine and Taiwan?” or “Should the ultimate goal of our China policy be victory over the Communist Party of China, or should it be détente?”—has less to do with one’s assessment of China and more to do with one’s assessment of the United States. What resources can we muster for competition with China? Just how large are our stores of money, talent, and political will?
Those on the right quadrants of my diagram provide pessimistic answers to these questions. They buttress their case with measurables: steel produced, ships at sea, interest paid on the federal deficit, or the percentage of an ally’s gross domestic product spent on defense. Against these numbers are placed fearsome statistics of Chinese industrial capacity and PLA power. Changes in technology, which favor shore-based precision munitions at the expense of more costly planes and ships, further erode the American position. This is a new and uncomfortable circumstance. The last time the United States waged war without overwhelming material superiority was in 1812.
To those who see American power through this frame, there is only one logical response: the United States must limit its ambitions. This means either radically reprioritizing defense commitments to focus on China or retreating from conflict with China altogether.
Those on the left two quadrants see things differently. Where the pessimists see settled facts, the optimists see possibilities. The optimists recognize many of the same trends as the pessimists, but view them as self-inflicted mistakes that can, and should, be reversed. An inadequate defense budget is not a law of the universe but a political choice. If Trump wins, he will choose otherwise. Implicit in the optimist view is a longer time horizon—there is still time to turn things around. But this window will not be open forever. Optimists fear that pessimistic assessments erode the political will needed to make changes while change is still possible.
The arguments between pessimists and optimists could be reframed as a matter of risk. The pessimists are most worried about the downside risks of a crisis with China in the near future (c. 2025–28). The optimists balance that possibility against the longer-term risks America will face as it withdraws from other regions of the world or abandons defense capabilities that are not needed in the Pacific theater. Optimists believe this second class of risks is large, and that the United States should not court them. Even an America in desperate need of defense reform has some capacity to “walk and chew gum at the same time.” This issue is at the crux of their arguments on Ukraine: in material terms, aid to Ukraine is not coming at Taiwan’s expense. It is relatively cheap. What stops America from helping both beleaguered nations?
The pessimists do not view that question purely in material terms. In their debates, the pessimists are quick to highlight the few weapons systems being shipped across the Atlantic that might be used in the Pacific, but their critique reaches higher than this. The costs of the war in Ukraine (and the Middle East) are measured not just in bullets, but in attention and effort: There are only so many minutes the National Security Council may meet. Washington can only have a few items on its agenda at any given time. The executive branch is stodgy, slow, and captive to bureaucratic interests; the legislative branch is rancorous, partisan, and captive to public opinion; the American public does not care a whit about the world abroad. Accomplishing anything meaningful in the United States—much less the drastic defense reforms both sides of the debate agree are necessary—requires singular attention and will.
If this seems like a pessimistic take on the American system—well, it is one. It is common for people in the optimistic quadrants to argue that the People’s Republic of China is riddled with internal contradictions. In a long-term competition between the two systems, they are confident that these contradictions will eat China from the inside out, and that America’s free and democratic order will eventually emerge victorious. None of the pessimists I interview make similar predictions. If they have anything to say about internal contradictions, it is American contradictions they focus on.
Power-Based vs. Values-Based Perspectives
So much for the optimist-pessimist divide. What of the y axis?
I think of this as a pole, with “power-based” perspectives on one hand and “values-based” perspectives on the other.
Republicans in the top two quadrants ground their arguments in cold calculations of realpolitik. From this perspective, international politics is first and foremost a competition for power. States seek power. The prosperity, freedom, and happiness of any nation depend on how much power its government can wield on the world stage. While states might compete for power in many domains, military power is the most important. A state frustrated by a trade war might escalate to a real war, but a state locked in deadly combat has no outside recourse. The buck stops with the bullet.
From the power-based perspective, then, the goal of American strategy must be the maximization of American power, with military force as the ultimate arbiter of that power. This force does not need to be realized in combat—ideally, its deterrent power will be strong enough that it is never actively used. The ideal means of American strategy is a military posture and alliance system strong enough to deter the Chinese from resorting to war.
The left and right quadrants of this perspective disagree on the best way to build that sort of power. The upper right quadrant—the prioritizers—do not believe America will ever possess power sufficient to compel China into submission; a stable détente between the two countries is the best outcome that America can attain. Even this modest aim will only be possible if the United States prioritizes the threat posed by China above all others.
Those who argue from the upper left quadrant—the primacists—also speak the language of realpolitik. They maintain, however, that the sacrifices the prioritizers propose will weaken American power. They believe that the existing American alliance system contributes to America’s strength today and will contribute to America’s potential strength in the future. Instead of limiting American aims, the primacists are more concerned with expanding American means. They are confident this can be done if the American people have the confidence to do so.
The lower two quadrants, whose arguments I label “values-based,” operate under a different frame. The people in these quadrants believe that American foreign policy should not be evaluated by a single variable. They see connections between what America does abroad and what America is like at home. They have strong values-based commitments to specific ways of life that are expressed in their vision for American strategy.
I have labelled those in the bottom left quadrant “internationalists” because of how often they invoke the phrase “liberal international order.” This group believes that America and its allies are knit together not only by shared security interests, but also by shared values. In fact, the values shared by the liberal bloc explain why these countries share security interests in the first place. China is an authoritarian power whose influence operations threaten the integrity of democracies across the world. Many internationalists view this political-ideological threat as the most dangerous that China poses. Those in this quadrant are especially skeptical of détente; they do not believe permanent compromise with China is possible. They attribute Chinese belligerence to the communist political system that governs the country. For them, tensions in U.S.-Chinese relations are less the expected clashes between a rising power and the ruling hegemon than a battle between two incompatible social systems. Pointing to the close cooperation that ties Iran, North Korea, Russia, and China together, the internationalists argue (contra the prioritizers) that the world is gripped in a general contest between liberal order and resurgent authoritarianism whose different parts cannot be disentangled from each other.
Those in the bottom right quadrant—the restrainers—also think about foreign affairs through a regime lens, but the belligerent regime in question is their own. Republican restrainers link the liberal international order to the free trade agreements all Trumpists despise and the administrative “deep state” all Trumpists distrust. They see the liberal international order as an international extension of the progressive order they are trying to tear down at home.
There are echoes of the 1960s New Left in the restrainer argument. Both the new left of yesterday and the new right of today are rebellions against “the establishment.” Both reject the pieties of their day; both see a bloated national security state as a symbol of the dehumanizing values they reject. Both groups correctly point out that there is no natural limit to the quest for primacy. Both argue that a totalizing foreign policy will lead to the bureaucratization of American life.
Only the most radical restrainers are ready for a 21st-century march on the Pentagon. Most aim for an easier target: a relatively modest foreign policy. Instead of defending an entire international order, it is enough to defend America. Instead of deterring authoritarianism, it is enough to deter China. China does not need to be defeated—it is enough to convince the Chinese to accept some sort of détente.
This is all pretty similar to the ends sought by the prioritizers. Little wonder so many of the primacists and internationalists I interviewed believed the prioritizers were restrainers in disguise! Again and again I heard this accusation made: prioritizer arguments are just an attempt to make isolationism sexy. The prioritizers do not actually believe in realpolitik—realpolitik is just a respectable way to attack the existing international order they despise.
There is an irony to this critique. Just as primacists and internationalists condemn the false face of the prioritizers, so the prioritizers and the restrainers condemn the false face of the primacists! Many of those I interviewed insisted that their primacist opponents made such-and-such argument not for the realpolitik reasons they professed, but because of their (hidden) commitment to liberal ideals. Ideals that cannot be defended on their own merits had to be prettied up with talk of hard power.
All of these suspicions of subterfuge are overblown. Both primacists and prioritizers believe the arguments they make. Yet their suspicions are revealing! All sides clearly believe there is political advantage in couching one’s arguments in realpolitik logic. That fact alone tells us something about the likely contours of a Trump presidency—and perhaps the beliefs of Trump himself.
ChinaTalk is a reader-supported publication. To receive new posts and support our work, consider becoming a free or paid subscriber.
The following is a guest translation from Sihao Huang, a current PhD candidate at Oxford.
Two researchers from Fudan University—renowned for its American Studies program—published an impressively detailed analysis in FT Chinese examining Trump and Harris's tech policies. As it turns out, Chinese analysts have been reading about the AI Manhattan Project, tracking the debate around Lina Khan, and even listening to the All-In Podcast [JS:god bless them…].
The analysts believe that Harris, the "AI czar," will pursue a more interventionist approach to AI regulation but will likely have a weaker hand than Biden in antitrust policies due to her active engagement with tech giants and her Californian roots. They also think she will take an "iterative strategy of checking for loopholes and filling gaps" with export controls.
In contrast, they expect Trump to run an "AI Manhattan Project" to compete against China, accelerate the development of military technology, repeal Biden's AI Executive Order, and encourage more flexible AI governance. They also see Trump as being more "radical" with export controls, blocking China's access to cloud compute and taking some action against Taiwan for ‘stealing’ America's semiconductor business. Ultimately, the two authors argue that regardless of who wins the election, "the current trend of pan-securitization in the U.S. digital technology industry is not affected by the change between the two parties." Given the ENFORCE and Remote Access Security Acts, the "march to suppress Chinese technology will continue."
Stick around till the end for a painting and some poetry by my favorite Chinese painter, Shitao.
Divergent Paths: Differences in Harris and Trump's Technology Policy Approaches
The technology policies that Harris and Trump might adopt represent two distinct paths for future U.S. technology governance. However, both share a consensus on technology policy toward China.
October 30, 2024 - Written by Yao Xu and Zhang Ao (Fudan University) for FT Chinese. Source. Translation abridged.
As candidates for the Democratic and Republican parties, Harris and Trump show significant differences in their attitudes and positions on technology policies. Harris is likely to continue Biden's technology policies, including artificial intelligence (AI) regulation, antitrust enforcement, increased taxes on the ultra-wealthy, opposition to racial and gender bias, cryptocurrency regulation, and promotion of digital equity. In contrast, Trump has explicitly stated that he will overturn Biden's AI governance policies as represented by the "Biden Executive Order," adopt relatively loose regulatory measures on the technology industry, give the green light to mergers and acquisitions, and continue tax reduction policies. Regardless of who is ultimately elected as the next U.S. president, their policy propositions will profoundly impact the global technology ecosystem and the geopolitical technology landscape.
Artificial Intelligence: Will Biden's AI Executive Order Be Abolished?
On October 30, 2023, Biden signed the Executive Order on the Safe, Secure, and Trustworthy Development and Use of Artificial Intelligence, the White House's first regulation on generative artificial intelligence. It aims to strengthen federal government safety oversight of AI, calling on federal departments to review AI's impact in their specific domains and find solutions to problems within each agency, thus becoming one of America's landmark policies on AI regulation to date.
Harris is inclined to continue this executive order, strengthening government supervision and public oversight of AI. She believes AI development needs regulation to ensure public safety and interests. Analysts believe that if Harris wins the election, she may adopt a more proactive and interventionist AI regulatory policy. The federal government is expected to be encouraged to actively procure AI tools while ensuring appropriate safeguards for their use across federal agencies. Additionally, Harris will urge Congress to legislate and improve the legal framework for protecting data privacy and cybersecurity.
Harris's political career began in California, with close ties to the tech industry, and she started focusing on data privacy and security issues in the technology sector early in her political career. Since becoming vice president, Harris has earned the nickname "AI Czar" for her active leadership in AI-related affairs. "It's very important that she's from California, from the Bay Area, and has served as a politician there," said Alondra Nelson, former director of the White House Office of Science and Technology Policy. "Silicon Valley is not unfamiliar to her." Harris has her own agenda in AI governance. "When an elderly person's healthcare plan is canceled because of an erroneous AI algorithm, isn't that a matter of survival for them?" Harris expressed concerns about AI algorithms' impact on marginalized social groups in a November 2023 speech. Harris emphasizes AI's social impact and safe development, striving to maintain a balance between technological progress and public interests.
Although she advocates stronger regulation, Harris is not as outspoken as Biden in calling for the breakup of tech giants or reshaping the current AI power structure at the federal level. As the election approaches, Harris is actively engaging with tech giants, continuously easing tensions between the Democratic Party and Silicon Valley while advancing the agenda from a middle-ground position. Box CEO Aaron Levie says Harris's positive attitude has achieved results, and "tech giant executives also need a stable leader who supports immigration and science." Levie states that as long as leaders respect technology and tech companies' own development logic, even if they implement tax and antitrust policies or strengthen AI safety supervision, they won't face strong opposition from the tech community.
Trump seeks to repeal Biden's executive order. In his campaign manifesto, Trump claimed: "We will repeal Biden's dangerous executive order, which has hindered AI innovation and imposed radical left-wing ideas on the development of this technology. Instead, Republicans will promote AI development based on free speech and human prosperity." Trump opposes excessive regulation of cutting-edge industries like AI while encouraging flexible "non-regulatory methods" such as policy guidance, pilot trials, and voluntary frameworks to minimize barriers to AI technology development and application, ultimately serving comprehensive competition with China. According to The Washington Post, in response to the Biden administration's AI executive order, the Trump team is drafting a new executive order proposing the implementation of an "AI Manhattan Project," aimed at promoting rapid development of AI technology, especially military applications, and will work to clear legal obstacles for implementation. Increased military investment in AI will benefit technology companies like Anduril and Palantir that have established good cooperation with the Pentagon. The main executives of these giants support Trump and have close ties with the Republican Party.
According to The Washington Post's exclusive report, the "AI Manhattan Project" will also create an "industry-led" agency to evaluate AI models and ensure they are protected from foreign adversaries. The plan includes a section titled "Making America Number One in AI," which proposes a strategy for the AI industry that differs significantly from the Biden administration's AI executive order. This suggests that if Trump returns, he may implement policies beneficial to Silicon Valley investors and tech giants. To win votes and realize his political agenda, Trump has become increasingly friendly toward Silicon Valley. In June of this year, Trump appeared on the "All-In Podcast" YouTube channel, hosted by well-known technology investors. After the show, Trump also participated in a fundraising event hosted by podcast co-host and former PayPal executive David Sacks. In the podcast, Trump stated that he has realized Silicon Valley's "geniuses" need more resources to promote AI development to compete with China. Chamath Palihapitiya, founder of venture capital firm Social Capital, believes Trump has gained more support in the technology industry than in the 2016 election. Trump's inclination also shows he needs to cooperate with technology elites and build a cabinet to "change the status quo."
Export Controls on Technology Industries Like Semiconductors: Will They Become More Radical?
Harris will continue the Biden administration's strategy of taking small, quick steps, gradually escalating, and fixing loopholes in the export control process for semiconductors and other sectors. The Biden administration's export control policy design around semiconductors and other fields reflects a pragmatic action strategy, which is more evident in the export control regulations issued by the Bureau of Industry and Security (BIS) of the U.S. Department of Commerce in October 2022 and October 2023, respectively. The former restricted the peak computing power of single chips and the data transmission performance between multiple chips, resulting in NVIDIA's most advanced GPU models A100 and H100 being banned from export to China. NVIDIA provided the Chinese market with cut-down versions A800 and H800 chips. To prevent NVIDIA from continuing to "exploit loopholes," BIS changed the restriction method in the 2023 new regulations, removing the previous "interconnection bandwidth" as an important parameter for identifying restricted chips, which directly led to the ban of A800 and H800. It can be predicted that if Harris comes to power, she will continue the iterative strategy of checking for loopholes and filling gaps, controlling and restricting some "emerging" advanced technologies while seeking a balance between blocking and suppressing China's technology industry and controlling differences.
Trump may become more radical on export controls. Based on his previous term, China was the first to be hit in semiconductor export controls under Trump. Since March 2018, when Trump launched the "301 investigation" and initiated trade and technology wars against China, the Trump administration began extensively using the BIS "Entity List," weaponizing export control regulations. Numerous technology companies, universities, and research institutions became victims, and the global supply chain suffered serious impacts. Additionally, in his last year in office, Trump targeted TikTok and WeChat's international versions, seeking to implement harsh ban measures on Chinese-based digital platforms in the United States. If Trump returns to the White House with extremely conservative vice presidential candidate J.D. Vance, he may continue using CFIUS (Committee on Foreign Investment in the United States) to restrict Chinese companies' acquisitions in semiconductors and other high-tech fields, continue restricting the export of high-computing chip products to China, and restrict exports of chip manufacturing equipment, parts, and chip design software upstream in the industrial chain, as well as computing power leasing services downstream. Beyond China as the main competitor, Trump's "America First" stance will also affect upstream and downstream interests in the semiconductor industry. On July 18, Trump commented on U.S. involvement in Taiwan's defense, saying that Taiwan "took away" U.S. chip business, and the U.S. should not act as "insurance" for Taiwan's defense. TSMC, the world's largest chip foundry, saw its stock fall that day. Overall, although Trump has shown a strong tendency toward export controls, his governance style is changeable, and if he comes to power, the scope and effectiveness of his policies remain uncertain.
Antitrust Enforcement: Can Silicon Valley and Wall Street Breathe a Sigh of Relief?
Harris may be inclined to promote antitrust regulation. On the one hand, the social impact and security development of cutting-edge technology are political priorities that Harris values highly. She tends to increase enforcement against Silicon Valley tech monopolies, believing these platforms abuse their strong market position and infringe on consumer interests, making it difficult to meet people's basic living needs. She has repeatedly spoken about the need to reduce inflation and provide more economic opportunities for all Americans. If Harris comes to power, she is expected to appoint strong law enforcers to key positions in core departments responsible for antitrust, such as the Department of Justice, Federal Trade Commission, or Federal Communications Commission. On the other hand, Harris also maintains contact with tech giants and tries to find a balance between government regulation and corporate innovation. During the 2020 election, Harris said that while strengthening regulation, the breakup of tech giants like Google should be opposed.
So far, Harris has not made a statement on how to handle antitrust issues. Wall Street hopes Harris can relax the antitrust enforcement seen during the Biden administration and create a new antitrust regulatory environment. Democratic donors such as IAC Chairman Barry Diller and LinkedIn co-founder Reid Hoffman believe Harris's position on this issue can be adjusted. As election day approaches, Harris is also under pressure to quickly enrich her position on key policy issues, and her final stance will depend on practical political considerations. Within the Democratic camp, firm antitrust policies are favored by party progressives such as Elizabeth Warren and Bernie Sanders, who hope Harris will continue Biden's policies. Elizabeth Wilkins, former director of the Federal Trade Commission's Office of Policy Planning, believes that despite Harris's ambiguous attitude, her other work protecting families and small businesses during her vice presidency is "fully consistent with the antitrust agenda."
Trump is not very active in antitrust enforcement. During his first term, Trump was rather cold toward antitrust matters. If he comes to power, he may continue to deal with current technology antitrust enforcement cases but will still give the green light to technology mergers and acquisitions. Trump's business background makes him generally more sympathetic to the business community, and his tax cuts and trade protectionist economic policies are quite popular with tech giants and their wealthy executives and middle-class employees. Out of consideration for their own interests, the Silicon Valley technology community, known as the "liberal bastion," has recently expressed support for Trump after the shooting, causing Trump's donations from the technology community to rapidly rise and surpass those to the Democratic Party.
However, the conservative camp to which Trump belongs also has its own antitrust agenda. His deputy, Vance, has publicly praised Federal Trade Commission (FTC) Chair Lina Khan for "doing a pretty good job" in antitrust work against tech giants such as Amazon and Google, and said that large technology companies need to be restrained. In February of this year, Vance called for the breakup of Google on social media. The conservative camp tends to reduce regulatory agencies while being willing to use antitrust supervision to check and balance tech giants. The contradictions in the conservative camp's antitrust stance will also impact Trump's antitrust policy.
The March to Suppress Chinese Technology May Be Difficult to Stop
The policies that Harris and Trump may adopt regarding technology industry issues represent two different paths for America's future technology governance. However, they have formed a consensus on technology policy toward China: internally promoting innovation policies while externally pursuing technological decoupling.
On one hand, the Biden administration is strengthening its technological competition with China. Since this year, the United States has used data as its entry point and data security as its justification to implement America First policies in technology industries and infrastructure construction, intensifying the competitive situation with China. On February 28, 2024, U.S. President Biden signed Executive Order 14117, "Preventing Access to Americans' Bulk Sensitive Personal Data and United States Government-Related Data by Countries of Concern," restricting the transmission of personal data to "specific countries." Subsequently, Chinese industries such as smart vehicles and shipping have also received focused attention due to alleged data security concerns. Furthermore, the United States has also restricted China's artificial intelligence development through means such as the Enhancing National Frameworks for Overseas Restriction of Critical Exports Act (also known as the ENFORCE Act) and the cloud computing bill.
On the other hand, Trump initiated the strong suppression of China's technology industry. During his term, Trump launched a trade war with China through the "301 investigation" and quickly extended it to a technology war. Through various means such as the "Entity List," presidential executive orders, and the promotion of China-related bills, the Trump administration adopted a strategy of universal coverage and focused attacks on China's technology industry. In 2018, the Trump administration launched the so-called "China Initiative" and conducted internal reviews to prevent research results from being "stolen." According to MIT Technology Review's study of prosecuted cases, the program fabricated numerous "unfounded" cases, many of which had little connection to technology and national security, negatively impacting American scientific research's reputation. During Trump's term, China's technology industries, including semiconductors, communications technology, artificial intelligence, new materials, and digital platforms, were generally "targeted." Multiple Chinese technology companies were placed on the entity list. WeChat and TikTok were once on the verge of being banned. Huawei was particularly targeted by Trump, causing serious damage to its global supply chain. Even in his final moments before leaving the White House, Trump signed an executive order directing federal agencies to assess the security risks of Chinese-made drones in the government fleet and to prioritize the elimination of Chinese-made drones.
Although the election is approaching, the current trend of pan-securitization in the U.S. digital technology industry is not affected by the change between the two parties. The overall direction of the United States' technological containment of China will not be affected by the final election result. The difference lies in the specific implementation methods and paths. Whether it is Harris's "gradual escalation" or Trump's "extreme pressure," their policy propositions will profoundly affect the global technology ecosystem and the geopolitical structure of technology.
(Note: Yao Xu is an associate researcher at the Development Research Institute of Fudan University, and Zhang Ao is a research assistant at the Development Research Institute of Fudan University. This article only represents the authors' personal views.)
ChinaTalk is a reader-supported publication. To receive new posts and support our work, consider becoming a free or paid subscriber.
A painting to take you into the night: Shitao’s Drunk in Autumn Woods from 1702.
Exploring the theme of intoxication, Shitao makes us reel with the unsteadiness of his figures and landscape. He probably created this picture to commemorate an autumn outing in the hills with friends. The three inscriptions testify to the pleasure and pride that Shitao took in his achievement; they make it clear that the exhilaration experienced by the friends was inspired by nature, poetry, conviviality, and creativity as well as wine.
White clouds and red trees amid the wild fields, those who go, go; those who come, come.
Yesterday in the open countryside, I let my gaze wander free, seven jewels and eight treasures compete with the green mountains.
People and plants are all drunk together, when the west wind strips all bare, that's when we try to sober up.
Who can say what true refinement is? In my old age, my nature tends toward seeking foolishness. (translation: Claude).
原题跋（三）：
頃刻煙雲能復古，滿空紅樹漫燒天。 請君大醉烏毫底，臥看霜林落葉旋。
In an instant, mists and clouds can return to their primeval form; Red trees fill the skies, spreading fire through the heavens. I invite you, sir, to get very drunk on my black brushstrokes; Lie down and watch the frosted forest as falling leaves swirl. (translated by 枫荣注)
Antoine, aka Duoduodiliao 多多底料, is a Mandarin teacher in France by day and a Chinese rap enthusiast by night. Today, he’s here with a setlist of his favorite hip hop tracks. His original songs can be found here.
Track 1: 芳草地 (The Fragrant Meadow) by DIGI GHETTO (艾志恒Asen/thomeboydontkill/mac ova seas/KIV/Mula Sakee/付思遥)
Duoduodiliao: DIGI GHETTO is a rap group based in Chengdu, composed of six members. Their official debut made quite a fuss because the six of them were already pretty famous before they began to work together. Some people say they are like a new-gen Higher Brothers. The mixtape is really nice — it’s club-friendly, the beats are good, and the flows are very inventive. The lyrics are a little bit corny, but there is great chemistry between all the members. You can tell they enjoyed themselves while recording the album.
Track 2: 威远故事 (The Story of Weiyuan County) by GAI周延
Duoduodiliao: The next song is by GAI — he’s one of the most successful rappers in China.
Jordan Schneider: For background, GAI is from Chongqing, and he initially became famous by winning the first season of “Rap of China,” which came out in 2017.
Duoduodiliao: GAI’s 2022 album is called 杜康 “Dù Kāng,” and it’s pretty uneven but it’s a proper rap album. This isn’t the GAI we see on CCTV singing Chinese New Year nonsense — this is the real GAI rapping his guts out, on some tracks at least.
The song I chose is 威远故事 “The Story of Weiyuan County,” and it’s one of the greatest Chinese rap tunes that came out in 2022. It's an introspective song where GAI talks about his past, his childhood, and his hometown — and it’s absolutely beautiful.
Jordan Schneider: GAI got canceled for having songs about drugs and doing crimes. Then he had this weird patriotic rehabilitation tour where he was making songs about the Great Wall and how awesome China is. It looks like GAI is going back to his roots with this album.
Track 3: 变蓝 (Turning Blue) by 也是福 (Eddie Beatz) feat. PO8 and 喜辰晨
Duoduodiliao: Next we have something different — this song is from a producer’s mixtape from 2022. The record is by 也是福, also known as Eddie Beatz — he’s one of the greatest producers in China. He has worked with notable artists like MaSiWei 馬思唯, Wang Yitai 王以太, and Xiao Lao Hu 小老虎. His tracks are usually quite jazzy and organic — he uses many live instruments rather than computer programming.
I discovered this record while randomly browsing NetEase Music. This album (也是蓝) is a collaborative mixtape featuring beautiful instrumental interludes. The song we're going to listen to is called 变蓝 (Turning Blue) featuring PO8 and Voision Xi 喜辰晨.
Track 4: 亚特兰蒂斯陷落 (Atlantis Surrenders) by 弗兰德斯坦/C-Low
Jordan Schneider: You mentioned browsing NetEase Music 网易云音乐 to find Chinese songs. Could you explain what NetEase is? How can people access it, and what makes it special compared to Spotify?
Duoduodiliao: NetEase Music is a Chinese streaming platform similar to Spotify, but it’s primarily available in China. There’s a modified international version, but it’s a different app, kind of like the distinction between Douyin and TikTok.
On NetEase, you can find Chinese artists, including underground artists who don’t focus on reaching international audiences. They simply release their music on Chinese apps. To truly understand Chinese music, especially underground, you need to use platforms like NetEase Music or QQ Music.
Jordan Schneider: You can access it by switching your app store to China, downloading the app, and then switching back to the country you live in. You can also use the desktop apps.
The apps themselves are fascinating to explore. They offer AI-generated playlists, user-generated playlists, and unique features like KTV (karaoke) functionality for every song. There’s even a national KTV leaderboard for each song, which keeps track of high scores like in an arcade game. Each song has discussion sections where people debate about the music and discuss specific lyrics. Artists maintain active profiles, similar to Weibo, where they share music updates and lifestyle content. It’s much more interactive than Spotify.
Duoduodiliao: The next song is by 弗兰德斯坦 (Flanders), a new rapper from Changsha. He recently appeared in the CSC Changsha City Cypher 2023, but for now, he’s still underground. This track, “Atlantis Surrenders” is only available on NetEase Music. It’s not on YouTube or Spotify.
The track features C-Low, former leader of the Beijing rap crew Easy Boys Gang. The use of autotune in this track is unique in Chinese rap. What makes this track special is the instrumental beat and Flanders’ deep, rich voice. It’s really impressive.
Track 5: 春雪采耳 (Ear Cleanse In The Spring Snow) by 施鑫文月 (SHII) and 小老虎 (Lil Tiger)
Duoduodiliao: Let’s continue with 施鑫文月 (SHII). He released an album in 2023 called “Sichuan Renaissance: Chapter Two” (巴蜀文藝復興：第二章), following Chapter One from 2021.
This record brings fresh air to an oversaturated rap scene. It crosses boundaries between musical genres like hip-hop, pop, and alternative. It’s also an ode to Chengdu local culture, discussing memories and intimate moments from the district where he grew up in Chengdu.
He talks about specific cultural elements, like elderly people playing mahjong, exercising, and dancing in public squares.
The song we’re discussing is from another one of his records, but it really captures an interesting slice of Chengdu culture — people in parks and squares offering to clean your eardrums with special tools.
Jordan Schneider: There’s a Douyin video of me getting that done.
Duoduodiliao: How does it feel? Is it pleasant?
Jordan Schneider: No, it’s not pleasant — it felt invasive. Your body produces ear wax for a reason. For days afterward, I felt like particles were getting into my ears because there wasn’t any wax to catch them.
TLDR; I’m not a fan. Maybe we need some randomized controlled trials to study it. Anyway, here’s the song.
Track 6: THE MESSAGE PT.2 by CREAM D and 艾热AIR
Duoduodiliao: The next track is by CREAM D from his album “Life After Life.” He's an OG rapper from Xi’an 西安 who started in the early 2010s. He’s Christian, and he discusses his spirituality a lot on this album.
Since CREAM D hadn’t released an album in many years, expectations were high for this one. He didn’t disappoint his audience. While he’s known for his technical skills, sharp flows, and lyrical ability, the introspective nature of this particular album makes it stand out.
The song we're going to listen to is “The Message Pt.2,” featuring a famous artist from Xinjiang called 艾热AIR. He’s a Uyghur rapper who won the 2023 season of Rap of China.
Jordan Schneider: Cool. I really like this one.
Track 7: 落幕 (Sunset) by Asen (feat. GALI, 堵琳Caroline)
Duoduodiliao: GALI is a Shanghai-based rapper who’s gained significant momentum through Rap of China. He went mainstream thanks to his natural charisma, clean flows, and well-written punchlines, making him hugely popular among Chinese rap audiences.
The song we’re going to listen to isn't actually from GALI’s album but rather features him on another rapper’s track. It’s called “Sunset” by Asen featuring GALI.
Track 8: 囚 (Cage) by 李佳隆 (JelloRio)
Duoduodiliao: The next album is 传奇 “LEGEND,” by Sichuanese artist 李佳隆 (JelloRio). He’s one of my personal favorites. I love what he's contributed to the culture these past few years.
In my opinion, this 2022 album is a flawless piece of work. The production team paid great attention to detail, with songs transitioning seamlessly from one to another. To fully appreciate it as the production team intended, you need to listen to the whole album in order.
The song we’re going to discuss is called “Cage,” and it blends Chinese folk 民谣 with hip-hop elements.
Track 9: 恨与爱 (Hate and Love) by AThree
Duoduodiliao: Next is Xinjiang rapper AThree with his 2022 album “Alpha 8.”
AThree’s record stands out for its lyrical quality - great poetry and smooth flow. He might be one of the few mainstream rappers in China who confronts political subjects in his songs.
The track we’re discussing, “Hate and Love,” addresses the Xuzhou chained woman incident 徐州铁链女事件, which sparked significant controversy in 2022.
Jordan Schneider: A man in Jiangsu province had a woman chained under his house for years, essentially keeping her as a sex slave. It was horrific. The discovery process revealed police negligence, and it became a weeks-long national discussion about how something so terrible could happen in modern China.
Jordan Schneider: What does AThree say on this track?
Duoduodiliao: His message is that rappers should be speaking out about these kinds of incidents. He criticizes how Chinese rap has changed since 2017, with many mainstream rappers only talking about cars and money.
Jordan Schneider: He’s a Uyghur, right?
Duoduodiliao: Yes. On all of his albums, there are always one or two tracks only in the Uyghur language.
Track 10: 不负责 (Why u blame on me?) by Capper and (ﾉI A I)ﾉ♡
Duoduodiliao: Our final song is from Capper's album. The English name is “Sword and Roses.” To me, it's one of the best Chinese rap albums of 2022.
Capper is a new-generation rapper based in Xi’an. He’s participated in several TV shows like Rap of China. His album is pretty incredible — both musically and production-wise, it’s really unmatched this year because it pushes musical boundaries to new heights. He experiments with nu-metal and hyperpop, and he executes it all perfectly.
His flow is on another level. He's a very promising artist with the potential to lead Chinese hip-hop toward new horizons. The song I've chosen is called 不负责 (Why u blame on me?). It’s super catchy.
Reflections on the Chinese Rap Ecosystem
Jordan Schneider: Can we discuss politics and hip-hop over the past few years? How have the boundaries changed, and how have rappers navigated these limits since 2021?
Duoduodiliao: The pandemic period in China revealed the extent to which rappers could address political topics before facing consequences. During this time, several rappers openly discussed the zero-COVID policy and Shanghai lockdowns. Because these were sensitive subjects, hearing critical voices was rare. Many WeChat groups faced bans.
Dr. Li Wenliang, who criticized the pandemic management approach, later died. Seeing rappers speak openly about these issues was refreshing — it suggested Chinese rap maintained connections to conscious rap traditions of addressing societal problems and government policies.
However, this trajectory didn’t last. One rapper, Sean ZH, based in Beijing but educated abroad, was banned from Weibo for a month after discussing the lockdown situation. This government response likely discouraged others from addressing political topics.
The boundaries remain unclear. Rappers can still discuss certain societal issues like work culture, but touching more sensitive subjects results in complete bans — their songs become impossible to post on any platform or social media.
Jordan Schneider: In 2018 and 2019, many rappers began traveling internationally and collaborating with foreign producers. How did travel restrictions impact Chinese music production? Has the situation changed since China reopened?
Duoduodiliao: The market changed significantly. Previously, many Chinese rappers toured North America, performing in Canada and the U.S., leading to numerous international collaborations. The lockdown completely halted this progress, preventing rappers from maintaining their international audience.
Now that travel has resumed, rappers are gradually rebuilding these connections. Many Chinese rappers attended the Rolling Loud festival in Thailand. However, progress remains slower than pre-lockdown levels.
Chinese rappers still face challenges in developing international audiences. The different social media platforms and apps used in China versus overseas create additional barriers.
Jordan Schneider: Any final thoughts on hip-hop’s evolution in China?
Duoduodiliao: Understanding Chinese rap requires recognizing the significance of rap TV shows like “Rap of China.” These shows traditionally offered underground rappers opportunities for mainstream success.
A rapper’s stance toward these shows defines their position in the scene. Some maintain “authenticity” by refusing to participate, viewing the shows as compromising artistic integrity. Others embrace these platforms for exposure.
The meaning of “keeping it real” differs between Chinese and U.S. rap contexts. Due to censorship and local conditions, Chinese hip-hop authenticity often centers on artists’ positions regarding these TV shows rather than traditional markers of credibility.
And one more for the road from Jordan:
ChinaTalk is a reader-supported publication. To receive new posts and support our work, consider becoming a free or paid subscriber.
Part of me finds Donald Trump’s behavior here unusually relatable. You think I want to sit up here talking about politics and war day after day?
You don’t know the temptation to just once, just for one week, turn this podcast into a drum and bass set, or play some of my favorite Kiasmos songs. But I don’t. Of course I don’t. It’s not what we’re doing here.
Well, Ezra, I hear you. Lucky for us, ChinaTalk is not a property of The New York Times — which means we can publish DJ sets whenever the mood strikes! So this week, we’re giving you what you need, not more election takes but a fantastic deep dive into Chinese shoegaze. This episode comes courtesy of ChinaTalk columnist and Jake Newby of the Concrete Avalanche Substack. They put together a wonderful radio hour playing some amazing tracks and walking you through the genre.
Have a listen to the show on our podcast on Spotify, Apple Podcasts, or search ChinaTalk on your favorite podcast app!
One of the stand-out acts from the Kind of Shoegaze Vol. 1 compilation focused on young Chinese bands that was released at the start of the year, NarrowLaneAngel formed in Inner Mongolia in 2023. In August of that year, they released an eponymous debut EP.
Track 2: Limpid · Forsaken Autumn · 卢佳灵
Based in Shanghai, Forsaken Autumn have been together since 2011, propelled by the talents of britlulu (who also founded the East Asia Shoegaze Festival) and singer Ecke Wu. Released at the tail-end of 2015, Forsaken Autumn’s record Whenere — dubbed “the Chinese Loveless” by one commenter on Bandcamp — is a classic in Chinese shoegaze circles.
Track 3: Nostalgia · Summer Daze
Founded in 2021, Summer Daze are another band who featured on the Kind of Shoegaze compilation series from Amemoyo. After a couple of early EPs, they've put out four new singles over the course of this year.
Track 4: Firework · The White Tulips
Xiamen music scene stalwart Chen Zhenchao (also known as Soda) has moved away from shoegaze into surf-rock and dream- and city-pop territory with his more recent projects, but in 2015 he and his band The White Tulips delivered the decidedly shoegazey Fondle. It’s re-release on vinyl in 2021 was a nod to its status as a Chinese shoegaze classic.
Track 5: float · 巧克力文件岛
Hebei five piece Chocland.doc apparently first came together at a former residence of Eileen Chang, but are seemingly yet to write a song based on Lust, Caution or any of her other novels. “Of course, the name of the band has no meaning,” they say. “What you understand is what you understand.”
Track 6: Is your dream still out-focus · Lunacid
Another one of China’s newer shoegaze acts, Lunacid was formed in 2023. The trio hails from Changsha and was also featured on the Kind of Shoegaze compilation series.
Formed just last year, 哲学思潮 hail from Nanning in Guangxi province, near the border with Vietnam. Their debut album Dazedtrek was recently made available on Bandcamp.
Track 8: Detached · The Numen
Shanghai-based quartet The Numen met on arts review platform Douban and have pursued a shared love of shoegaze and indie-rock since the summer of 2023. They namecheck shoegaze greats such as My Bloody Valentine’s Kevin Shields among their influences.
Track 9: Cat Tenant (Summer) (猫房客夏日版) · Baby Formula
Formed by “three boring people with no expectations for the unknown journey ahead,” Beijing band Baby Formula came seemingly out of nowhere, dropped an impressive eponymous debut album in the autumn of 2013, and then promptly disappeared again. Still, their music continues to resonate over a decade later.
Track 10: 星星 (Star) · 親愛的艾洛伊絲 (Dear Eloise)
As frontman for long-running band PK14, Yang Haisong is one of the godfathers of Chinese post-punk. Yet he’s also played a pivotal role in bringing more experimental, noisy, and yes, shoegazey sounds to the fore. Formed in 2007 with his wife (and one-time PK14 bassist) Sun Xia, Dear Eloise have released a string of atmospheric records over the years and remain an influential act in China.
If you like this playlist, you might also like thesecompilations from Jake and consider subscribing to his Substack!
The entire project of ChinaTalk, at the heart of it, is to help raise the quality of conversation and help inform policies that would most likely avert WWIII. I’m worried Trump would do a far worse job avoiding the big one than Harris. Here’s a piece on the topic I co-authored with Peter Harrell, formerly of the Biden NSC and NEC, with some riffs on export controls and acquisition reform below.
Former President Trump has made his promise to “prevent World War III” central to his campaign to return to the White House. But Trump’s foreign policy agenda is far more likely to drag the U.S. into catastrophic conflict than to prevent it.
Trump’s promise understandably resonates with voters. While President Biden did end the war in Afghanistan and keep U.S. troops out of direct involvement in conflicts in Ukraine and the Middle East, the world is less peaceful today than it was four years ago. In 2022, Russia launched the largest war in Europe since World War II. War in the Middle East makes daily headlines. China has increased the operational tempo of military exercises that threaten Asian allies.
However, one-off dealmaking with dictators from a position of weakness is a losing strategy to keep the peace between great powers.
Putin and Ukraine
For starters, let’s take Trump’s promise to negotiate an end to Russia’s war with Ukraine. What is Trump’s “secret plan”? Reporting indicates that he would pressure Ukraine to exchange land for peace by cutting off military support. This approach will fail: Even if Trump cuts off military assistance to Ukraine and Ukraine is forced to cede territory to prevent immediate military collapse, Putin’s ambitions will not be sated. His goal is to assert Russian dominance across Eastern Europe — which includes all of Ukraine, the Baltic countries and Poland. Degrading Ukraine’s military in return for a temporary ceasefire only for Putin to renew conflict against a weakened Ukraine months later is disastrous dealmaking.
Trump’s approach to Ukraine would be particularly dangerous given how Trump wants to treat NATO allies in Putin’s crosshairs. He recently said that he would give Russia an explicit pass to “do whatever the hell they want” if NATO members do not “pay [their] bills.” If Putin succeeds in taking parts of Ukraine, this language will embolden him to turn his war machine towards America’s treaty allies in Eastern Europe — which would either drag the U.S. into war, or see the U.S. acknowledge its alliance commitments are worthless. This will seed more conflict, not deter it. To avoid direct conflict with Russia, the U.S. should help Ukraine further advance its military capabilities to force Russia to acknowledge that its maximalist aims are impossible to achieve.
Xi and Taiwan
Trump is also risking war where the stakes may be even higher: East Asia. In a recent interview with Bloomberg Trump responded to a question about America’s commitment to Taiwan by stating: “Taiwan should pay us for defense. You know, we’re no different than an insurance company. Taiwan doesn’t give us anything. Taiwan is 9,500 miles away. It’s 68 miles away from China.” He sees Taiwan not as a democratic ally and bulwark against Chinese regional hegemony, but rather just another country who has wronged the US economically by taking “100% of our chip business.”
This all amounts to Trump turning America’s eighty-year commitment to East Asian security into a short-term lease renewal negotiation. What Trump plans to do isn’t savvy dealmaking, it’s diplomatic arson.
Mainstream Republican national security thinkers like Elbridge Colby and Mike Gallagher have advocated committing to Taiwan’s defense for principled (preserve a fellow democracy), geopolitical (counter Chinese expansion) and economic (ensure a stable global chip supply) reasons. The credible threat of U.S. military support for allies in Asia today helps dissuade Xi Jinping from invading Taiwan. This of course can swing too far — like Former Secretary of State Mike Pompeo’s assertion that Taiwan should declare independence. Under Trump, both China and America’s key allies in the region would feel far less confident that the US would actually help them resist a Chinese attack.
This loss of faith in America’s commitment to the region risks dramatic consequences. It will encourage Chinese adventurism against Taiwan, raising the odds of a war that ultimately drags the U.S. in. Even though U.S. allies including Japan and Korea have increased their defense spending in recent years, none of them can realistically deter Chinese aggression with conventional weapons without America’s backing. Pulling away from Japan and Korea would prompt them to either lean more towards Beijing or consider going nuclear, injecting more instability in a region where conflict could be truly catastrophic in terms of lives lost and global economic impact.
Preventing WWIII is not a one-off real estate deal, it’s a repeated game which requires signaling across decades to both allies and adversaries that the US is serious about preserving the peace. Since 1945, America succeeded in investing in our military might, alliances and global credibility to deter great power war. Today, America makes up a quarter of global GDP, while China comprises almost 20%. Together with its treaty allies, the US totals over 60%, while China’s closest thing to an ally, Russia, adds just one 1% to its ledger. Preserving this relative balance in national power will allow the US to keep the peace and further its interests far into the 21st century. Risking global realignment to juice up a few acquisitions deals plays right into Xi and Putin’s hands.
For all Trump’s bluster about preventing conflict, his actual policies, like the isolationism of the 1930s, would actually increase the chances of a war far deadlier than even today’s conflict in Ukraine. President Reagan worked with NATO and East Asian allies to show “peace through strength”. Trump’s strategy of treating America’s closest global friends like delinquent renters risks war through weakness.
Back to Jordan sole-authorship. A few more riffs on ChinaTalk-adjacent topics:
Export Controls
AI could really, really matter for long term national power. AI consists of algorithms, data, and compute. Algorithms and data are probably too hackable to drive relative long term national competitiveness. So, we’re left with computing power, which after o1 is even more likely to matter in pushing the frontier from an innovation and diffusion perspective.
What will Trump do with the Biden administration export controls on chips and semiconductor manufacturing equipment? Probably something like what happened to ZTE.
Step one: Trump’s bureaucracy finds a Chinese company flagrantly breaking US law whose continued use of American technology runs directly counter to US interests.
Step two: Trump saved ZTE because he gets a call from Xi.
I wouldn’t be shocked if he just lifted the controls month one as a goodwill gesture. If someone talks him into having a little spine on this issue, he’ll see a huge opportunity for a big deal as Xi would happily give Trump oceans of soybean orders and zoos-full of pandas for a few EUV machines.
There is no domestic political economy constituency for semiconductor export controls. The only political force today keeping them in place are national security professionals who rightly recognize their long term importance. An America First semiconductor policy, particularly in a market defined by scarcity, should aim to make sure all backordered are filled American fabs and datacenters before opening up an export market which would raise prices for US companies to acquire goods from a 3.5tn company and SME firms who have been rippingthe past few years. Instead, lobbyists will have far more success than they deserve framing the controls as some deep state conspiracy meant to skew a trade deficit.
Acquisitions Reform
Rule #1 of avoiding wars is having a military scary enough that no one wants to try it. It’s clear as day that the US needs a dramatic overhaul to how it buys capabilities in order to deter wars in Asia.
If I’ve learned anything over the dozensof shows we’ve recorded on defense policy over the past decade, it’s that absent sustained commitment from a president and secretary of defense, we don’t see the reform at the scale we need.
Trump picked his Secretaries of Defense because they had cool nicknames and looked the part. With wars in Ukraine and the Middle East, it will be hard enough for the next Secretary to preserve bandwidth to push the system to change without having to deal with monthly crises in civ-mil relations. I have zero faith in a Trump administration giving this issue the sustained attention it needs to make the progress the US needs to most effectively deter war.
EVs
Maybe less important for the future of humanity, but this one is just weird. Why is he so into BYD making cars in Michigan?
Not that there’s a single swing voter in the ChinaTalk audience … but thank you for indulging me. Back to our regularly scheduled programming later this week.
This summer, we teamed up with the Federation of American Scientists, and Chris Miller to hold a crowdsourced policy competition. We asked for ideas on how to deal with the problem of China potentially controlling the supply of foundational chips (also called “trailing-edge” semiconductors).
The U.S. has implemented export controls to try to stop China from getting a technological edge in advanced cutting-edge chips. But as I explained in a recent post, export controls have no hope of stopping China from building simpler types of chips — called “legacy chips”, “foundational chips”, or “trailing-edge chips”. These legacy chips are used for a huge number of things in our economy, from cars to smartphones to fighter jets.
And China is gearing up to build these legacy chips in absolutely staggering numbers. Check out this post by Jimmy Goodrich of the University of California Institute on Global Conflict and Cooperation and this post by the Rhodium Group for details. Basically, China is applying the same approach to legacy chips that it has successfully applied to batteries and EVs — massive scale and enormous subsidies.
This basically presents at least three potential dangers to the U.S:
First, China could deprive non-Chinese chipmakers of huge amounts of revenue by outcompeting them in the legacy chip market, making it harder for them to sustain their leading-edge chip businesses. Already investors are pressuring U.S. companies to avoid competing with China by canceling their semiconductor fabs.
Second, if China controls the legacy chip market, it could cut off our supply of chips in a war.
Third, Chinese security services might be able to put back doors into Chinese-made chips, using them to spy or even to attack U.S. infrastructure.
In other words, there are plenty of national security reasons for keeping Chinese-made legacy chips out of our supply chain. But how can we do it? It’s a tough problem.
First of all, as things stand, we don’t even know which products contain Chinese-made chips. If a Vietnamese-made phone or a Mexican-made PC includes Chinese-made legacy chips, the U.S. currently has no way of knowing.
Second, even if we did know, it might be politically unpopular to ban those chips. A lot of U.S. companies want to get chips as cheaply as possible, especially for new AI applications. We’d need some way to make chip restrictions politically palatable.
And finally, lots of Chinese legacy chips — and the products that contain them — aren’t going to be sold in the U.S. or our allied countries. How do we make sure non-Chinese chipmakers stay competitive in markets like Vietnam, Brazil, Indonesia, etc.?
We asked contestants to give us their ideas for addressing this problem and picked four winners. They each received $500 and will be working with FAS to continue refining their proposals.
The following write up is by , whose Substack consistently delivers excellent coverage on classic ChinaTalk themes like industrial policy and technology, interspersed with a little of my commentary. Overall, I was really impressed by the quality of submissions we received and am excited to run more of these in the future!
ChinaTalk is a reader-supported publication. To receive new posts and support our work, consider becoming a free or paid subscriber.
These winners are listed in alphabetical order by first author.
Winner #1: Weaponizing EDA and using targeted industrial policy
By: Zenghao (Mike) Gao, Charles Yockey, and Felipe Chertouh
Gao et al. point out an important weapon in the U.S.’ arsenal of export controls that hasn’t been used yet: Electronic design automation software (EDA). We hear a lot about where the production of chips happens, and some about where the production of chipmaking tools happens, but not very much about where the software used to design chips comes from. In fact, almost all of it comes from America, with a little bit coming from U.S.-allied countries like Japan and Australia. And this software doesn’t just design chips in the first place; it’s also what chipmakers use to correct problems with the fabrication process as they arise.
Gao et al. suggest that EDA could be “weaponized” by mandating that it run on U.S.-based cloud servers:
In hosting all EDA in a U.S.-based cloud—for instance, a data center located in Las Vegas or another secure location—America can force China to purchase computing power needed for simulation and verification for each chip they design. This policy would mandate Chinese reliance on U.S. cloud services to run electromagnetic simulations and validate chip design. Under this proposal, China would only be able to use the latest EDA software if such software is hosted in the US, allowing American firms to a) cut off access at will, rendering their technology useless and b) gain insight into homegrown Chinese designs built on this platform. Since such software would be hosted on a U.S.-based cloud, Chinese users would not download the software which would greatly mitigate the risk of foreign hacking or intellectual property theft. While the United States cannot control chips outright considering Chinese production, it can control where they are integrated. A machine without instructions is inoperable, and the United States can make China’s semiconductors obsolete.
This idea wouldn’t stop China from making foundational chips — Chinese companies could still use American EDA software. But it might give the U.S. one more piece of leverage to hold over China in case hostilities broke out — and another way to try to slow down the Chinese chip industry in general, if that becomes necessary.
On the defensive side of things, Gao et al. also call for the U.S. to form a trade bloc with Latin American nations to ensure safe supply of rare earths and NAND memory. They also have some additional ideas, such as forcing Chinese companies to release the source code for the firmware and other software for their chips. [Jordan: Enjoyed this one in particular for its creativity and density of ideas. Some of these to me seem more feasible than others but a ton of food for thought in this doc].
Full paper here: https://www.mikegao.net/public/chips-proposal.pdf
Winner #2: Working with other countries on industrial policies and tariffs
By: Andrew Lee
Lee sees the creation of a non-China foundational chip supply chain as the central problem to be solved. He envisions a program modeled after Lend-Lease — the system by which the U.S. delivered arms to the UK in World War 2, and by which it’s currently delivering arms to Ukraine. The program would license U.S. technology cheaply to friends and allies in exchange for cooperation in creating completely China-free chip supply chains:
The United States Federal Government could negotiate with the “Big Three” EDA firms to purchase transferable licenses to their EDA software. The U.S. could then “lend-lease” licenses to major semiconductor producers in partner countries such as Singapore, Malaysia, Vietnam, the Philippines, or even Latin America. The U.S. could license this software on the condition that products produced by such companies will be made available at discounted prices to the American market, and that companies should disavow further investment from or cooperation with Chinese entities. Partner companies in the Indo-Pacific could further agree to share any further research results produced using American IP, making further advancements available to American companies in the global market.
(Side note: It occurs to me that this might dovetail well with Gao et al.’s proposal for putting EDA on a U.S.-based cloud.)
[Jordan—very ambitious indeed! I feel like this would need some dramatic catalyst for a government to be aggressive enough to be able to push this over lobbyist objections. It also feels like there’s a vaccine analogy here.]
Lee also suggests coordinating with friendly countries in order to put tariffs on Chinese foundational chips. Recall that one of the big challenges here is that we don’t currently know which products contain Chinese-made chips, so we have no idea how many we’re importing. Lee’s solutions to this problem are 1) an international database of which products contain Chinese chips, and 2) reporting requirements for importers, enforced by random audits:
How would tariffs on final goods containing Chinese chips be enforced? The policy issue of sanctioning and restricting an intermediate product is, unfortunately, not new. It is well known that Chinese precursor chemicals, often imported into Mexico, form much of the raw inputs for deadly fentanyl that is driving the United States opioid epidemic. Taking a cue from this example, we further suggest the creation of an internationally-maintained database of products manufactured using Chinese semi- conductors. As inspiration, the National Institutes of Health / NCATS maintains the Global Substance Registration System, a database that categorizes chemical substances, along with their commonly used names, regulatory classification, and relationships with other related chemicals. Such a database could be administered by the Commerce Department’s Bureau of Industry and Security, allowing the personnel who enforce the tariffs to also collect all relevant information in one place.
Companies importing products into the U.S. would be required to register the make and model of all Chinese chips used in each of their products, so that the United States and participating countries could to impose corresponding sanctions. Products imported to the U.S. would be subject to random checks involving disassembly in Commerce Department workshops, with failure to report a sanctioned semiconductor component making a company subject to additional tariffs and fines. Manual disassembly is painstaking and difficult, but regular, randomized inspections of imported products are the only way to truly verify their content.
Finally, he suggests efforts to protect U.S. critical infrastructure by 1) identifying Chinese hardware within the infrastructure, and 2) improving cyber defense capabilities.
Winner #3: An “Open Foundational” design standard and buyers’ group
By: Alex Newkirk
Newkirk also sees Chinese disruption of the chip supply chain — along with possible backdoors and other security issues — as the main problem to be solved. He proposes two ideas. First, Newkirk would create an “Open Foundational” design standard for legacy chips, in order to ensure that China doesn’t get proprietary control over any type of computer chip. The chip companies who joined up to help create this standard would form a sort of cartel that could act to create a China-free manufacturing supply chain. Newkirk also suggests an international buyers’ group to create a strategic reserve of chips. This would serve the dual purpose of building up a chip stockpile and providing demand to encourage the adoption of the Open Foundational design standard.
He writes:
To secure supply of foundational chips, I recommend development of an “Open Foundational” design standard and buyers’ group…[T]he U.S. federal government…would establish a strategic microelectronics reserve to ensure access to critical chips. This reserve would be initially stocked through a multi-year advanced market commitment for Open Foundational devices.
The foundational standard would be a voluntary consortium of microelectronics users in critical sectors, inspired by the Open Compute Project. It would ideally contain firms from critical sectors such as enterprise computation, automotive manufacturing, communications infrastructure, and others. The group would initially convene to identify a set of foundational devices which are necessary to their sectors…and identify design features which…could be standardized. From these, a design standard could be developed…Steering committee firms will…be asked to commit some fraction of future designs to use Open Foundational microelectronics…[T]he buyers’ group would represent demand of sufficient scale to motivate investment, and that supply would be more robust to disruptions once mature.
Government should adopt the standard where feasible, to build greater resilience in critical systems if nothing else. This should be accompanied by a diplomatic effort for key democratic allies to partner in adopting these design practices in their defense applications. The foundational standard should seek geographic diversity in suppliers…The foundational standard also allows firms to de-risk their suppliers as well as themselves. They can stipulate in contracts that their tier one suppliers need to adopt Foundational Standards in their designs…
Having developed the open standard through the buyers’ group, congress should authorize the purchase through the Department of Commerce a strategic microelectronics reserve (SMR). Inspired by the strategic petroleum reserve, the microelectronics reserve is intended to provide the backstop foundational hardware for key government and societal operations during a crisis…The foundational standard provides the product specification, and the advanced government commitment provides demand…This demand should be steady, with regular annual purchases at scale, ensuring producers consistent demand through the ebbs and flows of a volatile industry….The SMR could also serve as a backstop when supply fluctuations do occur, as with the strategic petroleum reserve…This would ensure government access to core computational capabilities in a disaster or conflict scenario. But as all systems are built on a foundation, the SMR should begin with Foundational Standard devices.
It’s notable how Newkirk’s ideas support each other. The international chip design standard he would create would make it easier to build up a stockpile of reliable chips. And building up the stockpile would create the guaranteed demand that would encourage adoption of the design standard. That’s a very clever synergy. And as an added bonus, the consortium of companies that create and run the foundational chip standard would also be able to help carry out friend-shoring and de-risking, instead of leaving all the planning to the government.
Winner #4: A legal plan for blocking Chinese chips
By: Ben Noon
Noon focuses on the difficult problem of identifying and restricting Chinese-made foundational chips contained within U.S. imports from other countries. He vividly lays out the dangers of allowing China to control the foundational chip industry:
The list of examples of Chinese economic coercion is long…Washington faces less blatant coercion compared to its allies…This may be because Beijing does not believe it yet maintains necessary leverage over Washington…China’s growing position in the legacy semiconductor market could change that. How would Beijing’s behavior change if sales of the Ford F-150 relied on Beijing’s willingness to sell its semiconductors?
Noon argues that export controls have little or no hope of containing the Chinese foundational chip industry. And he argues that CHIPS Act-type subsidies alone are insufficient to maintain a U.S. foothold in the market because Chinese subsidies will always be larger. Thus, he concludes, protectionism is necessary in order to keep China from dominating the global market for foundational chips.
The question, of course, is how to restrict imports of Chinese foundational chips contained in other products. Noon goes through and explains a list of various legal and administrative vehicles that the U.S. government has at its disposal to accomplish that task:
Tariffs
Investigation of and restrictions on imported goods linked to unfair trade practices
Federal government purchasing restrictions
The Office of Information and Communications Technology and
Services (ICTS) at the Commerce Department, a recently created agency with broad authority to protect critical infrastructure from dangerous imported products
Noon believes that the most important legal justification for tariffs on Chinese chips is Section 301 of the Trade Act of 1974, which both Trump and Biden have used extensively in order to put tariffs on Chinese products.
The really tough question, of course, is enforcement. Noon recommends “a major expansion of supply chain analytical capabilities across the U.S. government,” but doesn’t say much more about that. He also suggests enlisting private companies as whistleblowers.
All of these proposals are quite interesting, and we’ve already contacted the authors to talk about following up on their development. I was very impressed by the diversity of ideas here — different contributors targeted different aspects of the problem, which helped them come at the issue from a variety of angles. I continue to be impressed by the creativity and technical acumen of Noahpinion readers. Expect more policy contests at some point in the future!
Thanks for reading ChinaTalk! This post is public so feel free to share it.
Critical minerals! The world needs more of them, and the US government can help. We interviewed Arnab Datta of the think tank Employ America to talk through the ideas he put forward in his piece co-authored with Daleep Singh on “Reimagining the SPR.”
Since we recorded, Senators Hickenlooper, Graham, Coons, and Young introduced the Critical Materials Future Act to establish a pilot program at the Department of Energy (DOE) to employ an array of creative financial tools and acquisition authorities to support domestic critical material processing, refining, and recycling projects. The DOE was authorized to receive appropriations of $750 million to support at least three domestic projects, with support for each going toward at least three types of critical materials.
The legislation includes many ideas first put forward in Arnab’s piece. Three exciting provisions are worth highlighting:
It gives the Secretary of Energy a wide array of authorities to support projects including the use of forward and futures contracts, options, and other transactions as deemed necessary;
It includes flexible hiring authority to ensure that the program can hire experts from the private sector to carry out the goals of the program; and
The program would be guided by a holistic set of objectives beyond mere “stockpiling” including to provide financial stability and reduce supply chain vulnerabilities. Furthermore, the report required by the creation of the program would evaluate the use of the flexible financing authorities to alleviating challenges like market volatility, boost price transparency, and other important goals.
The history and future of the strategic petroleum reserve, and how lessons from the oil market can offer insight about how to trade lithium, copper, and graphite.
How to engineer bipartisan support for moonshot policy ideas.
Lessons from a former kindergarten teacher on handling Congresspeople.
The Future of the Strategic Petroleum Reserve
Jordan Schneider: Let’s talk about critical minerals, stockpiling them, and smart ways to trade them.
Arnab, for background, what is the strategic petroleum reserve, and how has it evolved over time?
Arnab Datta: The strategic petroleum reserve was established in the mid-seventies, basically in the wake of the Arab oil embargo that came out of the Yom Kippur War.
As net importers of crude oil, we realized we were very vulnerable to these types of shocks. We established a reserve, which is essentially a set of salt caverns that are dotted all over the Gulf Coast that could store crude oil for us.
At the same time that this was happening, the International Energy Agency was established, and all the members were required to keep about 20-25 days of export capability in their reserves.
At max capacity, we probably held about 600-700 million barrels of oil in there, but it hasn’t been utilized extensively. There have been a couple of geopolitical disruptions where oil was released from the SPR, for example, during the civil war in Libya in 2010-11. We’ve used it in response to natural disasters like hurricanes and things like that here, but that hasn’t really been utilized to its full extent.
When it was established, people thought about it as this asset of oil sitting in a reserve that we can release. But the SPR was created with a whole other part of its mandate to basically boost our domestic industry. If you look at the acquisition authority specifically, there are a number of goals that are outlined in that, including maximizing domestic oil production and promoting competition.
Matt Klein: Historically, Congress basically just sold oil from the SPR for money to close budget gaps.
Arnab Datta: It’s funny — if you go back to 2014, there was a big review of the SPR during the Obama administration, and that’s when the recent set of mandated sales began. They were just there to fill budget gaps.
Right now, the legal capacity limit of the SPR is a billion barrels, but the physical capacity limit is 720 million barrels. Around 2014, there was a plan to establish a new site in Alabama. They canceled that. Then they just started selling off all the oil just to whatever budget deal needed a piggy bank. Now, hopefully, more and more people are realizing that is short-sighted.
Jordan Schneider: Talk a little bit about how domestic oil production changed the whole point of this thing and what you guys have been pushing in the wake of the invasion of Ukraine.
Arnab Datta: As I said, the SPR was designed when we were net importers of crude. But in the period since the SPR was established, we became net exporters of crude and we are now the number one oil producer thanks to the shale revolution.
As we’ve become this huge exporter, our vulnerabilities have shown in other places. Refining capacity is one — we haven’t built a new large refinery with capacity over 100,000 barrels per day since 1977. Exxon Mobil Holdings recently announced that they are trying to build one in Texas. But as you can see, our vulnerability is no longer really in crude oil.
A second thing that happened during that period was financialization. The WTI contract and the Brent crude contracts were established in the eighties, and they transformed the market dramatically. They’ve made it easier for producers to attract more financing. They created a mechanism for the bottom of the market to hit less quickly in the event of a supply glut or for the top to be a bit slower in the event of a shortage. More broadly, there’s a transparent market with important pricing and investment information that’s widely available.
Coming out of the COVID pandemic and recession, we started to get really worried about the prospect of a crude oil shock causing inflation and even a potential recession. We had been looking at investment patterns in shale and seeing investment drop off, so we were worried that prices would go really high.
We started to explore different authorities that we could use to boost domestic oil production in the short term and minimize some of that price pressure.
We zeroed in on the SPR, and whether contracting authorities and acquisition authorities at the SPR could be used to provide producers with the certainty they needed to go and produce.
Coming out of the shale revolution in the late 2010s — prices decreased and there were a whole host of bankruptcies among shale producers at each successive price level. Thus, investments eventually started to become less sensitive to price increases.
Even as prices coming out of the Russian invasion of Ukraine hit $130, we didn’t really see a huge investment response. Why? You heard oil CEOs talking about this. They had billions of losses over the previous decade that they needed to make up, and they couldn’t justify investments to their shareholders if the price increases were going to be just a momentary blip.
We needed some mechanism essentially to get at that capital discipline problem.
Matt Klein: When oil prices went extremely low and WTI went negative, there was actually a proposal to refill the SPR specifically to essentially bail out American oil producers.
That did not happen, which probably exacerbated the supply shock problem a year or two later.
Arnab Datta: It shows you how short-sighted politicians can be on this stuff. They took the stance of “We’re not going to give a bailout to oil companies,” but the oil companies ended up profiting off of that in the end.
Matt Klein: I know a lot of the work that Employ America has published is about using the SPR for managing crude, but is there a capacity with the existing legislation of the SPR to also stockpile and trade refined products as well?
Arnab Datta: It depends on how creative you want to be with an authority. We do have smaller kinds of refined product reserves — they’re not utilized in the same sense. There are storage issues that come with storing refined products.
A lot of this is oriented around the question of how much oil price instability — both for refined products and for crude — affects exchange rate stability. We made the case that this is something worth taking seriously and accordingly, that the Exchange Stabilization Fund at Treasury should be utilized for this purpose.
Now, that’s not something that it’s typically seen as outside of the scope of some of the folks at Treasury. I understand their concerns. But outside of some of those authorities, it’s difficult to do this.
Matt Klein: You mentioned the storage issues — do you mean that refined products just literally would not be suitable to put in the salt caverns that the SPR has?
Arnab Datta: Yes, they degrade differently. To be clear, we could build facilities to store refined products. But they degrade outside very, very specific conditions. They degrade much more quickly.
Jordan Schneider: Why don’t you talk through the tension of working on this and pushing forward climate goals as well?
Arnab Datta: I’ll credit my colleagues, Skanda Amarnath and Alex Williams, for making the case that this is important for decarbonization.
When you think about the energy transition, the price volatility of crude oil is as much of a challenge as crude oil itself. There’s a tendency among folks on the further left to engage in “keep it in the ground” rhetoric.
It ignores the fact that people need oil. This is not something that’s going away overnight. We need to think about how we are managing that stable transition and making sure that EVs, for example, are price-competitive with ICE cars.
This type of flexible contracting allows you to essentially set a soft floor on prices. If prices go to $40 again, EVs become less price-competitive. Making sure that the levers of policy are being utilized towards price stability over time is important.
Matt Klein: China is another very large oil consumer, of course, and they are much less transparent, but they also have very large strategic reserves and seem to use them actively to manage prices. What your sense is of how that works?
Arnab Datta: I have some friends who are very interested in this and they share data when they can find it, which as you said, is not easy. Generally, when I see some big development that they’ve started purchasing a lot, I try to just think about it in the context of what it means for prices going forward and whether we’ve got enough supply coming on.
There have been times when some action has happened where they started purchasing and storing a bunch of refined product. There’s always a worry about what that might mean — What are they gearing up for? I just can’t get a sense of their motivations from that. I really try to look at what it means in the broader picture for the supply-demand picture.
Jordan Schneider: Let’s come back to the legislation. What is the actual contracting change?
Arnab Datta: Put simply, the DOE can now utilize forward fixed price contracts when it’s acquiring oil for the Strategic Petroleum Reserve. In the past, prior to this rule change that happened at the tail end of 2022-2023, the regulation for acquisition basically said, “When acquiring oil for the SPR, DOE shall use a price index to account for fluctuations in market prices between the time of contracting and the time of delivery.”
What does this mean in practice? If I am DOE and I write a solicitation for a million barrels of oil nine months from now, the clearing price at delivery is not pre-contracted for. It’s based on the spot price of oil at that time. What that means is if you’re a producer and prices go up in that period, great, you’ve earned a little bit more money, but if prices go down, you'‘e at a loss. At the time that this practice was established, it wasn’t necessarily a bad thing. In some ways, it made sense. It was before a futures market. But now, particularly in the context of sending an investment signal, which is one reason why the DOE undertook this change. You want certainty. That’s the thing that producers need. They need certainty, particularly at the floor that they're going to be able to get a certain price.
The big change that DoE made in this regulation was changing that single sentence from “shall” to “may.” They can still do the market index pricing, but they’ve started to use fixed price forward contracts, which they’ve been engaging in for the better part of a year now.
Matt Klein: The type of oil that’s produced in the US by shale producers is meaningfully different from the type of oil that historically the US has imported and put in the caverns. Does the SPR make allowances for things like the sulfur content and viscosity? Are they aiming to specifically say, we want to support the US shale industry by stockpiling the light sweet versus the heavy sour? How is that process unfolding?
Arnab Datta: Unfortunately, their solicitations are all in sour crude right now. There’s a good reason for that - it’s what they have storage capacity for in the SPR. They do take sweet and they have stored it, but it can’t be commingled from what I understand. You basically need to keep the sour with the sour and that’s where the capacity is right now.
The other reason they tend to prefer to purchase sour is because our refining capacity is much more oriented towards sour. That’s a challenge that we still have to work through frankly.
If we’re thinking about incentivizing domestic investment, we’re really thinking about the shale patch because that has the shortest investment cycle - just nine to twelve months. We need to think about how we can maybe reorient some of our existing SPR caverns towards the sweet so that we can use it more as an investment signal that way.
Matt Klein: You mentioned you’re from Canada — the tar sands oil is very sour, right? Is there a possibility of doing some kind of cooperative arrangement with the authorities there to say, “Well you have the really sour stuff, we got the sweet stuff, so let’s work out a deal?”
Arnab Datta: It’s something I would certainly like to see. There are authorities and ways that you could do this. You could utilize the exchange authority. Whether it’s possible technically - the people who manage the SPR are really the only ones who know that.
But it’s a broader point that’s really important - because of how globalized this market is, we should be thinking more about how we can utilize allies and their resources to achieve a shared goal here.
One big development in the oil industry of the past couple of years is that the Dangote refinery in Nigeria finally got online and that’s purchasing West Texas intermediate - shale sweet crude. We do have capacity coming online that is able to take more of our shale patch product.
From Strategic Oil to Strategic Minerals
Jordan Schneider: How do the dynamics we’ve been talking about around energy map to critical minerals?
Arnab Datta: One thing that is similar is that a lot of the dynamics that you see when it comes to investment among producers are true for a lot of commodities, like copper and lithium. They tend to be characterized by these supercycles that are in some ways even more extreme. In the shale patch, you have a nine-to-twelve-month timeline from investment to production. In a lot of mining contexts, that time period can be years.
You don’t have a lot of price certainty. There is a risk that you’re going to start producing at a time when there’s too much supply already on the market and you’re not really going to get a great price for it. These are the dynamics that characterize all commodity investment. That's the challenge we really need to work with here.
There’s a tendency to under-invest, because the cost of underinvesting is suboptimal profits, while the cost of over-investing is potential bankruptcy.
You saw this after the big shale investment in the late 2010s when oil prices went down, but you also saw it in lithium. In the late 2010s, China had basically scaled back a bunch of its EV subsidies, so there was a huge supply glut of lithium on the market. You saw producers from Australia and Quebec basically go bankrupt because they couldn’t sell their product on the spot market in a way that was economical.
Jordan Schneider: All right, so what’s the answer?
Arnab Datta: Daleep Singh and I wrote an article about this in FT — this was before he rejoined the administration, but our idea was to reimagine the SPR as a strategic resilience reserve. The concept is that we have these vulnerabilities in other places with other commodities, whether it’s lithium, copper, steel — a whole host of things that we need for the energy transition, but also in refined products. Could we reimagine the SPR to basically address the similar vulnerabilities and disruptions in different markets?
Jordan Schneider: Let’s go a few levels deeper, Arnab.
Arnab Datta: It’s worth stepping back and just thinking - put simply, we’re trying to manufacture a lot of the tech needed to decarbonize here in the US - EVs, solar panels, small modular reactors. But to manufacture that stuff, we need a steady supply of a bunch of different commodities, let’s call them energy transition commodities. Things like lithium for EV batteries, graphite for casings for nuclear reactors and copper for electrification.
For many of these commodities, the US producers are reliant on a single source - China. They’re the only ones who really have developed these markets. They've designed them in a lot of ways around their own domestic goals and policy. So when we think about having our own manufacturing renaissance, basically from a risk management perspective, securing the supply chain for these commodities is a key challenge that we have to undertake.
If you’ll indulge me in going a little bit deeper on that, this issue runs from a market level all the way down to individual projects. If you think about energy transition, commodity demand is rising all over the world. But the supply is controlled by China from the raw ore, which sometimes they can control anywhere from 60% to 90%. Even more important is the refining capacity, which they often control 90% of. They really have a lot of the supply themselves.
What they can do is they can tactically strategically flood the market for these commodities, driving the price down globally. That both changes the market, but it makes it really difficult for new entrants to invest. If we’re thinking about domestic miners in the US, miners in Australia and Canada, it’s very difficult for them to invest when the current price of lithium, for example, is really low. So that makes it difficult. It also reduces the value of those assets, and it creates opportunities for China to go in and buy these distressed assets and continue that domination.
But aside from these market dynamics, the market infrastructure itself is also very dominated by China. If you take something like lithium or cobalt, the only physically cleared benchmark contracts are on the Guangzhou Futures Exchange. Basically, there’s a couple for some of them - I mean, there are ones on the LME as well. But if you’re looking to hedge, that’s where you have to go.
The existing cash-settled contracts are also imperfect. If China has a localized supply glut, it’s going to impact the price of those contracts. If you’re a lithium producer in Quebec and you’ve got a signed offtake agreement with Texas Tesla Gigafactory, the price index in your contract, which is normal practice, is highly correlated with Chinese local market supply. When their prices go down, the value of your contract with the Texas Tesla Gigafactory goes down. That makes it very difficult. It can make your project more distressed.
Stockpiling is an answer a lot of people are throwing around. It’s useful because you can give producers some of the certainty they need, but it’s not necessarily sufficient, because what we really need is market infrastructure. We need something more transparent.
Matt Klein: If China is the biggest market for physical lithium, changes in Chinese demand will affect the price of lithium, which will affect the Quebec-Texas trade — that makes sense to me. But I don’t understand why the fact the futures market is based in China would make a difference. Can you kind of walk me through that?
Arnab Datta: I should have been a bit clearer there. The fact that the physical contracts mostly exist in China is much more of - if you’re a producer looking to hedge, that’s where you have to go. That market is not necessarily the best governed or necessarily the most reliable in terms of meeting contract obligations. I’ve talked to producers who have tried to go there and use it to hedge, and they haven't had great experiences with the actual honoring of contracts. In 2019, a lot of these offtake contracts, whether they were on exchanges or not, just weren’t honored. Companies were left holding the bag.
Jordan Schneider: You had this cool riff on lessons from the Fed and this idea of crisis prevention versus crisis mitigation. Why don’t you talk a little bit about that?
Arnab Datta: This case, zooming out — the thing we’re really trying to deal with here is uncertainty. Uncertainty can come up in a bunch of different contexts. There are a lot of different market failures that can happen — agency issues, asymmetric information — but you're basically dealing with a bunch of different tail risks.
What we analogized to was the Federal Reserve. When it’s managing financial risks, it essentially has a toolkit that it can utilize to prevent crises and to mitigate them when they occur. We can’t predict everything, obviously, but if we can build a toolkit that’s capable of conducting operations in both those spaces, you can build something more resilient.
In the context of the Fed, crisis prevention looks like things like capital requirements for banks to make sure that if there’s some kind of a crisis, they can bring some equity to bear. Then there are a whole host of other bank regulations. In past crises, in 2008 and 2020, they’ve basically served as the lender of last resort. To keep liquidity in the system, they purchased distressed assets - I mean, indirectly, at least.
In the context of commodity markets, you have similar dynamics. You need to evaluate, in real time, if there is enough resilience in the system so that we’re getting the level of production we need. You can do that with things like forward price contracts, like put options. You can make sure that there’s enough incentive to produce to a certain level so that we don’t have a shortage.
But if it does occur, or if you get the other challenge, where if you get too much production, you can have the federal government step in and just start purchasing it and serving as the buyer of last resort so that those companies that did produce don’t end up going into bankruptcy. They have a place where they can still remain viable as companies by selling their product.
Matt Klein: Arnab, you said that Employ America got into this originally because you were rightly concerned about the possibility of an oil shortage leading to unwelcome inflation in ‘21 and ‘22, which was right. I’m curious, are you guys next going to cook up a plan for the Fed to trade commodities futures as the logical way of doing this?
Arnab Datta: We, in our paper, laid out a couple of ways this could go. One is you can cobble together a set of executive branch authorities to basically undertake the work necessary to do this. There’s some new offices that have been created. These are offices that potentially could be set up to evaluate markets in real time. There's a foundation that was established at the Department of Energy with a pretty wide ambit to support the Department of Energy’s work. Could they be evaluating what’s happening in lithium and copper and saying, we need to hit this vulnerability? Possibly.
Then you have a number of different funding streams, like the Office of Clean Energy Demonstrations. You have the Loan Programs Office. Could they be undertaking activities? We put out a recent proposal for LPO to basically capitalize a special-purpose vehicle that would engage in the trading we’re talking about to give producers the certainty they need to go produce. It’s not a perfect version of this, but it's a pretty good one, and it’s one you can do with the authority you have.
Broadly, though, I’d love to see Congress equipping an office with the capacity in terms of staff to evaluate these challenges and vulnerabilities in real time and giving them the toolkit to actually act through market channels. Basically, to both purchase either directly with producers and tie it to capital commitments, or to just purchase on exchanges would be really useful.
Regarding our SPR intervention — what is happening now is great in terms of the fixed-price contracts, but it’s still a bit clunky. Besides the sour-not-sweet issue, it happens kind of beside the market rather than through the market. If we’re thinking about market intervention, if DoE could just directly be trading WTI contracts, that’s a more efficient channel to the market. Equipping the SPR office of the DoE to basically set up a trading account at CME and start trading this stuff would be pretty useful.
They can keep pushing for an incentive to build even more storage capacity, potentially. There are new contracts being established. There’s a new company - I mean, it’s been around for a while, but there’s a company called ABX that just launched their first contract last week or a couple of weeks ago. It’s an LNG physically cleared contract. They’re looking to really make a splash in battery metals in physical contracts, which don’t really exist outside of China right now.
Jordan Schneider: We saw this work pretty well with the CHIPS program office, but that required a new organization and 200 people coming in from outside of government who have been actively playing in these markets for their careers. Curious about your take on that, Arnab.
Arnab Datta: It’s always tough. CHIPS has done a great job. One thing I will say is just stepping back to your point about the political consensus around this — it is changing.
I speak with a lot of miners and folks in the industry — they speak to Republicans about this.
More and more Republican staffers I talk to are starting to understand, particularly because these are not functioning free markets. This is a place where another country is playing a very active role, and there is an understanding that we need to respond in kind. We can leverage our skills in building financial markets to do that.
To your point about staff capacity, when we first started doing this work, advocating for this with the SPR, there was a lot of resistance from people internally because it is very different. It’s a totally different type of contracting. It’s not something that had been done. But you also do find that over time, bureaucrats do learn how to do this stuff, and they can figure it out. Ideally, you want to set up something that has the capacity to do this - like the Fed has economists and people from the banking industry who know how to set up purchasing facilities in a crisis. You want that built in, probably, but there are also ways to get it done without that.
Jordan Schneider: One thing about the other dynamic that is a little underappreciated about this is you can attract talent to do cool stuff. This is a whole lot more exciting than whatever was happening with the Strategic Petroleum Reserve ten years ago. If you allow people sort of room to run and be creative and thoughtful and use their energy in an active way — you can attract a different type of talent.
Arnab Datta: Traders are included in this context. I’ve spoken to traders who lament the fact that Joe Biden is now the greatest oil trader of all time. It’s him and Marc Rich basically, next to each other. When you think about the necessity of doing this for public purpose and that you're not going to get a percentage of the trade on the other end of it, but you're going to execute something really cool, people get excited about that.
Matt Klein: I’m reminded of something I worked on a long time ago when I was an intern somewhere else, which is the farm subsidy system we have in this country. People obviously have a lot of issues with it in various ways. Nevertheless, there’s a pretty strong cross-partisan consensus that it's good to have a whole system of supports that are all different for different kinds of crops. It's not just one system - it’s very customized for things like whether it’s peanuts or wheat or soy or whatever.
Yet that seems like something built up over 90 years. Do you see other people thinking about this commonality, like what we can learn from that and whether that is an appropriate analogy?
Arnab Datta: It’s right. The Commodity Credit Corporation is the last of the government corporations established out of the New Deal that’s still standing. That’s because farmers rely on those tend to come from red districts. You could imagine something like that for mining districts.
One important thing is that people in that context understand that we need to be creative and have flexibility in designing different contracts for different goods, as you said. What happens in the lithium market is going to look very different than graphite.
We did a series last year on why the federal government has an interest in establishing a liquid lithium benchmark. In the context of graphite, that’s not really true, because graphite is used so differently, basically in everything, that its form and function look so different that it doesn't make sense to have this commoditized version of it.
That analogy is helpful for that point as well. It takes time to build these things. After we wrote the FT article, a lot of folks started to reach out, people from mining in the defense industry and commodities that are coming into defense with the green energy transition. Building that type of coalition, though, is something that can be done. It's just that everyone needs to see something in it for themselves.
Part of that is where you get into the contractual flexibility, the financing flexibility, so that every kind of industry can get something appropriate for them. One of the problems that we have with legislating these days is something very narrow gets constructed. It's a loan authority. Loan authorities can be great. Lending right now is helpful because the cost of capital is very expensive. But it's not the only thing. It's not the thing that's going to bring a project online necessarily, because you also need a demand piece to it. So making sure that something is built that has that flexibility is really key.
Matt Klein: Right. The demand piece is why the cost of capital is so expensive.
Arnab Datta: Exactly.
Jordan Schneider: We got the inside game. We got the outside game. We've got PDFs, we've got Financial Times articles, we got blog posts, we got private memos, we got podcasts.
How do you sort of orchestrate something like this, a relatively bipartisan, technical-ish thing into existence?
Arnab Datta: Orchestrating is very generous. I'm just incredibly lucky to work with amazing people. Skandar, our executive director, was very early in identifying this as a potential problem. It helps to be first and be good and be right early on.
The way that we approach our work here, particularly in the inside-game aspect of it, is we do a lot of work and apply a lot of rigor to make sure that what we're proposing is within the realm of possibilities, both from a legal perspective, from a policy implementation perspective, and from a political perspective. You need to evaluate all of those fears and put something forward.
I was very lucky. My first job was with Michael Bennet, who's a very thoughtful senator. My first boss was a guy named Charlie Anderson, who thought through this stuff very deeply and technically as well. So I was lucky to learn that skill from him and other people in that office. When you're thinking about balancing the different pieces, the first piece of it is really getting those details right. Everything after that can follow, but it needs to have some level of viability and rigor to it early on.
Our very first version of this was the weekend after Russia invaded Ukraine. Skander and I wrote a two-page memo that we sent to NEC and NSC outlining the contours of the original proposal, which is different than what it ended up being. Then realizing we heard some positive responses to it, we got some more negative responses to it. We tried to then think about what the public case looks like.
We need to think about how to reach every different kind of stakeholder there that might have some agency in this. We started to make that public case. Then over time, when we felt that they weren’t being responsive enough, we got a little bit harsher. Then when they were responsive, we started to ease up a bit and try to find ways to work through some of the details. You need to be nimble.
Jordan Schneider: Can you expand on making sure this is actually legal and knowing when to be mean?
Arnab Datta: On the legal side, it is common practice for the White House to suggest something to an agency and for that agency to come back and say, “This is unlawful.” That happens in every administration. It is not always true, but sometimes it is. But a lot of times it’s “This isn’t how we do this now.” There’s a conception that the way we do this now is the only legal way to do this. It's evolved over time.
You need to be able to have the confidence to say, that's not always the case. Then it just goes to how strong a legal argument you're able to craft. I'm a lawyer, and I relish the opportunity to write a legal memo laying this stuff out. Generally, in the context of the spending power, that's something where the executive branch has a pretty high amount of leeway.
When it comes to contracting, the law will define and constrain what you are able to do, but it’s rarely going to say you absolutely can’t do this. With modern-day financial contracts and lending contracts, you can usually craft something that is flexible enough to meet your goal and also comply with the law.
That’s an important thing. I’m sure there are general counsels at the Department of Energy who would disagree with my interpretations of things, but at least showing that you’ve done the work also goes a long way because you want the principal, the champion who’s pushing this forward, to at least have some of that confidence as well.
As for when to be mean, it’s going to be different for everyone. We got a bit harsher in our tone. Skander went and did some podcasts where he was a bit harsher. We put out a couple of blog posts where we were critical of Secretary Granholm and what she was saying publicly. For us, our North Star was the policy response. When we felt that they weren’t being responsive and they weren’t being correct in how they were applying this, we started to get a bit harsher. Then when they were communicating with us a bit more, we eased up.
I’ll single out a guy at NEC named Neale Mahoney, who became the champion for this and really saw it through to execution. He was someone who was very responsive and he helped us think through some of the logistical challenges.
Jordan Schneider: Resources and time are limited and you can’t do everything. As you’re thinking about trading off between another one-on-one meeting with a staffer versus writing another thing, is there a heuristic? Is it all just by feel?
Arnab Datta: I don’t know if there’s a heuristic. I will say that very quickly you run into the capacity limits of one organization and a couple of Twitter accounts who care about this thing. When it comes to the things that I’m happiest to try to invest time in and hand off, it’s some of that amplification effort.
We were really lucky to talk to and get some journalists who have influence to write about this stuff at important junctures, folks like Matt Yglesias or Robinson Meyer who really get this policy and get the importance of it. That’s something where I’ll certainly invest more time because I know that’s going to reach an audience that is helpful for me.
There are two pieces specifically about this that you can go on our website and see. One is about managing logistical risk at the SPR and the other is about auction design. I would imagine that maybe ten people read the entirety of these reports. They are long, they took days to write. I’m really proud of them. They’re really good. But that work is partly to get something out there, even though no one’s going to read it. It’s helpful for people within the administration to say this is out there now, there is an argument here that is credible, that has some work behind it. I’ll put a lot of work into that even though no one’s going to read it.
One thing I would say as a piece of advice is something I wish I did more: ask people “What does your boss read?”
Getting direct time with the boss, with the actual representative or senator, is difficult. It’s certainly difficult for outside people. The way you can do that is by getting into the publications that they read themselves.
You also need to manage bad press. Marco Rubio, for example, does care about the press that he gets on the Wall Street Journal op-ed page. In the past, there have been cases where he spent a lot of time on bills and he’s been publicly supportive of them, and then just changed a day later because one bad article came out.
The Future Of ChinaTalk and Advice for Fatherhood
Jordan Schneider: ChinaTalk is basically not going to be able to pay for my incipient child with just philanthropic support and paid subscriptions. That requires me to look for sponsors that are not foundations.
Where would you two draw the lines if you were me?
Arnab Datta: I’m highly nonjudgmental about sponsorship stuff. When I was in high school, I was the kid who just didn’t care that Wilco was in Starbucks or whatever, because I just liked Wilco and I wanted more people to listen to their music.
Your product is good and people care about it, and it’s really good that you keep doing it. Other than ethically bad companies, you should just go nuts and get what you can to keep it going.
Jordan Schneider: I love the permission slip. I’ll take it. My favorite iTunes review I’ve ever gotten is, “Jordan’s a US government shill. He gets all this money from the CIA.” I wish!
ChinaTalk is a reader-supported publication. To support Jordan’s new baby, consider becoming a free or paid subscriber.
Jordan Schneider: Matt, I’m about one month out from fatherhood. What’s your advice?
Matt Klein: There’s a really good book, later adapted to a movie, called Lone Survivor. It’s about a Navy SEAL in Afghanistan. The whole first part of the book is about Navy SEAL training. At the culmination of the training, they go through something called hell week. This is when overwhelmingly most people drop out and don’t become official Navy SEALs.
The challenging part isn’t that they ask them to do anything harder than they’ve been doing. They’ve already done long runs and other tasks like being chained down in a pool to see how long you can hold your breath underwater. The hard part about hell week is that you're doing the normal stuff while being sleep-deprived for basically the entire week. They also play loud noises at you, including the sounds of crying babies.
What this means is — you are going to be a Navy SEAL soon, so that’s kind of cool.
The other thing to think about is, if it seems really hard, a lot of other people have managed to survive it. You’re in good company.
Jordan Schneider: That was the best line from the baby class you take before delivery. I was in this room with all these overachieving couples in their early thirties, and then the head nurse who was doing the class said at the end, “Some teenagers have kids. I have faith in you. It's gonna be okay.”
Arnab Datta: I used to teach kindergarten, and I had a couple of kids who had close to teenage parents. They turned out fine.
Jordan Schneider: How has your experience in a kindergarten classroom helped you in your policy entrepreneurship journey, Arnab?
Arnab Datta: Being able to address every level of immaturity or maturity is helpful, especially with Congress and people in Congress, from baby staffers to representatives and senators themselves.
Is China’s model of economic development compatible with democracy? Were events like Tiananmen Square and the rise of Xi Jinping inevitable under China’s governance structure?
Welcome back to part 2 of our interview with Yasheng Huang 黄亚生, the author of The Rise and Fall of the EAST: How Exams, Autocracy, Stability, and Technology Brought China Success and Why They Might Lead to Its Decline.
In this installment, we discuss…
The aspects of imperial China’s governance Mao chose to embrace, and those he chose to abandon,
The factors enabling Mao’s radical policies compared to imperial rulers,
Why China was able to grow so much faster than India, despite the setbacks of the Cultural Revolution,
Statistical approaches for evaluating the effectiveness of autocratic development models,
China’s economic reforms and rural development policies in the 1980s,
How the events of 1989 permanently altered China’s trajectory,
Whether the rise of Xi Jinping was inevitable.
Apple Podcasts:
Spotify:
Co-hosting today is Ilari Mäkelä of the On Humans podcast. Here’s part one if you missed it.
Xinhai revolutionaries ambush Qing forces at Nanking in 1911. Painting by T. Miyano.
Mao’s China — Dynastic Traditions and Leninist Innovations
Jordan Schneider: Last week, we discussed Chinese history up to 1949. Ilari, could you give us a brief recap of our first episode?
Ilari Mäkelä: To understand China’s history, Yasheng Huang emphasizes the importance of focusing on the little-studied period after the collapse of the Han dynasty in the third century. During the next few centuries, China resembled Europe, with no single power base and competing ideologies. Huang’s studies have shown that this was also when China was at its most inventive.
Unlike Europe, this political fragmentation ended when China was unified again in the 6th century under the Sui dynasty. Everything changed with the introduction of civil service exams on a large scale. These exams allowed a significant portion of the population to participate and created stability. Many Chinese believed that through hard work, they could gain political power.
While this system prevented the development of a European-style aristocracy, it funneled all intellectual efforts into serving the emperor. Consequently, there were no power struggles like Henry VIII versus the pope, nor was there an Enlightenment-style intelligentsia. Instead, China produced numerous Confucian-educated civil servants.
Jordan Schneider: Professor Huang, what aspects of this imperial tradition did Mao adopt or reject when he began to rule in 1949?
Yasheng Huang: Mao took over China in 1949, inheriting certain aspects of imperial China while departing from others. He maintained and extended autocratic controls, implementing strong economic controls through central planning, state ownership, and government price-setting. Politically, communist China’s power reached beyond the county level, extending to villages through land reforms and the communist system.
However, Mao actively went against imperial tradition by de-emphasizing bureaucracy and meritocracy. This shift began in 1957 and intensified during the Cultural Revolution in 1966. Universities were closed, examinations and expertise were downplayed, and ideology was elevated. While ideology was important in imperial China, Mao’s approach differed significantly, relying on political and ideological studies rather than the traditional examination system.
Jordan Schneider: It’s interesting that you argue the more codified a system is and the fewer power bases that exist, the more important the leader’s actions become. However, even powerful Ming or Song dynasty emperors couldn’t have executed the level of radical changes that Mao implemented from the mid-1950s through the end of his reign. Mao seemed to combine the worst of both worlds — centralized personal power without bureaucratic and cultural safeguards that would have prevented disasters like the Great Leap Forward or the Cultural Revolution.
Yasheng Huang: Excellent observation, Jordan. Two additional factors contributed to this difference.
Modernity — imperial China lacked modern means of transportation and communication. Indoctrination was the primary tool, and once officials were dispatched to remote regions, central control diminished. In contrast, modern systems allow for instantaneous and ongoing control through air travel, trains, and buses, facilitating frequent indoctrination sessions in Beijing.
Transformative agenda — imperial China, like many traditional polities, focused on maintaining the status quo. They often lacked knowledge of potential transformations or how to preempt them. Mao, however, had a highly transformative agenda. In 1958, he famously aimed to surpass Britain and America in steel production — a concept foreign to Ming dynasty emperors.
Communist China was more ambitious and transformative, aided by Leninist techniques and organizational capacity. It wasn’t a carbon copy of imperial China, but incorporated elements borrowed mostly from the Soviet Union.
Vladimir Lenin’s theory of the vanguard of the revolution emphasized the ability of political elites to rapidly transform society. While Karl Marx predicted that communist revolutions would occur after capitalism, Lenin argued they could be launched directly from feudalism. This transformative agenda was absent in imperial China but central to Mao’s rule.
Jordan Schneider: The contrast between the Ming dynasty and Mao’s Leninism is fascinating. You argue that when Marco Polo arrived in Yuan Dynasty China, he felt transported 500 years into the future. During the voyages of Zheng He 鄭和, the Chinese viewed other cultures as backward and lacking.
Human nature may play a role here. When elites feel they’re at the top and things are stable, they may be less inclined to explore, change their minds, or question whether their system has all the answers. In contrast, from the fall of the Qing dynasty through the Republican period to Mao’s era, there was a keen awareness that China needed to catch up with the rest of the world.
To what extent is this perspective determined by China’s relative position as more or less advanced than other societies?
Yasheng Huang: I agree that the factor I mentioned is one of many contributing to this perspective. However, even given the disparity between East and West 500 years ago, it doesn’t fully explain China’s isolationist stance. The mental model you describe likely applies to the political elite who, upon seeing nothing impressive abroad, decided to remain insular.
Today, people from wealthy countries often explore poorer regions out of curiosity. When Christopher Columbus embarked on his voyages, he had personal motivations and potential gains alongside nationalistic agendas. This personal stake was entirely absent in Zheng He’s voyages, which were purely state-sponsored activities with national objectives.
Jordan Schneider: That’s an apt comparison. European colonization in the 18th, 19th, and 20th centuries — despite the lack of respect for native populations — still involved scientists and observers who documented their experiences. This information flowed back into the broader ecosystem. In contrast, China lacked this type of engagement for over a thousand years.
Autocratic vs Democratic Development Strategies
Jordan Schneider: Ilari, would you like to bring us back to Mao?
Ilari Mäkelä: One interesting aspect is the package Mao inherited from the imperial era (albeit in a mutated form) is the autocratic tradition, which we often view as partially negative.
One of the most fascinating questions in comparative economics is why China has grown so much faster than India. Some argue that democracy doesn’t work well for such a large country as India, suggesting that autocracy, like in China, is necessary. You have a different perspective, focusing more on China’s human capital resources.
I recall you mentioning in your TED talk that China grew at a pace similar to or even faster than India, even during Mao’s era. This suggests that growth occurred despite policy, not because of it. You attribute this to an inherent push for growth in China, absent in India, linking it primarily to literacy. Is this correct?
Yasheng Huang: The baseline judgment on China and India is based on almost a single data point from the World Bank website — the GDP per capita of China and India in 1980 or 1990. It shows that India had a higher GDP per capita than China at that time. Twenty or thirty years later, China’s GDP per capita surpassed India’s. Many people used this data point to argue that autocracy (China) could grow faster than democracy (India).
However, there are a couple of problems with this conclusion. First, the data is almost certainly wrong. One issue with Chinese GDP data is that they used a Soviet system called the material product measure, which didn’t include the service sector. In poor countries without industrialization, you have agriculture and a sizable service sector. For many years, the service sector was very large in the Indian economy.
Ilari Mäkelä: But isn’t it still true that China has undergone a more dramatic transformation than India since the 1980s?
Yasheng Huang: Absolutely. The point I was making earlier disputes the initial discrepancy. Given that China grew faster than India, it’s actually more impressive if you believe that China started from a higher per capita base. However, there are many differences between China and India. It’s puzzling why people immediately jump to the conclusion that it was because India was a democracy and China was not.
To properly evaluate this, we need to look at global evidence to compare at least two competing hypotheses, which are human capital and the political system. Academic literature increasingly shows that autocratic and democratic systems don’t have a significant impact on growth rates. Some autocracies grow very fast, like China, Taiwan, and South Korea in the 1960s and 1970s. There are also spectacular failures, such as the Philippines, Indonesia, and many African countries.
The problem with the formulation by many Chinese and some Western intellectuals is that they escalate a factual statement about China growing faster than India to a statement about the superiority of autocratic systems.
When discussing systems rather than specific countries, you need evidence from other countries with the same system as China. That’s just common sense.
Jordan Schneider: Where do you think this argument comes from then?
Yasheng Huang: Some people believe in autocracy and are grasping for evidence to show their belief has economic benefits. I often point out to them that China was also autocratic under Mao, but it didn’t grow very fast during that period. An honest person would conclude that there must be something else different between Mao and Deng’s eras. Perhaps globalization, open policies, or private sector development played a role.
I’m dissatisfied with media coverage because they often go for sensational statements without examining the logic behind people’s views. In academic seminars, these arguments wouldn’t hold up for five minutes because basic logic would demolish them. Unfortunately, media attention often goes to those making broad, sensational statements rather than those presenting logical arguments and evidence.
Subscribe to ChinaTalk to support non-sensationalist journalism.
Liberalism and the Future of Chinese Democracy
Ilari Mäkelä: One reason for the hyperbolic “it’s because of autocracy” conclusion might be that some Westerners have inadvertently contributed to this by repeatedly claiming that autocracies cannot grow fast. When faced with a fast-growing autocracy, it’s easy to jump to the opposite extreme conclusion. Would it be fair to say that there are certain conditions under which autocracies can grow fast? For China, factors like the imperial legacy of civil service exams (Keju) leading to high literacy and human capital, and a post-Mao autocratic political environment that was more conducive to growth due to increased pluralism and diversity, played significant roles?
Yasheng Huang: That’s a very good observation, Ilari. In my current book, which is a revision of my 2008 book, I argue that the right way to look at China or any country is to focus on the direction of movement rather than static conditions. That is, we need to look at whether the is country moving towards more openness or more closure.
It’s not accurate to say that China couldn’t grow because it’s not a Western democracy, or that it could grow because it’s not a Western democracy. The key is to look at the movement of both Eastern and Western countries. Are they moving in an open or closed direction? In statistical terms, this is called a difference-in-differences approach.
“The future prospects of the village 农村远景图,” Great Leap Forward poster, 1958. Source.
Under Deng Xiaoping, China was undoubtedly an autocracy, but it was moving away from Mao’s totalitarianism towards a more authoritarian society. For people living there, this shift made a huge difference. The incentive effect comes from that movement.
Imagine an entrepreneur in 1978. They wouldn’t be deterred by the lack of Western-style institutions. Instead, they would be encouraged by the fact that they could now start a business without fear of arrest, unlike in 1976. The incentive effect came from China becoming less autocratic, not from being autocratic.
As a policy question, the real issue is whether to move the country more towards or away from autocracy, even if the decision isn’t about fully embracing democracy.
Jordan Schneider: Can we relate this to your personal history? If you were born five or ten years earlier, do you think you would have even bothered learning English? How did this change in the political climate reflect in your youth, particularly in terms of learning and the drive to acquire knowledge?
Yasheng Huang: In my personal case, I was always interested in knowing more rather than less, regardless of the environment. After coming to the US, I spent hours in the library reading Western writings about China, particularly about the Cultural Revolution and the Great Leap Forward, because I couldn’t access that knowledge in China.
Regarding business motivations, people need to feel secure to make investments. If you know you’ll be arrested immediately after making an investment, you won’t do it. In my book, I emphasize the importance of not underestimating the incentive effect of not being arrested. Many people focus on China’s lack of secure property rights law, neglecting the fact that Deng Xiaoping significantly increased the personal security of the Chinese people. This increased security was a crucial factor in encouraging Chinese entrepreneurs to start businesses.
Ilari Mäkelä: The era after 1978 is indeed special and quite different from everything else that happened under the CCP. There wasn’t a significant rural-urban gap at the time, with rural areas growing as fast as urban areas. There was also a diversity of voices within the party, ranging from old-school economic planners to reformers like Zhao Ziyang 赵紫阳. Since you lived in China in the 1980s, could you share your personal experience of that period?
Yasheng Huang: I first came to the US in 1981, but I returned to work for the World Bank in China during the 1980s. However, my understanding of that period didn’t come from my lived experience. Like many Western observers, I didn’t have a clear view of the 1980s until I began working on my 2008 book.
After publishing a book on foreign direct investment (FDI) in 2003, I was left with the question — what was driving Chinese growth in the 1980s, before FDI became crucial? This led me to research the 1980s extensively for my 2008 book, Capitalism with Chinese Characteristics, which is also summarized in my 2023 book.
I discovered that reforms in rural China went far beyond agricultural reforms. There were contracting reforms, land reforms, village elections, and even rural financial liberalization. I was astonished by how proactive the Chinese leadership was about rural financial liberalization. Top officials from various banks were issuing support for these reforms.
Large-scale financial institutions developed extremely fast in rural areas during the 1980s. This affected 800 million people, as China was 80% rural at the time. Yet, Western scholarship largely ignores this period. The prevailing narrative is that China grew without financialization or with a state-owned financial system, disregarding the extensive documentation and statistical analysis showing otherwise.
It’s frustrating and saddening that there’s so little curiosity about this crucial period in China’s development, despite its immense impact on hundreds of millions of people.
Ilari Mäkelä: You’ve piqued my curiosity. Before we discuss what happens next and why this trend doesn’t continue, I’d like to ask a question. I mentioned earlier that one of the interesting aspects of the ATS is that it’s not always a matter of the CCP making decisions based on a grand strategy. Often, it’s because there is no grand strategy. There are internal power struggles, such as Deng Xiaoping versus Chen Yun 陈云, or Zhao Ziyang versus Li Xiannian 李先念. Based on your research into those documents, to what extent was this a unanimous decision in Beijing versus the liberal factions of Hu Yaobang 胡耀邦 and others being able to operate under the cover of confusion?
Yasheng Huang: People like Hu Yaobang and Zhao Ziyang didn’t suddenly decide to implement financial liberalization. What’s remarkable about 1980 is threefold.
There was political fragmentation at the very top, as you mentioned.
The policy ethos was focused on tangibly improving people’s lives rather than solely on GDP growth. Any policy or practice that improved people’s lives was considered the right approach by definition. For example, small-scale sellers on Beijing streets selling chickens and various goods were allowed because it improved their lives by providing income.
There was a prevalence of initiative-taking. A concrete example is the Rural Credit Foundation, an almost national-scale rural financial institution. The central bank refused to recognize it, fearing disruption of financial control. However, the Minister of Agriculture stepped in and provided legitimacy. Such local bureaucratic initiative is unimaginable in China today or even in the 1990s.
These practices weren’t initiated by Deng Xiaoping, Zhao Ziyang, or Hu Yaobang. Deng Xiaoping famously said about township and village enterprises, “The Central Committee of the Chinese Communist Party takes no credit for it. It was a total surprise to us.” Essentially, there was space for organic and spontaneous practices, which almost entirely disappeared, especially in rural China, after 1989.
Jordan Schneider: Let’s discuss 1989 then. You have this concept of axiomatic legitimacy, where after 1500 years of imperial rule over the broader Chinese landmass, people expect and defer to the legitimacy of the center, except during rare occasions like dynastic falls or major invasions. Professor, to what extent do you believe a counter-revolutionary reaction was inevitable?
Yasheng Huang: I don’t believe 1989 was inevitable. I concede that China’s post-1989 direction was more likely than before, but it wasn’t 100% determined by axiomatic legitimacy. Every culture starts with certain presumptions. In Chinese culture, due to history and autocratic rule, there’s a presumption that whatever the government does is legitimate and beneficial to individual welfare.
This view was present in both the 1980s and 1990s. The difference lies in the ability to question and debate these presumptions in the 1980s. Sometimes, you might end up with the same view you started with, which is fine. However, the opportunity for discussion and inquiry existed.
In contrast, without freedom of discussion or what Amartya Sen called “discussion democracy,” people tend to maintain their initial views without examination.
This is evident today, where many young Chinese people, even those living in the West, refuse to engage in different mental processes and cling to their axiomatic views.
Jordan Schneider: The real question is whether a discussion democracy can exist under CCP rule. Is it possible to have a system where people like young Yasheng Huang can question the legitimacy of Beijing’s rule? Can a governing body be comfortable with this level of questioning, or was the 1980s an anomaly due to the aftermath of Mao’s era, which ended when the system responded to people asking too many questions?
Yasheng Huang: I’m of two minds on this issue. It’s debatable whether 1989 was an inevitable result of a looser autocratic system or a consequence of individual decisions by leaders and students. When I study history, I’m struck by how often individual events shape its course.
Looking at Taiwan and South Korea, both experienced massacres but didn’t reverse their economic and social progress. At some point, their leaders decided to let go and allow change.
Taiwanese students protesting for democracy at Chiang Kai-shek Memorial Hall, March 1990, during the Wild Lily movement 野百合學運. Source.
Today, Taiwan and South Korea are vibrant democracies, arguably performing better in some aspects than Western democracies. During the pandemic, for example, people in these countries trusted their governments more readily.
Regarding China in the 1990s, it wasn’t necessarily imperative for the regime to reject all political reforms proposed by Zhao Ziyang. Hu Jintao moved China further in a statist direction, and Xi Jinping continued this trend. I attribute much of the blame to the leadership that came to power in China after 1989, whom I refer to as the “Shanghainese” due to their origins and viewpoints.
This new leadership focused on globalization, privatization, and urban development while neglecting rural areas. They viewed the countryside as a burden, believing peasants couldn’t contribute significantly to GDP growth or technological development. This perspective shaped the “China model” of urban development, high-tech industries, and infrastructure that we see today. The rural reformers, including Zhao Ziyang, were sidelined or marginalized in the aftermath of Tiananmen.
Ilari Mäkelä: I’d like to challenge your earlier statement that 1989 wasn’t that important. It seems it could be the most crucial event. The conservative old guard wanted to maintain their way, while Deng Xiaoping aligned with liberal reformers like Hu Yaobang and Zhao Ziyang. The Tiananmen Square protests allowed the conservative faction to drive a wedge between the political and economic liberals and Deng Xiaoping, who appeared to be an economic liberal but a political conservative.
This shift resulted in everything you described, but it also led to another significant change. You mentioned the ideologically stubborn youth in China today. Some argue this is also a post-Tiananmen symptom. The government decided to emphasize ideological training over economic focus. New curricula were introduced, emphasizing national humiliation. Wang Zheng’s 汪铮 book, Never Forget National Humiliation, discusses how this phrase became a classroom mantra for Chinese children.
The words on the board say, “Never forget national humiliation, unyieldingly defend our homeland 勿忘国耻，强我国防” and, “September 18th incident 九一八事变,” which refers to the Japanese false-flag operation staged as a pretext to invade Manchuria.
The narrative of Chinese humiliation and rejuvenation under the Communist Party became a way to prevent another Tiananmen Square incident. By teaching children about the horrible past and the great present, while instilling suspicion of foreign influence, the government shaped a new generation. The Xi Jinping era seems to be a natural progression of two trends — nationalism in textbooks and the lack of political pluralism in the CCP.
Yasheng Huang: I apologize if I wasn’t clear earlier. I absolutely believe that 1989 was extremely important. In fact, I think it was more significant than many scholars have acknowledged. They usually recognize its political impact but overlook its economic consequences. In my revised 2008 book, I presented statistics demonstrating that Tiananmen also had substantial economic effects.
The question is whether it had to turn out this way. There’s an unconfirmed rumor among Chinese intellectuals that Deng Xiaoping considered inviting Zhao Ziyang back at one point. If true, and if Zhao had returned, would China have taken such a divergent path? My observation is that while there was a 90% chance of China ending up this way, there might have been a 10% chance that leaders weren’t as constrained by 1989 and could have continued reforms.
Taiwan and South Korea, for instance, had their own massacres but managed to move on. Perhaps some distance from the event is necessary. However, China remains stuck in a Tiananmen mindset long after the event, which I believe is largely, but not entirely, determined by Tiananmen.
Jordan Schneider: The key issue, relating to your lifespan argument, is that reform in Taiwan and South Korea occurred after the death of their autocratic leaders. China’s problem was that it had more people with charismatic, revolutionary legitimacy who lived exceptionally long lives. If Deng Xiaoping and Chen Yun hadn’t been around in the late 1980s, the slightly younger generation, more open to different political perspectives, might have had more power. My counterfactual is that if Tiananmen had happened in 1994 or when these older leaders were more incapacitated or absent, the balance of power within the system might have been different. Even after Mao’s death, there were still enough people with strong beliefs in the old system and charismatic credibility to maintain their influence, preventing the new generation from asserting themselves.
Yasheng Huang: I agree with your analysis about the overlapping generations, which was absent in Taiwan and South Korea. Interestingly, the Cultural Revolution profoundly affected many revolutionary elders. While many were staunch conservatives, a few were quite liberal – more so than the technocrats who came to power in the 1990s. They were willing to think about issues systemically, which is how I define liberalism: considering solutions at a system level rather than focusing solely on leadership changes.
Take Xi Zhongxun 习仲勋, Xi Jinping’s father, for example. He was liberal not just in supporting liberalization but in making procedural arguments. At the Politburo meeting that ousted Hu Yaobang, he argued that the process violated agreed-upon rules. This was a revolutionary way of thinking in such a system, more profound than simply advocating for a market economy or democracy. His insistence on process and rules was more revolutionary than economic liberalization, although ultimately, he was defeated.
The lesson China should have learned from the Cultural Revolution was that it wasn’t just Mao’s mistakes, but the system that allowed him to make those mistakes.
Earlier Chinese Communist Party documents reflecting on that period showed this understanding. They instituted some separation of powers, with the presidency and party secretary general held by different individuals. However, this lesson was forgotten after 1989 when Jiang Zemin was brought to Beijing. Worried about his weakness, they abolished other power centers to consolidate his authority, leading to the China we see today.
Jordan Schneider: Professor, I searched for your name in Chinese and found many English language interviews about your book, but only one in Chinese. Have you considered starting your own YouTube channel or podcast series?
Yasheng Huang: I’m not very good at that, to be honest. It’s worth noting that before Xi Jinping, and even during his first term, I did many interviews with Chinese media outlets. I could go into great detail about data and evidence, much more so than with Western media. I was very impressed with the quality of Chinese financial journalism. However, the situation has changed now, not because the quality has declined, but because the political environment has shifted.
Ilari Mäkelä: Let’s examine China today then. In some ways, Xi Jinping seems to be a natural progression of these trends…
In part 3 we discuss the rise of Xi, succession dynamics, and prospects for liberalization in China’s future
We discuss…
The Steelman case for why China needed a leader like Xi Jinping,
What sets Xi apart from his predecessors,
Succession challenges in authoritarian states and likely scenarios post-Xi,
Why engagement with China failed to produce political liberalization,
How the US could have better leveraged economic relations with China,
Creative approaches to human rights advocacy in China.
To discuss Trump’s VP pick, ChinaTalk chatted with Employ America’s Arnab Datta and economist Matt Klein, who writes the excellent Overshoot newsletter.
We get into…
Vance’s beliefs about the economy - from dollar hegemony to tariffs and industrial policy,
Psychological dynamics at play within the Trump-Vance ticket,
The case for weakening the US dollar, and how the executive could make that happen.
Spotify:
Apple Podcasts:
The Death of Dollar Hegemony?
Jordan Schneider: Matt, kick us off – what can we expect from JD Vance economically? I hear he’s read your book, Trade Wars Are Class Wars: How Rising Inequality Distorts the Global Economy and Threatens International Peace.
Matt Klein: You can see this in the questions he asked Federal Reserve Chair Jerome Powell — “Why should the dollar be the reserve currency? Why would we want it to be the reserve currency? Is it a problem that foreigners like using the dollar?”
This indicates that he and his team have read and absorbed some points from the book I wrote with Michael Pettis. We argued that while the US economy is large globally, the use of the dollar vastly exceeds the size of the US economy.
Concretely, our economy is between 20-25% of the world, but our currency and financial system meet the needs of 80% of the world. That leads to big mismatches between what makes sense for global savers and investors versus Americans.
There’s an argument to break that link — to have the dollar used mostly by Americans — and have the US financial system focus on what makes sense for Americans, not on creating assets for foreigners. This is an argument from our book which Vance seems interested in.
Jordan Schneider: If a president wants to devalue the dollar and ensure the US is not the global reserve, how would they go about realizing that practically?
Matt Klein: If your goal is to make US assets less accessible and attractive to foreigners, legislation would help. Without it, you could arbitrarily enforce the rule of law, use sanctions, or confiscate wealth. Many countries have trouble attracting foreign investment by doing this unintentionally.
Bipartisan legislation was proposed five or six years ago to give the Federal Reserve authority to impose a tax on any foreign purchase of US assets. The Fed could adjust the tax based on their judgment of foreign demand for US assets to balance financial inflows and outflows.
Another option, used by many countries to lower exchange rates, is to have the central bank buy foreign assets. You create dollars and buy something else. The Swiss National Bank has bought dollars and euros, including stocks, to manage exchange rates. Singapore’s Monetary Authority targets the exchange rate instead of setting interest rates.
There’s no inherent reason why the Fed couldn’t do this, but legislation would help.
To change the wording — the dollar is not strong, it’s expensive. It’s overpriced.
Jordan Schneider: What would that do to the global and American economy?
Matt Klein: For the US economy, if you make the dollar less expensive relative to other currencies, Americans should find US-made goods relatively cheaper than foreign-made goods. People outside the US should find US-made goods more affordable. This should increase buying of US-made products and decrease American buying of non-US products.
Second-order effects could include increased profits, investment, and hiring in export industries, while import-dependent businesses might do less well. Globally, companies competing with US firms might lose out.
This could create an inflationary impulse, which sometimes is welcome. The Bank of Japan, Swiss National Bank, and Swedish Central Bank have intervened in currency markets to generate inflation when it was too low.
The actual impact could be more complex. Some countries, like China, use sterilization techniques to offset some effects of currency interventions.
Jordan Schneider: What are the first and second-order global effects?
Matt Klein: It partly depends on the magnitude. A 10-20% move isn’t unusual over a multi-year span. For larger moves, there’s debate among economists. Some argue a stronger dollar is bad for emerging markets because it tightens financial conditions. Others say a weaker dollar hurts exports from these countries.
The net impact is unclear and context-dependent. Some argue it doesn’t matter because international trade is invoiced in dollars anyway, but this ignores the eventual conversion to local currencies for profits.
A massive move could lead to asset sales and potentially higher US interest rates. Usually, significant currency moves are responses to other factors, like the dollar’s appreciation during the financial crisis due to widespread dollar-denominated debt.
Jordan Schneider: Let’s talk tariffs. Matt, did Vance learn anything from your book in this department?
Matt Klein: We actually wrote that many arguments for tariffs don’t make sense in a world of floating exchange rates. The textbook argument is that tariffs penalize exporters because imports remain constant while the exchange rate appreciates, making exports less competitive.
If tariffs act as tax increases and tighten the government budget, it could reduce American spending and the trade deficit, but by making Americans poorer. Some in the Trump administration have suggested using tariff revenue to offset tax cuts elsewhere, complicating the net effect.
Jordan Schneider: All right, let’s talk industrial policy.
Arnab Datta: I’ve worked with Vance’s office on some initiatives. He’s thoughtful and diverges from Republican orthodoxy, particularly on spending. He’s willing to invest in strategic sectors. The question is how much he’ll define that conversation in a future Trump administration.
Even in areas with bipartisan consensus, like next-gen geothermal energy, many Republican senators support the industry but don’t want to spend money. I’m curious how a future Vice President Vance might influence thinking on these issues.
Jordan Schneider: He has said he supports the CHIPS Act, framing it in an isolationist context about reducing dependence on Taiwan. He has a bias toward treating manufacturing as the only “real” part of the economy.
Matt Klein: He didn’t come from Silicon Valley. He came from Ohio.
Arnab Datta: I’m always curious about what Vance actually thinks because he tends to look for the most partisan aspect of an issue. The semiconductor example you used about not wanting to protect Taiwan is one version. Another is his argument about immigrants driving up housing prices, which has no evidence. I’m never quite sure how to evaluate these things.
Jordan Schneider: He’s been playing that game really aggressively for the past eight years
The Psychology of a Trump Apprentice
Jordan Schneider: We’ve done a lot of Shakespeare emergency episodes over the past few months, which explore the question of how well politicians can really “know themselves.”
Vance is clearly very smart and thoughtful. But he has also been able to put on the clothes of some guy who does great on Tucker Carlson, and he’s been completely inhabiting that role for a number of years now. Does that end now? Is it in his marrow at this point?
Matt Klein: It’s not Shakespeare, but there’s a relevant line in “A Man for All Seasons.” The line is: “But for Wales” - basically, you’re giving up all of your principles for a little bit of power.
Sir Thomas More: There is one question I would like to ask the witness. That’s a chain of office you’re wearing. May I see it? The Red Dragon. What’s this?
Judge: Sir Richard is appointed Attorney General for Wales.
Sir Thomas More: For Wales. Why Richard, it profits a man nothing to give his soul for the whole world. . . but for Wales.
Jordan Schneider: But the difference is, Vance now has a good chance of being president. Not just if Trump dies, but running in four or eight years. Your Faustian bargain here has the potential to have a pretty high payout from a power acquisition perspective.
There’s been some reporting about how all his staffers are really tall and smoke a lot. Can we get a vibe check on the broader energy of the JD Vance universe?
Arnab Datta: That is not something I have experienced. The folks that I know from his world are average height, and as far as I know they don’t smoke. But I don’t hang out with them enough to know that for sure, I guess.
Jordan Schneider: Fair enough. One interesting discovery from my aggressive googling over the past 48 hours is that his wife spent a year in China after college as a Yale China fellow, teaching at Sun Yat-sen University.
Matt Klein: I mean, Matt Pottinger spent a long time in China, too.
Arnab Datta: My big takeaway from this is Yale Law School just keeps winning, controlling the country every other way.
Jordan Schneider: Anyway, it’s reported that everyone wants to have this really tight relationship with Trump and ask him for favors all the time. In a Shakespearean way, it seems like Vance was very strategic about this. He didn’t even ask Trump to endorse him. The only thing he asked for was an endorsement of a transportation safety bill or something.
If and when Vance is sitting in the White House and thinks “This is my time to spend some political capital and push something with Trump,” I wonder if Trump will end up seeing his own mortality and feel threatened.
That mentor-protégé dynamic may get a lot more complicated as time goes on, perhaps becoming more contentious and dark.
Arnab Datta: I’m just thinking about Trump in a mentor relationship, and that sounds not great.
Is he gonna get annoyed when Vance gets some good press and try to cut him down a little bit? I don’t know.
Jordan Schneider: Well, if he starts pushing things Trump doesn’t want to be pushed, then all of a sudden he goes from surrogate son to someone who’s trying to do unpopular things that are obnoxious. Matt, can you talk about this better than I can?
Matt Klein: I’m just surprised that someone who’s gone as far as he has in talking about January 6 and so forth would bother to push Trump in a place that Trump wouldn't want to go anyway.
Jordan Schneider: Perhaps it’s more strategic for him to hide his talents and bide his time.
Matt Klein: There’s nothing to hide. He’s been very clear over the past few years about what he thinks about the Constitution and what he thinks his role as a senator is and so forth.
ChinaTalk is a reader-supported publication. To receive new posts and support our work, consider becoming a free or paid subscriber.
We don’t do a ton of econ coverage on ChinaTalk, but had to make an exception for this truly epic piece of independent research by Jon Sine on LGFVs. He writes the Cogitations substack.
In this essay we will tell the story of China’s LGFVs: that special type of state-owned enterprise (SOE) behind China’s infrastructure building bonanza. Along the way, we will look at the political and economic processes in China that facilitated their rise, and assess what we know and what we don’t know about them.
Data Before Narratives: The Lay of the LGFV Land
Let us begin with data. Today, there are nearly 12,000 LGFVs in China, 3,000 of which publicly disclose financials.1 Collectively, they are gargantuan. As of 2020, according to bottom-up surveys of LGFV financials, aggregated assets and liabilities equal 120% and 75% of China’s GDP, respectively.
LGFV Assets and Liabilities
IMF, Local Government Financing Vehicles Revisited, February 2022, https://www.elibrary.imf.org/view/journals/002/2022/022/article-A003-en.xml
LGFV’s financial situation is, to put it frankly, very bad. In aggregate, earnings (before interest, taxes, and depreciation, i.e., EBITDA) do not cover even their interest payments. Including government subsidies only occasionally pushes the interest coverage ratio above one. Moreover, the average borrowing cost for LGFVs, 5% or so, far outpaces their 1% return on assets, posing obvious sustainability problems. 2
Interest Coverage (LHS) and Borrowing Rates (RHS)
Source: WIND, Jameson Zuo and Siqi Lin, “Assessing the Overall Debt Risk of Chinese Local Government Financing Vehicles,” CSPI, September 2023, https://www.arx.cfa/~/media/834250EAA848452D82966537A082CAE5.ashx
Cash flows paint an equally troubling portrait. Every year 80 to 90 percent of LGFV spending is funded by new debt. On the whole, LGFVs operating inflows do not come close to covering operating expenses. New debt is routinely added simply to make up the gap and sustain current operations.3
LGFVs Cash Flows
IMF, Local Government Financing Vehicles Revisited, February 2022, https://www.elibrary.imf.org/view/journals/002/2022/022/article-A003-en.xml
As is predictable from the above financials, the stock of interest-bearing LGFV debt has just about unceasingly expanded.4According to statistics from the IMF, LGFV’s interest-bearing debt has grown from 13% of GDP, or RMB 8.7 trillion, in 2014 to 48% of GDP, or RMB 60.4 trillion, in 2023.5
China’s Interest-Bearing LGFV Debt
Source: IMF Article IV Consultations for China, years 2018. 2019, 2021, and 2023.
LGFV interest-bearing debt is even larger than the IMF data above suggests. An unavoidable limitation of assessing LGFVs via bottom up data, as all of the above sources do, is that it only captures the 2-3,000 LGFVs that have issued bonds and published associated financials. Another 9-10,000 smaller LGFVs have never accessed the bond market and are therefore simply missing from the data.6 A simple way of trying to estimate the rest of the interest-bearing LGFV debt is to assume LGFVs follow a Pareto distribution.7 Doing so suggests IMF estimates probably understate debt by 25%. A more reasonable, if still likely conservative, estimate of interest bearing LGFV debt is probably 60% of GDP, or RMB 75 trillion, in 2023.8
Things Data Can’t Say
While the data tells a story of its own, there is also much it cannot tell us about LGFVs.9 Staring at data and getting proscriptive according to normative neo-classical doctrine risks turning us into IYIs.10 Understanding how we got here and the likely path forward requires other inputs.
What happens if we bring in the complexities of political economy? The incentives of organizational and social systems? The complexities of the past may shed light on the road ahead.
A Single LGFV Can Start a Prairie Fire
The protagonist of our story was born in a little known city in central Anhui province called Wuhu. The city, situated on the bank of the Yangtze, is just three hours south from another little known place with an important legacy: Anhui’s Xiaogang county, the place credited with pioneering China’s household responsibility system in 1978. Twenty years later, Wuhu would join its Anhui brethren in pioneering another defining feature of “socialism with Chinese characteristics”: the LGFV.11
If the household responsibility system unleashed market forces into China’s economy, then the LGFV brought the state surging back in—something Party-state leaders would likely have been keen on following destabilizing bouts of inflation in the late 80s and early 90s.12 But the direct desires of leadership were not the proximate cause for this novel feature of socialism with Chinese characteristics.
The LGFV, as we will see below, rather arose as an indirect result of three conjoined and systemic issues: fiscal centralization and decentralized local developmentalism, the Party-state’s organizational and incentive system, and the weakly institutionalized nature of China’s political system.
A spark only starts a fire in a flammable environment.
The Confederated States of China
In 1781, the United States was governed under the Articles of Confederation. That system had a glaring flaw: an utter lack of centralized fiscal revenue, which neutered central governance capacity. Within eight years a constitutional convention was called to forge a new system, which I would argue turned out better.
When China began allowing markets after Mao, the economic structure shifted beneath Beijing’s feet. SOEs crumbled, the tax base eroded, and tax evasion increased. Total government revenue and the center’s revenue share massively declined.13 Though one imagines the Soviet Union’s recent dissolution was top of mind, one could have also seen parallels between Beijing’s situation in the early 1990s and America under the Articles of Confederation.
The Partial Red-Herring of Beijing’s Fiscal Re-Centralization
In the early 1990s Beijing sought to re-affirm central authority. One of the more reliable ways to do so is to take control of the purse. Of this view were General Secretary Jiang Zemin and his Premier Zhu Rongji, who pushed forward a major fiscal and tax restructuring in 1994. Two big changes were wrought. First, Beijing decisively centralized revenues while keeping expenditures decentralized. Second, Beijing barred local governments from running deficits, and from borrowing.14 China, the most fiscally decentralized country in the world in terms of local expenditure share, overnight developed a stark vertical fiscal imbalance, with unfunded mandates at the local level.15 Nearly an inversion of the original problem. But Beijing was happy and that’s what mattered (to Beijing).
Andrew Batson, “The real source of China’s local government money problems,” Personal Blog, July 2015, https://andrewbatson.com/2015/07/21/the-real-source-of-chinas-local-government-money-problems/
Ultimately, Beijing wanted revenues routed through its own coffers for purposes of control, most importantly over subordinate levels of government, and redistribution. That becomes evident once you realize the central government simply transfers back just about all the money. Indeed, once transfers are accounted for the oft-cited central-local fiscal gap disappears.16Unfunded mandates did materialize following the 1994 budget reform, but in a more nuanced way.
Andrew Batson, “The real source of China’s local government money problem,” Personal Blog, July 2015, https://andrewbatson.com/2015/07/21/the-real-source-of-chinas-local-government-money-problems/
The problem was (and still is) in the nature of the intergovernmental transfer system. Beijing bureaucrats apportion funds to the provinces, who are in charge of apportioning funds to prefectural cities, who are in charge of apportioning among county level units, who are in charge of apportioning among townships.17 But sometimes the provinces send funds directly to the counties, by passing the cities. Each level also needs their own funds. And each level can take months before passing on the funds it received. By the time funds get from top to bottom, a year or more can pass.18
An Example of Intergovernmental Transfers
Christine Wong and Xiao Tan, “Anatomy of intergovernmental finance for essential public health services in China,” BMC Public Health, 2022, https://bmcpublichealth.biomedcentral.com/articles/10.1186/s12889-022-13300-y/figures/1
It’s fairly easy to understand how funds may also leak, be misappropriated, or miss their mark. The severity of funding problems ends up being highly geographically variable. The resource gap for many local governments is real, if overstated at the macro-level if transfers are not considered.
When we talk about China’s fiscal revenue, what is likely more important to understanding the evolution of China’s political and economic trajectory is the production bias baked into the system: 60% of taxes are on goods and services, just 6.5% on income.19
Share of major taxes in total tax revenue
Source: NBS; Shuanglin Lin, The Fall and Rise of Government Revenue. In: China’s Public Finance: Reforms, Challenges, and Options. Cambridge University Press; 2022. https://www.cambridge.org/core/books/chinas-public-finance/fall-and-rise-of-government-revenue/D88A32934133E189C34D89C3071FCCF0
Local government’s fiscal structure highly incentivized them to figure out how to boost production. So they turned to the task with their highly decentralized expenditure capacity and, more importantly, decentralized powers of off-budget liability-creation.20
The Decentralized Developmental State Gets a New Tool
The expenditure portion of the government income statement is more important than revenue in understanding the rise of LGFVs.21 This owes to the pervasive and multi-faceted role of local governments in economic activity, mobilizing for development.22
Local officials in China have long competed intensively within a hierarchical system that is both decentralized and weakly institutionalized. A combination of top-down and bottom-up incentives attracting them to or diverting them from enacting the center’s ambitions. This developmental modality post-Mao evolved into what is colloquially called the “mayor economy.”23 It is in large part an institutional outgrowth from China’s decentralized Maoist-Leninist system, and arguably traces even further back to imperial-institutional legacies.24 Along with previously noted fiscal incentives, a particularly noteworthy top-down feature has been the Party’s Organization Department and its control and influence over local cadres promotion.25 Mao’s death enabled the re-direction of the Party-state’s high-powered incentive systems to a new goal of production-oriented tournament style competition.26 Arriving at a period of burgeoning hyper-globalization, localities soon began to fight tooth and nail to lure FDI.27 A period of post-Mao capital deepening, urbanization, and transition beyond agriculture occurred at unprecedented speed and scale.
Local officials hunted within this incentive ecosystem for ways to increase their economic role.28 One seemingly win-win way for local governments to deepen their involvement was infrastructure—a role that arguably became their most important.29 To fully step into this infrastructure supplying role, however, local governments needed a new financing tool. Given the enormity of the task, no typical fiscal revenue stream would suffice. That’s where a little ingenuity from financial big whigs at China Development Bank (CBD) came in to play. Chen Yuan, son of famed Chen Yun and then head of CBD, helped local officials in Anhui and Wuhu concoct a simple but brilliant way to expand their economic footprint: create a corporate entity owned by, though at arms length from, local governments that could borrow unconstrained by the 1994 budget and regulatory regime. The entity’s raison d'etre was simple: (1) raise money and (2) build infrastructure. This was the “Wuhu model,” and in 1998 it birthed China’s very first LGFV: Wuhu Construction Investment Company.30
Over the next twenty years the Wuhu spark would engulf the country.
Land, Meet Finance
Leveraging land was Wuhu Constuction Investment Company’s distinguishing characteristic. Indeed, it was land finance that made the rise of LGFVs possible.
Local governments began selling land-usage rights in the 1990s, around the same time as Wuhu’s LGFV popped into existence. Once the land auction system was approved centrally and rolled out in the early 2000s, local government sales of land-use rights exploded.31
Proportion of state-owned land transfer revenue relative to local public budget revenue
China Land Resources Statistical Yearbook, graph via Lan Xiaohuan, Embedded Autonomy (置身事内), Chapter 2 Section 2.
For good and for ill local governments, uniquely empowered to requisition and sell land-use rights, used all means at their disposal to get land and ready it for sale, including mass evictions of those currently occupying it.32 The sordid history of local government land theft from China’s rural populace (re-zoning the land and selling it property developers) manifested in many cases as a Georgist’s worst nightmare, with self-immolation a not infrequent form of protest in the countryside.33 CPC led land reform in the early 1950s gave peasants their land, communization in the late 50s took it back. Marketization in the 1980s let the peasants work their land again, local governments began stealing much of it away in the 1990s. Such are the ebbs and flows in China’s countryside.34
By the late 90s, the local government revenue split began to take its contemporary shape, roughly equally divided into three parts: (1) central government transfers, (2) locally generated revenue, and (3) land sales.35
Source: Fitch Ratings, Ministry of Finance, Wind; via FitchRatings, “Central Transfers Ease China’s Local Government Fiscal Strains,” March 2022.
Land sales and locally generated revenue are highly intertwined. In the late 1990s, local governments began selling industrially-zoned land cheaply to attract companies that they could tax harvest. Selling land low seems counterintuitive until you realize locally derived revenue mostly comes from value-added and corporate income tax. Thus local governments became ever more perfect price discriminating monopolists: they sold residentially zoned land (居住用地) high to maximize one off revenue, but sold industrially zoned land (工业用地) low to maximize recurring revenue.
Average quarterly transaction price of land transfer in 100 key cities. Source: Wonder Database, graphic via Lan Xiaohuan, Embedded Autonomy, 置身事内, Chapter 2 Section 2.
LGFVs took on two very important roles in the burgeoning land finance system. First, the pivotal role of preparing land for sale. Such preparation not only involves evicting rural residents and coordinating with the local governments on compensation (or lack thereof), but also requires multiple land development steps, known as the “seven connections and one level” (七通一平). LGFVs were used to build (1) road connections, (2) water supply connections, (3) electricity connections, (4) drainage connections, (5) heating connections, (6) telecommunications connections , and (7) gas connections. They also leveled the land in preparation for development. The extensive requirements of land development meant that often local governments were losing money on the land use sale (a problem that has only gotten worse in more recent years).36
Second, to finance these land-prepping activities, LGFVs ramped up use of the Wuhu land finance model. Local governments transferred more and more land use rights to these proliferating vehicles, who would in turn go to banks for loans using the land rights as collateral. And thus the land-industrial-complex began. Re-zone land. Transfer land. Develop land. Sell land. Buy land. Collateralize land.37
One need not be a Georgist to understand the unmatched power of land. Tax revenue would never have been capable of resourcing LGFVs the way land-finance did.
LGFVs, Property Developers, and Beijing’s Banking Balloon
No story on the rise of LGFVs could be told without a discussion of property developers and China’s banking system. If LGFVs are Batman, then property developers are Robin, and the banks are Wayne Enterprises (mileage may vary with this analogy).
The Wuhu model LGFV could take land from local governments and develop it for sale, but who would buy? Well, property developers. By the early 2000s, thousands of property developers had been established, eager to participate in China’s epic-scale urbanization process. Both sides of the would-be transaction, however, needed credit—particularly prior to the heyday of pre-sales in the 2010s. The LGFV-Property Developer crime fighting duo therefore needed their Wayne Enterprises.
Banks are the cornerstone of China’s financial system and were really the only game in town capable of facilitating a burgeoning land finance system. Most know about the big banks, but the stories slightly more complicated. To get localities on board with the 1994 fiscal overhaul, Zhu Rongji made a “grand bargain” with localities: in exchange for acquiescing to fiscal centralization, localities would get the right to establish their own locally controlled banks.38 As Adam Liu, whose research focuses on this topic, puts it: what is “rarely discussed is the most vital component of Beijing’s compensation package [for the 1994 budget reform]: local governments were offered the privilege of entering the banking sector.”39 When Beijing closed the front door with its budgetary restriction on lending in 1994, it opened a window.40
One can see the results: city commercial banks and other types of locally controlled banks proliferated. In 1995 the first city commercial bank was set up, and two decades later over 130 were in existence. The loan share of the big six banks collapsed from nearly 100% in the 1980s to roughly 25% today as a result of rising competition from locally controlled banks, e.g., city commercial banks, for deposits (as well as the 12 joint stock banks). The proliferation of state-owned local banks is an overlooked but consequential feature of China’s political economy, and core to the LGFV story.41
In the early 2000s banks were the primary funding source for the land finance system, on both the supply and demand side. Approximately 50-60% of real estate developers’ funding, for example, come from bank loans by 2004.42 That money, in turn, paid for land-use rights that funded local governments and, in part, their LGFVs. Meanwhile, LGFVs used their land-use holdings as collateral for loans (though in China’s weakly institutionalized system sometimes banks would lend without collateral).43 Bank loans made up about 90% of LGFV liabilities in that same period.44 China Development Bank and the big six state owned banks were core lenders, but locally controlled banks were players too, and their role would only grow more prominent.
LGFVs, property developers, and the banking system co-evolved.
Pouring Gasoline On a Fire: The 2008-10 Stimulus
Structural fiscal and developmental incentives likely would have continued to steadily drive LGFVs into ever greater positions of prominence within China’s economy.45 But it was the Great Financial Crisis of 2008 and China’s response to it that sent LGFV growth into steroidal overdrive.
No longer simply acquiescing to their use, Beijing began directing local governments to deploy LGFVs for countercyclical infrastructural stimulus and tasked the banking system with funding them.46 The People’s Bank of China explicitly called on local governments to use LGFVs to borrow, the banking regulator (CBRC) explicitly encouraged LGFV use, as did the Ministry of Finance.47 Buyer of narratives therefore beware, lest you fall victim to the woe is me central government fairy tale. This puts a bit of a lie to the notion that Beijing is effectively a responsible parent figure always trying to constrain misbehaving local officials.
When the center boldly announced its RMB 4 trillion stimulus plan to ward off recession, it did not intend to fund the stimulus directly but instead turned to local governments, now replete with bank licenses and financing vehicle. To this day we don’t know exactly how much China actually spent on its stimulus, though its clear the vast majority came off-budget via LGFVs. Only a trillion yuan shows up on the central government balance sheet.48 Beijing’s calculation of “official debt” accrual to LGFVs in 2009 and 2010 was RMB 3.6 trillion. But this number leaves out a substantial and non-transparent amount of LGFV debt.49 Some estimate LGFVs took on roughly a third of all new bank loans issued in 2009, and continued on in 2010 to account for 40 percent.50
As one example of what funds were used for, LGFVs mobilized to double-down on the build out of industrial parks and development zones. Parks and zones are used by local governments to attract companies, and thus revenue, jobs, and pad cadre evaluation, while from the central government perspective they are suppose to catalyze industrial clustering effects. From 2006 to 2018, officially recognized national and provincial level zones alone increased by 1,180, from 1,363 to 2,543.51 These numbers, large though they are, understate the extent of the build-out: they only include officially recognized zones above the city-level and do not reflect ongoing consolidation, merging, and cancellation of zones.52
One analysis of the post-2008 expansion stated that it “demonstrates that local governments are responsive to central commands.”53 Such a statement must be severely qualified. Local governments could not have been more eager to oblige Beijing in this instance, they tend only to respond with alacrity when doing so also corresponds with their interests. The host of epithets—from foot dragging, to bureaucratism, to formalism—is testament to the legion examples of local government non-compliance. But unleashing the LGFVs was also a boon for local cadres, their promotion metrics, and arguably for local development overall. It also made a fair amount of sense: local governments are closest to the ground and best understand what projects might work. And so China’s counter cyclical stimulus ran through local governments and their LGFVs.
Predictably, Beijing quickly lost what little control it had of the LGFV expansion process. Not only was the credit expansion much greater than Beijing intended, but LGFVs institutional role expanded and embedded deeper into the sinews of China’s economy. The analogy of LGFVs as Sorcerer’s apprentice is imperfect but apt.
Bend it Like Huarong
As they expanded, LGFVs moved out well beyond their initial infrastructural and land development remit. One analogy is that LGFVs have become akin to twelve thousand little Huarongs. Huarong, the country’s largest asset management company, was established just a year after the first LGFV in 1999. Colloquially referred to as a “bad bank” because it was set up to take non-performing loans off the Industrial and Commercial Bank of China’s balance sheet.54 Initially an asset recovery firm, Huarong metastasized into a massive conglomerate with dozens of subsidiaries involved as many different industries. The crazed expansion got so crazed under former Chairman Lai Xiaomin that the CPC decided to execute him.55 LGFVs have similarly spread out well beyond their original mission under increasingly complex corporate structures.
One categorization schema for LGFVs, offered by analyst Glenn Luk, divides LGFVs as follows: (1) The Infrastructure LGFV, (2) The Real Estate Asset Management LGFV, (3) The Structured Asset-Backed Warehousing LGFV, (4) The Financial Intermediary and Investment Holding LGFV, and (5) The Conglomerate “All-of-the-Above” LGFV. If the names don't immediately make senes to you, I invite you to read his post (see footnote).56 Suffice to say that after the 2008 explosion, NAO’s first audit in 2011 discovered that only half of LGFVs were strictly focused on government infrastructure projects, with 18% partly focused on them and a full third entirely focused on market-oriented projects.57 The issue has only exacerbated since.58
A recent Bank for International Settlement paper on the post-stimulus transformation of LGFVs shows just how far they have meandered beyond infrastructure. Between 2004 and 2018, only 20% of the investments made by the 4,432 they analyzed were into public goods sectors. They focus on Guizhou’s Dushan County as a concrete example. This LGFV operates across as many public goods sectors (e.g., electricity, heat production & supply, health, social security) as it does market sectors (e.g., software and IT services, retail, wholesale and real estate). The authors argue there is a U-shaped return to this sort of diversification, with some amount potentially helpful to growth but too much being harmful.59 Others, however, are much less sanguine, and see this expansion as potentially at the core of capital misallocation in China and the country’s overall productivity slowdown.60
Jianchao Fan, Jing Liu and Yinggang Zhou, Investing like conglomerates: is diversification a blessing or curse for China’s local governments?” BIS Working Papers, January 2021, https://www.bis.org/publ/work920.pdf.
LGFVs fed off a mix of moral hazard, unfettered access to credit, and indeterminant state responsibility for liabilities—that is, substantial state involvement under conditions of weak institutionalization. On their expansionary march LGFVs, from a more macro perspective, have cultivated an impressive web of investment linkages:
IMF, China: Selected Issues: Local Government Financing Vehicles Revisited, 2021, page 41.
Red LGFV Over China: The Case of Zunyi Road and Bridge
Zunyi, a small city located in China’s southwestern province of Guizhou, does not often make the news. The city’s claim to fame dates back to the Long March in 1935, when it played host to a meeting that, according to Chinese Communist Party (CPC) lore, decisively established Mao Zedong’s leadership over the Party. Other than that, Zunyi is relatively unremarkable, a city of middling population and economic development, firmly in the 3rd tier of China’s unofficial city ranking system. One recent development, however, has once again brought attention to the city: the increasingly urgent Party-state effort to deal with LGFV debt. Zunyi Road and Bridge Construction (遵义道桥建设) is the star of the show. A snapshot of Zunyi’s corporate structure offers hints at some of the poblems.61
Zunyi Road and Bridge holding structure, 遵义道桥建设（集团）有限公司, 股权穿透图, via 爱企查 https://aiqicha.baidu.com/company_detail_69261055076241
Road and Bridge’s subsidiaries also have subsidiaries. Take for example, its largest subsidiary: Zunyi City Baozhou District Urban Construction and Investment (遵义市播州区城市建设投资经营), with registered capital of RMB 1.6 billion. Baozhou Urban Construction itself fully-owns a diverse array of 10+ businesses, ranging from a funeral service company, to a financial leasing company seemingly focused on industrial equipment, to a water services management company, as well as a 49% stake in a property management company and a 5% stake in another diversified LGFV holding company.
Holdings of Road and Bridge Largest Subdiariy, Baozhou Urban Construction
Zunyi City Baozhou District Urban Construction Investment Management (Group) Co. (遵义市播州区城市建设投资经营（集团）有限公司), 股权穿透图, via 爱企查 https://aiqicha.baidu.com/company_detail_62311309937102
The financing practices to go along with such a web of holdings have been similarly convoluted. One company may acquire loans or go to the bond market only to on-lend to its affiliated entities. There are hundreds, perhaps thousands, of LGFVs like Zunyi Road and Bridge and like the one in Dashan County mentioned earlier. These conglomerate LGFVs undertake what economist David Daokui Li describes as a “nested layering approach” to levergae. Companies at one level borrow funds, use that borrowed capital to secure additional loans at the next subsidiary level, and amplify debt layer by layer.63 All the while moving into more and more lines of business.
Evolving Liabilities: Bonds and Shadow Finance
Beginning almost immediately after their massive GFC-era expansion, Beijing turned from hot to cold on LGFVs. The center unleashed the National Audit Office (NAO) twice upon localities to try and scrape together the extent of the LGFV bonanza, first in 2011 and then again in 2013.64 The then bank regulator, CBRC (which then became the CBIRC and, as of 2023, the NFRA) also began keeping a comprehensive list of all LGFVs around that time, dividing them into “good” and “bad” and ostensibly demanding banks refrain from many types of lending to those on the naughty list.65
Constraints on bank lending, combined with the need to rollover short-term (~4 year) loans post-08, created huge incentives for LGFVs, local officials, and many other actors to devise alternative financing methods for these increasingly systemically important vehicles.66 A cat and mouse game had begun between Beijing regulators and localities.
LGFVs and China’s financial system proceeded to co-evolve.67 As LGFVs moved beyond plain vanilla bank loans, they played a huge part in spurring two other major financing channels: municipal corporate bonds and shadow banking.
Municipal Corporate Bonds
Municipal corporate bonds (MCBs) are bonds specifically issued by LGFVs (they are sometimes called chengtou). One informed analysis of them wryly describes them as the quintessence of socialism with Chinese characteristics: “MCBs are a perfect example of how planning and the market mix in the contemporary Chinese economy: They are implicitly backed by local government (hence “municipal”), but legally speaking they are issued by LGFV entities just like other regular corporations (hence “corporate”).”68
In 2006 only 17 LGFVs had issued a bond. By 2010 it was 1200. By 2013, 1700.69
The rapid growth owes to the central government’s stimulus-era decision to allow and encourage LGFVs to begin issuing bonds (enterprise bonds, a subset of corporate bonds, which at the time were regulated by NDRC, but as of 2023 are now regulated by CSRC).70 Data compiled by Fitch cuts a striking image:
Fitch, China Corporate Bond Market Blue Book, 2019, page 21, https://your.fitch.group/rs/732-CKH-767/images/china-corporate-bond-market-blue-book_fitch_10083315.pdf
The municipal corporate bond market really exploded in size, though, as 1-to-2 year short-term bank loans to LGFVs needed to be rolled over and re-financed.71 While the stimulus was the spark, the real acceleration came when it was time to rollover short-term debt.72 One study combed through bond prospectus of MCB issuers and disaggregated issuance according to stated purpose (prospectuses require issuers to state the purpose for the funds). The data clearly indicate a rapid rise beginning in 2013, precisely when the average short-term bank loan would have been coming due.73
Municipal corporate bond (MCB) Issuance by Purpose (2004-2016)
MCB_repay refers to repayment as purpose and MCB_inv to new investment. Chen, Z., He, Z., & Liu, C, “The financing of local government in China: Stimulus loan wanes and shadow banking waxes,” Journal of Financial Economics, 2020, page 52.
Rather than close, most “bond buyers preferred LGFV bonds because they offered higher yields than corporate debt, but were considered government guaranteed, even though they funded projects that typically had no capacity to repay the debt.”74
Most significant to the story, though, is who bought the bonds and how.
Shadow Finance: The Banks Shadow
Shadow banking is financial activity outside regulated channels. But in China, shadow banking takes on a double entendre of literally being the shadow side of banks. China’s shadow banking mirrors the rest of the financial system in being massively bank dominated. LGFV municipal corporate bond issuance was, as it turns out, mostly bought up by banks via shadow banking products.
Wealth Management Products (WMPs), or special funds set up to skirt regulations on deposit rates, are the most important shadow banking product. WMPs invested heavily in municipal corporate bonds. Chen and He et al calculate that 62% of all MCB proceeds came from WMPs. And it was banks that created those WMPs, of course with money that ultimately belongs to household depositors/lenders.75 Shadow banking and WMPs, much like MCBs, may not have precisely arose via LGFV stimulus, but massive pressure to rollover debt and refinance short-term stimulus loans created an extreme accelerant.76
Systemic incentives aligned to shift a substantial share of banking into the shade. Banks already engaged in regulatory arbitrage to get around the deposit rate ceiling and intermediate funds they otherwise wouldn’t have been able to.77 LGFVs got funds they wouldn’t have otherwise had access to. And households got higher returns via WMPs, which they viewed as fail-safe investments (owing to rampant moral hazard).78 Competition among banks to issue WMPs and compete for deposits, specifically between the big four and small and medium sized banks (SMBs), also intensified, going from luring deposits via offers of cooking oil, gold bars and cash to issuing higher-interest WMPs.79 As the figure below shows, WMP issuance as a share of all bank assets takes off in 2011, precisely when LGFV rollover needs are emerging.
Floating WMPs, where yield is not guaranteed, saw particular increase because they could be off-balance sheet and thus not trigger regulatory loan-deposit ratio. Viral V. Acharya Jun “QJ” Qian Zhishu Yang, “In the Shadow of Banks: Wealth Management Products and Issuing Banks’ Risk in China,” February 2017, page 45, https://jrc.princeton.edu/sites/g/files/toruqf2471/files/qian_jun-shadowbank-china-aqy-10feb17-all.pdf.
Entrusted loans, meanwhile, were the second most prominent category of shadow banking, making up roughly 20 percent of shadow lending relative to WMPs 52 percent.80 Entrusted loans are loans between two non-bank entities that are facilitated by banks, a service for which the bank takes a fee but no exposure. The main characterstics of entrusted loans is firms with privileged access to China’s banking system, such as SOEs and larger LGFVs, channeling capital to those with less access.81 Most such transactions occur between affiliated parties, as between parent and subsidiary, such as in the case of Zunyi. This type of lending increases when credit conditions tighten, precisely as happened when Beijing tried to limit the credit deluge following its GFC-stimulus.82 Entrusted loans, a marginal part of total social financing relative to bank loans, clearly take off with LGFV refinancing.
Post-Stimulus Rise of Entrusted Loans (2002-2020)
Franklin Allen, Xian Gu, C. Wei Li, Jun “QJ” Qian, Yiming Qian, “Implicit Guarantees and the Rise of Shadow Banking: the Case of Trust Products,” Forthcoming Journal of Financial Economics, April 9, 2023, page 69, https://papers.ssrn.com/sol3/papers.cfm?abstract_id=3924888
Beyond WMPs and entrusted loans, trust companies also came onto the scene. Trust companies are “conduits that connect financial and non-financial institutions with asset classes that their licenses would otherwise prohibit.”83 Many trusts served as yet another mechanism to intermediate bank financing, in large part to LGFVs. The quintessentially convoluted process goes something like this: a bank makes a loan then sells it to a trust, the trust packages multiple such loans into a “trust plan,” then another bank’s WMP invests in that trust plan—investment banks would later serve as an additional layer, intermediating the second bank’s WMP investment into the trust plan.84 Such is the financial cat and mouse game.
Thus, all told, MCBs and shadow banking explode together around 2012 when LGFVs needed post-stimulus refinancing. The stimulus-through-LGFVs and follow-on regulatory tightening had the incidental effect of hastening development of China’s bond and shadow banking markets. LGFVs and the financial system co-evolved.85
Growth of shadow banking in China (2010-2020)
Franklin Allen, Xian Gu, C. Wei Li, Jun “QJ” Qian, Yiming Qian, “Implicit Guarantees and the Rise of Shadow Banking: the Case of Trust Products,” Forthcoming Journal of Financial Economics, April 9, 2023, page 69, https://papers.ssrn.com/sol3/papers.cfm?abstract_id=3924888
The evolution of LGFV liabilities reflects the tale. Based on a bottom-up estimates, Zhang and Xiong estimate loans fell from 79% of LGFV liabilities to 60% by 2015. That ratio has remained roughly stable up to today. In 2023, for example, global asset manager PIMCO estimated that the composition of LGFV debt was 60% bank loans, 30% bonds, and 10% other financing sources.86 Clearly bank loans still dominate, but given the extent of resources in play, the compositional shift was pivotal.
Evolving LGFVs Liability Composition
Zhiwei Zhang and Yi Xiong, “Infrastructure Financing” in Handbook of China’s Financial System, 2020, page 213.
Beijing’s effort to conjure then control LGFV borrowing contributed to a sort of diversification of the financial system, including the massive expansion of the bond market. Many called for just this sort of change to Beijing’s bank dominated financial system. It’s a bit like a simulacrum of diversification, though, as banks remained the ultimate buyers, just with additional steps. But, at least when it comes to bonds, much debt was refinanced in substantially more sustainable manner. What’s more most of the ultimate lenders, via WMPs, were rich coastal Chinese folks. A potential avenue for common prosperity, and partial resolution to LGFV debt, could be defaulting on those bonds. That, however, could also risk seismic upheaval in a system laden with moral hazard and wherein not a single LGFV bond has yet to default.
The question of how to resolve the seemingly unstoppable freight train of LGFV borrowing lingered on.
Fiscal Reform Round 2: “Solving” The LGFV Debt Problem
As with property developers, the central government has been aware of problems with LGFVs for a long time and has been moving to resolve what they call LGFV’s “hidden debt.” As mentioned, regulations were first put in place in 2010 to impede bank lending to LGFVs. But the most important effort to try and resolve LGFVs debt and financing problems came with the country’s second big change to its budget framework in 2014, precisely two decades on from the last one. This time, a core purpose of the change was to deal with off balance sheet LGFV debt.
Most pertinent to resolving the LGFV situation, local governments were given a mandate and resources to swap off-balance sheet “hidden” LGFV debt with on-balance sheet bonds. Local governments were now given a clearer path to issuing debt themselves via bonds, and being actively encouraged to do so. The goal was to recognize contingent debt, refinance it with longer duration, lower interest rate bonds, and eliminate LGFVs within three years. Beijing’s “new budget law prohibits local government and their branches from borrowing in any other form, and unless otherwise specified by law, from offering any credit guarantee to any organization or individual.” Then the State Council issued Document Number 43 in September 2014 which aimed “to make these rules explicit” by stating LGFVs did not have “the authorization to borrow on behalf of the local government. If the only business of an [LGFV] is to borrow on behalf of the government, it should be shut down.” The goal, according to economists Chong-En Bai, Chang-Tai Hsieh, and Zheng Michael Song, “was to entirely eliminate [LGFVs] by replacing the debt of the [LGFVs] with local government bonds within three years.”87 If this was indeed the goal, then the reforms clearly failed.
China’s Interest-Bearing LGFV Debt
Source: IMF Article IV Consultations for China, years 2018. 2019, 2021, and 2023.
The failure has several causes. From the start, there remained debt classification issues with regard to what Beijing and local governments considered contingent liabilities. As a result, and the first big cause of the failure, is that a sizable amount of LGFV debt—equivalent to at least 13.4% of GDP, which the IMF’s 2018 Article IV derived from NAO estimates—was likely never transferred.88 Officially recognized local debt did shoot up as part of major bond swap program, and LGFV debt in turn fell from an estimated 32.1% of GDP in 2013 to 13.4% in 2014. But then LGFV debt as percent of GDP shot back to its original levels within three years.
Second, the new bond financing mechanism remained basically the same as the old one. The only difference was that Beijing actively used it. As before, all bond issuance has to be approved by the central government, via the State Council (and approved by National People’s Congress). In turn, provinces are in charge of issuing the debt and distributing the proceeds to all lower level governments. While controlling potential wanton lower-level behavior, it also creates an immense coordination problem lower levels and upper levels. Not only might upper levels not know what lower levels really need, lower levels may have incentive not to fully disclose how much they need. Low-level local governments would thus clearly find it easier to continue relying on channels they control (their banks) rather than official channels like special purpose bonds that are highly onerous, take a long time to get funds, and subject local officials to greater scrutiny.
Third and most important: nothing was changed with regard to local developmental incentives. Local officials were still highly involved in the economy, still controlled local banks and shadow institutions, and still stood to benefit in a variety of ways from credit finding its way to LGFVs. Predictably, new lending did not stop and LGFV debt continued to pile up.
Policies and Counter Policies (上有政策，下有对策）
The center’s hands have been far from clean in the LGFV clean up process. Throughout the winding road of LGFV growth the center has repeatedly stoked the LGFV fire only to then play fireman. The 2009 LGFV stimulus followed by crackdown is one example, but the same dynamic happened just prior to 2014 and again just after. In 2013 the State Council called upon local governments to increase infrastructural investment once more.89 Then, just after the big 2014 Budget Law change, “the State Council issued a new decree in May 2015 (document 40) that reversed its attempts to crack down on [LGFV] borrowing,” urging “financial institutions to continue to lend to [LGFVs].”90 Not only do local governments counter Beijing’s policies, Beijing often counters Beijing’s policies.
In more recent years, however, Beijing has gotten more focused and the drum beat of regulatory restrictions has intensified. Interestingly, though, they have also become more secretive.91 In 2018 the State Council made explicit the PRC government’s ambition to wipe out all local government hidden debt, calculated per 2017 numbers, within 5 to 10 years.92
In the ten years from 2014 to 2023, amid the LGFV diversification process we discussed earlier, local SOEs began to be classified and regulated according to three categories: competitive, functional and public welfare.93 The center is trying to rationalize the operations of LGFVs, limiting state subsidization and backstopping of those oriented toward the market while phasing out the weakest performers. In practice, the policy efficacy is dubious at best. LGFVs have mixed and matched assets not only to create conglomerates that might stand on their own feet in the market, but also as part of a cat and mouse game with regulators with the ultimate goal of keeping credit flowing and their operating capacity in tact. Conglomerates with diversified holdings (from pure public-welfare to pure market) become too big to fail for local governments. LGFV complexity and debt burdens have only grown.
Since 2018 an increasing number of regulatory documents have ceased to be made public. Some speculate there is fear and hesitation within China’s bureaucracy of taking ownership over this massive problem. Amid increasing clamor to better address the LGFV issue, the PRC government issued and circulated two guiding documents within its bureaucracy in 2023. Neither Document 37 or 47, as they are shorthanded, have been made public. But substantial information has leaked within China, and discussion on WeChat is ample.
One organization, an asset management and consulting company that focuses on local governments and their SOEs, has put together a helpful guide on how struggling LGFVs might “transform and develop in the context of today’s debt package.” The firm, Nanjing based Zhuoyuan, suggests:“If an urban investment platform in a region has weak qualifications and declining financing capabilities” it might want to consider “injecting more high-quality assets and integrating weak platform assets.” Alternatively, it might try acquiring high-quality listed companies so as to “broaden investment and financing channels for urban investment companies.”94
A new round of audits is apparently under way to re-assess the scope of the LGFV debt situation. A number of other steps have also been announced, including $1 trillion worth of bonds to bring LGFV debt on-budget and refinance at lower rates, to be handled by provincial governments in the 12 most struggling provinces. More policies and announcements are expected. But what will their efficacy be?
Conclusion: The Most Complex Economic Problem?
Having arrived at the end of our story on the rise of LGFVs, the fall remains untold. The winds of change have been blowing. But a precipice remains elusive. Metaphorical cliffs do exist in China’s weakly institutionalized, moral hazard laden system. And Beijing has shown a willingness to push things off. Need we discuss the rapid fall of China’s property developers?95 With the land market contracting, local officials and LGFVs will be hard-pressed to keep a land-financed based LGFV system intact.
Why though, one wonders, has there been no LGFV policy equivalent of the Three Red Lines? Is it because they are more embedded in “the system”? Consider how much greater the on-balance sheet exposure to LGFVs is than to property developers.
IMF Article IV China 2023, page 26
Or maybe it’s because, despite escalating central regulatory action against them, LGFVs remain useful to Beijing in a variety of ways. Beijing, as we have seen, often counters Beijing (上有政策，上也有对策). LGFVs are handy not just as shock absorbers against economic recession, nor just as levers to buttress growth and employment, but also as vehicles for carrying out country-wide goals such as extreme poverty elimination and shanty-town revitalization. But the costs of using LGFVs mounts.96
Whether or not shock therapy comes for LGFVs, the incentive ecosystem is shifting. Beijing has decided its bureaucracy must mobilize not principally to rectify backwards productive forces, but rather to remedy the “contradiction between people’s ever-growing need for a better life and China’s unbalanced and inadequate development.”97 At least one implication is clear: the land-finance system that for the last two decades has been the lynchpin for much of China’s development is being directed toward the dustbin of history. High-powered top-down incentives to boost growth are no longer so high-powered. Property developers have taken their hit, but LGFVs remain at large.
Ambiguous mandates render the future of LGFVs difficult to divine. What constitutes a better life, precisely? Growth is still important, but must be balanced with security and needs to be “high-quality.”98 Local experimentation is still encouraged, even demanded, but must comport with stricter top-down preferences.
Despite the diversification we have seen in the holding schemes, asset structures, and liability composition of LGFVs, there remains a lack of diversification where it matters. Local Party-state officials continue to control LGFVs and the local banks, and possess immense influence over many other facets of their local economy. In a remarkably concise passage from his book 置身事内, Lan Xiaohuan argues that “the root cause of the government’s debt problem,” and by extension of low-quality growth, “is not insufficient revenue, but rather excessive spending, as the government has taken on too many roles in developing the economy. Therefore, the debt problem is not simply a ‘soft budget constraint’ issue or a problem solved by modifying the government’s budget framework. Instead, it is fundamentally a problem of the government’s role.”99 LGFVs are the crux of this problem. And it is why property developers, most of which were private companies, have not posed nearly as vexatious a problem for Beijing.
A decade ago Beijing not only set out to constrain LGFVs, but to eliminate them. Fiscal restructuring proved insufficient. Today, localities still have dauntingly expansive roles and mandates, will new sources of financing materialize or will they be forced to abdicate? In this evolving context, will local officials face new incentives to keep their all-purpose handy man, the LGFV, alive and kicking? Will LGFVs whither away, as Lenin once promised the Soviet state would? Who will make them? With a new round of audits sweeping the nation alongside top-down inspection tours and the ongoing anti-corruption campaign, what might become of China’s 12,000 LGFVs?
A common saying in the Party these days holds that it is sometimes necessary to “turn the knife inward and scrape poison off the bone.” A fundamental solution to the LGFV problem, it seems, requires a deeper cut into the system.
The 12,000 number is derived from China’s own statistics. China’s banking regulator has, since 2010, attempted to compile a comprehensive list of China LGFVs. In 2021, the CBIRC (now the NFRA) counted 11,736 LGFVs, seemingly unchanged from 2018 total of 11,734. But LGFVs can be very mysterious entities. They do not follow a specific naming system, though they normally include something about urban construction (ergo their typical Chinese acronym, urban investment company, or chengtou [城投]). They are all local, state-owned enterprises. With the center using this list to, in large part, try to constrain lending to many of these entities, LGFVs and their local governments would clearly have some incentive to limit transparency, likely contributing to undercounting.
Multiple organizations exploit the spec of transparency offered by bond-issuances to estimate the LGFV debt problem. Estimates differ seemingly based on how many bond-issuing LGFVs get included in their bottom-up estimates.
At the upper bound of comparables, Rhodium Group, collated and analyzed annual reports from 2,892 LGFVs that have issued bonds.
A 2022 study by the IMF, presumably using very similar methodology as the Article IV consultations, used CapitalIQ’s more limited database to analyze 2200 such LGFVs. PIMCO used Wind Financial and perhaps some proprietary information to come up with its own bottom-up estimate of LGFV and other government debt.
Nearly every year IMF Article IV report on China changes its estimate of LGFV debt levels. For instance, these are the estimates of China’s 2017 LGFV debt:
2018 Article IV: 24.1%
2019 Article IV: 24.1%
2020 Article IV: 32.0%
2021 Article IV: 32.2%
2022 Article IV: 37%
This is annoying though understandable given limited transparency and a changing pool of LGFVs with financial disclosures. Part of the explanation, at least for the big change in 2020, is a methodology in that yea: “IMF has historically used a top-down approach to estimate China’s LGFV debt based on the results of the National Audit Office (NAO)’s 2013 audit of LGFV debt. Beginning in 2020, the IMF switched to a bottom-up approach based on the firm-level financial statements of bond issuers classified as LGFVs by the bond market regulatory agency NAFMII, in line with observed market practice.” https://www.elibrary.imf.org/view/journals/002/2022/022/article-A003-en.xml
Another possible explanation is offered in Chen (2020): “WIND classifies MCBs following the ChinaBond Pricing Center. As a subsidiary wholly owned by China Central Depository & Clearing Co., Ltd., ChinaBond provides authoritative pricing benchmarks of Chinese bond markets. Whenever ChinaBond changes its MCB component list, WIND adjusts its classification retroactively, causing the number of MCBs in our study to potentially differ from other studies on MCBs.” See: Chen, Z., He, Z., & Liu, C., “The financing of local government in China: Stimulus loan wanes and shadow banking waxes,” Journal of Financial Economics, 2020, page 48.
The only data sources that include LGFVs beyond those issuing bonds are China’s official audits, conduced in 2011 and 2013, by the National Audit Office (NAO). See: National Audit Office, “2013 Announcement 32: Nation-wide governmental debt audit results,” (2013年第32号公告：全国政府性债务审计结果), 2013. https://web.archive.org/web/20140104011301/http://www.audit.gov.cn/n1992130/n1992150/n1992500/n3432077.files/n3432112.pdf
We have no more recent bottom-up estimates of the entire galaxy of LGFVs. Prior to 2020, the IMF appears to have simply projected forward those estimates.
Those data, however, are limited. As Bai et al (2016) describe: “the data on the Audit Office only covers "official" debt of the LGVs, which the Audit Office defines as "the debt that government has responsibility to repay or the debt to which the government would fulfill the responsibility of guarantee or for bailout when the debtor encounters difficulty in repayment.”” Thus Party-state data may tell us how much LGFV debt was used on infrastructure, but it does not give us the whole picture. As Bai et el (2016) note, the total debt in their smaller sample of bond issuing LGFVs is larger than the total given by the NAO See: Chong-En Bai, Chang-Tai Hsieh, and Zheng Michael Song, “The Long Shadow of a Fiscal Expansion,” Becker Friedman Institute, November 2016, page 11.
Here is an overview of the data accumulated in the NAO audits:
Chen, Z., He, Z., & Liu, C., “The financing of local government in China: Stimulus loan wanes and shadow banking waxes,” Journal of Financial Economics, 2020, page 47.
LGFVs accounted for in the data, 2-3000, are equivalent to roughly 20% of all 12,000 LGFVs. They are highly likely to be the largest. If we assume these 20% account for 80% of interest-bearing debt, then most bottom-up debt estimates leave out roughly 20%. Thus to get a total estimate we simply multiply the original estimate by 1.25. I.e. IMF-Estimate x (1/0.8).
This is still probably an underestimate. Relative to Rhodium’s survey of 2,982, the IMF appears to include only 2,200. If we apply the same pareto assumption to Rhodium’s 2022 estimate of LGFV debt of 59% (which also includes accounts payable), we would expect LGFV debt to be approaching 75% of GDP, or RMB 90 trillion, as of 2022.
David Daokui Li’s methodology arrives at the conclusion that local debt is likely 50% higher than IMF estimates, which would be roughly equivalent to the augmented Rhodium estimate above of 75% of GDP.
If we analyze LGFVs strictly in a financial sense, we miss a major part of the picture. Namely, the divergence between economic and financial returns. Many LGFVs invest in infrastructure projects with positive economic externalities but low financial returns, precisely the kind of investments private investors would not be willing to make and wherein market failures, in the neo-classical bent, are admitted to exist. A narrow financial assessment of LGFVs would therefore fail to capture potential positive social and economic externalities.
In addition, the above data should not necessarily give cause for concern over an acute crisis. LGFVs have an abundance of real assets that could provide some amount of income. In addition, some of those assets could be liquidated to pay down some of the debt. With an asset to liability ratio of nearly two-to-one (i.e., 125% to 70%), there’s substantial buffer. And most important, not only is nearly all LGFV debt held internally but most of the lenders are also state-owned. Counter-party risk is minimal.
Intellectual Yet Idiot, coined by Nassim Taleb. “Typically, the IYI get the first order logic right, but not second-order (or higher) effects making him totally incompetent in complex domains.” Some of his other examples of IYIs in his chapter from Skin In The Game are themselves quite dumb, but the acronym is still great. Nassim Nicholas Taleb, “The Intellectual Yet Idiot,” Medium, 2016, https://medium.com/incerto/the-intellectual-yet-idiot-13211e2d0577.
Others argue that the first LGFV was developed in Shanghai six years prior. For example, Andrew Collier in his book Shadow Banking in China, states: “The first of these companies was established in Shanghai in 1992. Called the General Corporation of Shanghai Municipal General Corporation, it was set up to coordinate construction of municipal infrastructure projects, including water, sewage, roads, and other utilities. It received both municipal funds and the authority to borrow from banks…” Andrew Collier, ShadowBanking in China, page 54.
However, as Sanderson and Forsythe note, and as discussed later in this essay, the Shanghai financing vehicle did not exploit the core characteristic that distinguished the Wuhu model LGFV: land-finance. Therefore I go with Wuhu as the first LGFV. But one could reasonably disagree. Lan Xiaohuan also refers to the Wuhu LGFV as the first in his book 置身事内.
Inflation was 18% in 1988 and 1989, and after calming spiked back to 24.1% in 1994:
Mundell, R. A., "Introduction" in “In Inflation and Growth in China,” IMF, 1996, https://www.elibrary.imf.org/display/book/9781557755421/ch001.xml
The high inflation of this period was perceived as a major problem among China’s leadership, considered one of Zhao Ziyang’s failings, and even a root cause of Tiananmen protests. See Julian Gerwirtz, “Never Turn Back,” Harvard University Press, 2022, pages 212, 214, 281.
In 1978, general revenue was 30.8% of GDP (with extra-budgetary contributing an additional 8%). By 1993, general revenue had fallen to just 13% of GDP. Of that, the central governments’ share had fallen to just over 20%. Shuanglin Lin, “The Fall and Rise of Government Revenue,” In: China’s Public Finance: Reforms, Challenges, and Options. Cambridge University Press; 2022, https://www.cambridge.org/core/books/chinas-public-finance/fall-and-rise-of-government-revenue/D88A32934133E189C34D89C3071FCCF0.
Technically, local governments could still borrow but only if the central government expressly allowed it. In practice, they were almost never allowed to borrow. That would have change during the 2014/15 fiscal reform, which kept the same system but better defined the pathway for local government on-book borrowing and began approving more of it. For clarity on this point see: Donald C. Clarke, “The Law of China's Local Government Debt Crisis: Local Government Financing Vehicles and Their Bonds,” George Washington University Law School, https://scholarship.law.gwu.edu/cgi/viewcontent.cgi?article=2472&context=faculty_publications.
For more depth on the 1994 fiscal reform, see: Philippe Wingender, “Intergovernmental Fiscal Reform in China,” IMF Working Paper, 2018; Wang, Shaoguang. “China’s 1994 Fiscal Reform: An Initial Assessment.” Asian Survey, 1997, https://doi.org/10.2307/2645698
On China’s broader fiscal system, this is one of the best overviews I’ve read: Baoyun Qiao, Xiaoqin Fan, Hanif Rahemtulla, Hans van Rijn, and Lina Li, “Critical Issues for Fiscal Reform in the People’s Republilc of China,” ADB, June 2023, https://www.adb.org/publications/fiscal-reform-prc-fiscal-relations-debt-management.
Most analysts, however, simply note the overall fiscal imbalance. For example: “Inter-governmental fiscal imbalances are at the root of this increase as local governments faced persistent revenue shortfalls relative to their spending obligations.” In otherwise excellent piece: Waikei R Lam and Marialuz Moreno Badia, “Fiscal Policy and the Government Balance Sheet in China,”August 4, 2023;
See also: “CDB’s lending to local governments stems from the failure of Zhu Rongji’s 1994 reforms, which left local governments with huge spending burdens—everything from providing water to roads—but no way to raise funds apart from leasing out state land. The prohibition set on borrowing by local governments was a rule observed only in the breach, just pushing the borrowing off the budget and into the arms of the state banks.” Henry Sanderson and Michael Forsythe, “China's Superbank: Debt, Oil and Influence—How China Development Bank Is Rewriting the Rules of Finance,” Wiley, 2012, page 30.
And: Nicholas Borst, “China’s Balance Sheet Challenge,” China Leadership Monitor, 2023, https://www.prcleader.org/post/china-s-balance-sheet-challenge
On the issue of intergovernmental transfers, see: Linda Chelan Li and Zhenjie Yang, “What Causes the Local Fiscal Crisis in China: the role of intermediaries,” Journal of Contemporary China, 2014, https://www.tandfonline.com/doi/full/10.1080/10670564.2014.975947; OECD, Urban Policy Reviews: China 2015, https://read.oecd-ilibrary.org/urban-rural-and-regional-development/oecd-urban-policy-reviews-china-2015_9789264230040-en#page191
For a case study of how intergovernmental transfers work in practice see: Christine Wong and Xiao Tan, “Anatomy of intergovernmental finance for essential public health services in China,” BMC Public Health, 2022, https://bmcpublichealth.biomedcentral.com/articles/10.1186/s12889-022-13300-y/figures/1
Personal income tax, by comparison, was 54.39% in the United States, 40.41% in Germany, and 17.25% in Russia. Also worth noting, China has no capital gains tax, estate tax, or gift tax. And most importantly: no personal property tax. For comparison, property tax was 14.78% of tax revenue in the US and 12.4% in the United Kingdom in 2019, and 11.95% in Japan in 2018. For much more see: Shuanglin Lin, “The Fall and Rise of Government Revenue,” In: China’s Public Finance: Reforms, Challenges, and Options. Cambridge University Press; 2022. https://www.cambridge.org/core/books/chinas-public-finance/fall-and-rise-of-government-revenue/D88A32934133E189C34D89C3071FCCF0
Andrew Batson also sees decentralized liability creation, coupled centralized revenue, as quintessential to the system. He makes his opinion on the system rather straightforward: “A combination of centralized revenue-raising authority and decentralized liability-creating authority is the worst of both worlds, and the sooner China gets away from it the better.” See Andrew Batson, “The fiscal consequences of a unitary state,” Personal Blog, October 2023, https://andrewbatson.com/2023/10/18/the-fiscal-consequences-of-a-unitary-state/.
This is also the argument Lan Xiaohuan makes in 置身事内
Also see Tsinghua ACCEPT’s very good paper on China’s post reform and opening development, which has a good explanation of how over-eager officials getting involved in the economy and stimulating excess production, among other maladies, is at the root of the economic problem as much, if not more, than fiscal imbalances. See “Summing Up 40 Years of Economic Reform and Opening,” 改革开放四十年经济总结, 清华大学中国经济思想与实践研究院 Academic Center for Chinese Economic Practice and Thinking, December 2018, page 17, http://www.accept.tsinghua.edu.cn/_upload/article/files/7a/3c/976c22ff48cfb1860f7288f6bd05/612c2ab9-2527-4237-b28e-7ce36ee2b379.pdf;
Local government spending favors production and growth oriented infrastructure rather than service oriented public goods, clearly shown via date from 1999 to 2006: Chengri Ding, Yi Niu, Erik Lichtenberg, “Spending preferences of local officials with off-budget land revenues of Chinese cities,” China Economic Review, 2014, https://www.sciencedirect.com/science/article/abs/pii/S1043951X1400131X
See Kristen E. Looney, “Mobilizing for Development: The Modernization of Rural East Asia,” Cornell University Press, 2020.
This is campaign style governance, or goal pursuit via systems of bureaucratic and popular mobilization. For example, in the context of rural development, “governments may employ such tactics as sending official work teams to the villages, ratcheting up propaganda, setting core tasks, designating model sites, training local activists, and rewarding the most fervent participants with prizes, media attention, and other benefits designed to foster competitive emulation.” See Looney, page 6.
China’s intensely decentralized implementation system, its Leninist institutional heritage, and its extensive experience with campaigns made mobilizing for development a natural fit. The expectation from Beijing, once it changed the principal contradiction facing Chinese society in 1978 away from class struggle and toward remedying the backwardness of productive forces, was that cadres at all levels would mobilize for this most fundamental, top-down mission.
The results of China’s campaign mobilization were mixed for the countryside in particular, where extensive appropriation from rural residents was common. As Looney notes: “China’s rural modernization is a story of huge successes and huge failures. During the 1980s, marketization, along with decentralization and decollectivization, resulted in unprecedented growth and poverty reduction…Over time, however, housing became the primary target of local efforts, and despite an initial emphasis on moderate change in this area, the policy evolved into a top-down campaign to demolish and reconstruct villages. Rural resource extraction continued in the form of land grabs, the rural-urban gap grew wider, and problems with the quality of goods and services surfaced. The Chinese case underscores how easily campaigns can spiral of control.” Looney, page 8.
The notion of regionally decentralized local governments and local officials competing within a context of developmental incentives is central to many analyses of what has distinguished China’s economic and political systems. See:
Chenggang Xu, “The fundamental institutions of China’s reforms and development”, Journal of Economic Literature, December 2011;
Pierre Landry, “Decentralized Authoritarianism in China,” Cambridge University Press, 2008.
For more on the mayor economy, or the constructive role of competitive local officials in economic development, see also Keyu Jin, “The New China Playbook”
China’s model operates via ‘good-enough institutions. See: Yuen Yuen Ang, “How China Escaped the Poverty Trap” and Yuen Yuen Ang, “Beyond Weber: Conceptualizing an alternative ideal type of bureaucracy in developing contexts,” Regulation & Governance, 2017.
Earlier analysis saw a federalist bent to China’s policies, though most are less sanguine this is the most useful frame for China’s uniaary system: Barry Weingast, Gabriella Montinola, and Yingyi Qian, “Federalism, Chinese Style: The Political Basis for Economic Success in China,” World Politics, 1995.
China’s model is perhaps sui generis. As Kellee Tsai writes: “Rather than promoting market-preserving federalism or a coherent developmental state, China’s fiscal reforms unleashed a remarkable diversity of informal adaptive practices and developmental strategies among local governments.” Kellee S. Tsai, “Off balance: The unintended consequences of fiscal federalism in China,” Journal of Chinese Political Science, 2004.
China’s Maoist-Leninist institutions touched on here were neo-traditional, in Ken Jowitt’s framing. These were harnessed or allowed to transmutate into a form of embedded and decentralized state capitalism, with neoliberal elements.”
For a deeper understanding of the institutional nature of China’s Leninist system, Ken Jowitt’s “Leninist Extinction” is a valuable resource.
For the argument applied to post-Mao China, see Jean C. Oi, “Rural China Takes Off: Institutional Foundations of Economic Reform,” May 1999.
Pierre Landry, “Decentralized Authoritarianism in China,” Cambridge University Press, 2008; Hongbin Li and Li-An Zhou, “Political turnover and economic performance: the incentive role of personnel control in China,” Journal of Public Economics, 2005.
Individual psychosocial benefits, e.g., status and distinction, and the ample potential personal gain also matter greatly.
Or, as the newly divined principal contradiction at the 1978 3rd Plenary of the 11th Centtral Committee put it, to remedy: backward productive forces” and return to taking economic development as the central task. Shi Dan (史丹), “Evolution of Principal Contradiction Facing Chinese Society and the CPC Leadership over Economic Work,” Institute of Industrial Economics (IIE), Chinese Academy of Social Sciences, May 2022, http://www.chinaeconomist.com/pdf/2022/2022-5/Shi%20Dan.pdf.
For a corrective against the excessive lionization of Deng’s role in bringing about this shift, also see: Julian Gewirtz, Never Turn Back, 2022.
The crazed race to lure production resulted in intensive expenditure. Keyu Jin states that between 1998-2007 foreign manufacturing companies were the most subsidized type of firm in China, on average getting multiples more than SOEs. See Keyu Jin, “The New China Playbook,” 2023, page 101.
The local role in economic development is both the curse and blessing of China’s development. For a time in the 90s and aughts, after the Party changed its principal contradiction to focus cadre competition on economic development, the decentralized authoritarian economic model conjoined the interests of the central government, localities, local officials, corporations, and most of the populous eager to benefit from rapid growth.
Whether local governments offered more of a helping hand or grabbing hand is still debated. But the results speak for themselves. Clearly an immense amount of development happened even in the presence of corruption and a certain amount of grabbing. Following the 1994 reforms, officials still had ample incentive to work with enterprises and, of course, to get involved themselves via LGFVs and other activities.
For a negative view of China’s fiscal centralization on switching local officials from helping to gabbing see: Chen, Kang & Hillman, Arye & gu, Qingyang, “From the Helping Hand to the Grabbing Hand: Fiscal Federalism and Corruption in China,” 2002, 10.1142/9789812778277_0008.
For a middle of the road assessment and the potential diversion of productive entrepreneurial and business activity to rent-seeking, see: Zhiqiang Dong, Xiahai Wei, Yongjing Zhang, “The allocation of entrepreneurial efforts in a rent-seeking society: Evidence from China,” Journal of Comparative Economics, 2016, https://doi.org/10.1016/j.jce.2015.02.004.
The other contender for most significant local government role is their aggressive courting of businesses to locate within their jurisdictions, particularly foreign firms that could contribute knowledge spillover and on-the-job training. Much more so than in other East Asian developmental states, courting FDI was a core contributor to China’s development, particularly the portion directed as export oriented manufacturing. Suffice to say that building infrastructure for companies to benefit from was often a core part of the enticements, meaning these roles—attracting business and building infrastructure— are not totally separable.
For more details on China Development Bank’s role in creating the Wuhu model LGFV, see the first chapter of Henry Sanderson and Michael Forsythe, “China's Superbank: Debt, Oil and Influence—How China Development Bank Is Rewriting the Rules of Finance,” Wiley, 2012.
And again, as noted in a previous footnote, some may argue Shanghai created the first LGFV in 1992.
“As Figure 1.1, taken directly from a CDB presentation, shows, land expropriation and the transfer of land rights are central to making the machine work, used for paying back the loan. “The city had land but no way to turn it into cash, so the government couldn’t get money,” researcher Yu says. “At that time, no one realized what Chen Yuan knew: that once the land price goes up, you have a second source of income.”” Henry Sanderson and Michael Forsythe, “China's Superbank: Debt, Oil and Influence—How China Development Bank Is Rewriting the Rules of Finance,” Wiley, 2012, page 8.
The role of Chen Yun, economic czar for decades, in creating the LGFV model is also noteworthy.
Henry Sanderson and Michael Forsythe, “China's Superbank: Debt, Oil and Influence—How China Development Bank Is Rewriting the Rules of Finance,” Wiley, 2012, page 8.
For good useful information the legal and institutional set up enabling local government’s to exploit the land, see: Chengri Ding, Yi Niu, Erik Lichtenberg, “Spending preferences of local officials with off-budget land revenues of Chinese cities,” China Economic Review, 2014, https://www.sciencedirect.com/science/article/abs/pii/S1043951X1400131X
There were 41 instances of villagers self-immolating just between 2009 and 2012. NPR, “Desperate Chinese Villagers Turn to Self-Immolation,” October 2013. https://www.npr.org/sections/parallels/2013/10/23/239270737/desperate-chinese-villagers-turn-to-self-immolation.
China today maintains four books as part of its budget: the general public budget, the government funds budget, the state capital operations budget, and the social security fund. Locally generated revenue is booked under the local portion of the general public budget. Central funds are also transferred via the general public budget. Land sales now show up in the government funds budget. For more see Tianlei Huang, “Lessons from China's fiscal policy during the COVID-19 pandemic,” PIIE Working Papers, March 2024.
For a discussion of local government’s in the negotiation process, see: Tsinghua’s ACCEPT,改革开放四十年经济学总结, 清华大学中国经济思想与实践研究院 Academic Center for Chinese Economic Practice and Thinking, ACCEPT, December 2018, page 49, http://www.accept.tsinghua.edu.cn/_upload/article/files/7a/3c/976c22ff48cfb1860f7288f6bd05/612c2ab9-2527-4237-b28e-7ce36ee2b379.pdf
For the seven connects and one leveling seeing: 七通一平 https://baike.baidu.com/item/%E4%B8%83%E9%80%9A%E4%B8%80%E5%B9%B3/801741
Tsinghua’s ACCEPT also wrote a rousing defense of the importance of the Party-state stepping in to help with land development: “land conversion is a crucial factor in the process of economic development that has been grossly under-emphasized in modern economics. Land is indispensable for the development of most economic activities, especially for developing countries that have not yet completed industrialization, and how quickly land can be converted from agricultural to non-agricultural land has an important impact on the process of industrialization and urbanization. Modern economics assumes that the process of land conversion is spontaneous through Coasean negotiations, but in reality, the transaction costs of Coasean negotiations are often high, so the process of land conversion, if spontaneous, will be expensive and slow.” See: 改革开放四十年经济学总结, 清华大学中国经济思想与实践研究院 Academic Center for Chinese Economic Practice and Thinking, ACCEPT, December 2018, page 35, http://www.accept.tsinghua.edu.cn/_upload/article/files/7a/3c/976c22ff48cfb1860f7288f6bd05/612c2ab9-2527-4237-b28e-7ce36ee2b379.pdf
Additional resources: Fulong Wu, Land financialisation and the financing of urban development in China, Land Use Policy, Volume 112, 2022 (https://www.sciencedirect.com/science/article/pii/S0264837719306313)
Yi Feng, Fulong Wu, Fangzhu Zhang, The development of local government financial vehicles in China: A case study of Jiaxing Chengtou, Land Use Policy, Volume 112, 2022, 104793, ISSN 0264-8377, https://doi.org/10.1016/j.landusepol.2020.104793 (https://www.sciencedirect.com/science/article/pii/S0264837719313730)
Adam Y. Liu, Jean C. Oi, and Yi Zhang, “China’s Local Government Debt: The Grand Bargain,” The China Journal, 2022, https://www.journals.uchicago.edu/doi/abs/10.1086/717256; Adam Y. Liu, “Beijing’s Banking Balloon: China’s Core Economic Challenge in the New Era,” The Washington Quarterly, July 2023, 10.1080/0163660X.2023.2223838
See Adam Y. Liu, “Beijing’s Banking Balloon: China’s Core Economic Challenge in the New Era,” The Washington Quarterly, July 2023, 10.1080/0163660X.2023.2223838
Fuel was added to the banking system proliferation when the administrative structure of the big four/six bank branches was changed around the time of the Asian Financial Crisis in 1998. Beijing apparently wanted to deny local governments the unfettered control over branches of the big four banks that they had been exploiting. Provincial branches of state banks were abolished, replaced with supra provincial entities, and most importantly: appointment authority of lower level bank officials was stripped from local Party cadres and given to the higher ups within the banking system itself.
This major change in bank personnel management represented a switch in what is called “vertical management” (part of ever shifting tiao/kuai governance). Local leaders could no longer strong arm same administrative level bank branches for loans at a whim. As tapping deposits at the big banks became more difficult, local officials found even more incentive to expedite development of local banks. The local government mouse skirted the Beijing cat.
See: Nicholas Lardy, State Strikes Back; and Chong-En Bai, Chang-Tai Hsieh, and Zheng Michael Song, “The Long Shadow of a Fiscal Expansion,” Becker Friedman Institute, November 2016, page 6.
See Adam Y. Liu, “Beijing’s Banking Balloon: China’s Core Economic Challenge in the New Era,” The Washington Quarterly, July 2023, 10.1080/0163660X.2023.2223838
Zhang Xiaojing (Institute of Economics, Chinese Academy of Social Sciences) Sun Tao (Financial Stability Bureau, People's Bank of China), “China's Real Estate Cycle and Financial Stability (Preliminary Draft),” Hong Kong Institute for Monetary Research 3rd Seminar on Mainland China Economy Real Estate and China's Macroeconomy, July 2005, page 12. 中国房地产周期与金融稳定 (初稿) 张晓晶 (中国社会科学院经济研究所) 孙 涛 (中国人民银行金融稳定局) 二 00 五年七月 https://www.aof.org.hk/uploads/conference_detail/767/con_paper_0_203_zhang-xiaojing-paper.pdf;
Henry Sanderson and Michael Forsythe, “China's Superbank: Debt, Oil and Influence—How China Development Bank Is Rewriting the Rules of Finance,” Wiley, 2012, page 13.
See some of the examples in the first chapter of Henry Sanderson and Michael Forsythe, “China's Superbank: Debt, Oil and Influence—How China Development Bank Is Rewriting the Rules of Finance,” Wiley, 2012.
Indeed most LGFVs were likely already established prior to GFC era stimulus. In one study of county-level LGFVs, the authors note: “contrary to the common belief, most LGFVs did not emerge with the stimulus plan: 3,724 out of the 4,432 county-level LGFVs were founded before 2009.” Jianchao Fan, Jing Liu and Yinggang Zhou, “Investing like conglomerates: is diversification a blessing or curse for China’s local governments?” BIS Working Papers No 920, January 2021, page 15, https://www.bis.org/publ/work920.pdf.
In 2009 PBOC Document No.92 explicitly calls on local governments to use LGFVs
See: Feng, Y., Wu, F., & Zhang, F, “The development of local government financial vehicles in China: A case study of Jiaxing Chengtou,” Land Use Policy, 2020, page 4.
Meanwhile, here’s what CBRC said in 2009, “Encourage local governments to attract and to incentivize banking and financial institutions to increase their lending to the investment projects set up by the central government. This can be done by a variety of ways including increasing local fiscal subsidy to interest payment, improving rewarding mechanism for loans and establishing government investment and financing platforms compliant with regulations.” Document No. 92, CBRC, March 18, 2009.
“Allowing local government to finance the investment projects by essentially all sources of funds, including budgetary revenue, land revenue and fund borrowed by local financing vehicles.” Document 631, Department of Construction, Ministry of Finance, October 12, 2009.
For the regulations and quotes see Chong-En Bai, Chang-Tai Hsieh, and Zheng Michael Song, “The Long Shadow of a Fiscal Expansion,” Becker Friedman Institute, November 2016, page 10.
On the credit estimates see: Andrew Collier, “The Rise of the LGFV” in Shadow Banking in China, page 55; see for related discussion: Henry Sanderson and Michael Forsythe, “China's Superbank: Debt, Oil and Influence—How China Development Bank Is Rewriting the Rules of Finance,” Wiley, 2012, page 14.
Full catalogues were only released in 2006 and 2018. There was a net increase of 330 national-level development zones and a net increase of 850 provincial-level development zones, the only two levels included in the list. In 2018, tthere were 552 national-level development zones, 1991 at the provincial level, and countless more at the city and county level. See the NDRC’s catalogue of such parks and zones published in 2018: 中国开发区审核公告目录（2018年版）https://www.ndrc.gov.cn/fggz/lywzjw/zcfg/201803/t20180302_1047056.html; see the NDRC’s 2006 catalogue: https://www.ndrc.gov.cn/xxgk/zcfb/gg/200704/W020190905487497735524.pdf.
Additional analysis: 聂晶鑫 and 刘合林, 中国省级以上开发区分布变化数据集（2006–2018), https://www.geodoi.ac.cn/WebCn/HTML_INFO.aspx?Id=431c7a32-7b65-42f8-bbaf-1a542ab402df
For the point on consolidation, see: 聂晶鑫 and 刘合林, 中国省级以上开发区分布变化数据集（2006–2018), https://www.geodoi.ac.cn/WebCn/HTML_INFO.aspx?Id=431c7a32-7b65-42f8-bbaf-1a542ab402df
Yi Feng, Fulong Wu, Fangzhu Zhang, “The development of local government financial vehicles in China: A case study of Jiaxing Chengtou,” Land Use Policy, 2022, page 6, https://www.sciencedirect.com/science/article/pii/S0264837719313730.
The financial engineering by which this was accomplished is expertly, and at times comically, discussed in: Carl Walter and Frasier Howie, “Red Capitalism: The Fragile Financial Foundation of China's Extraordinary Rise,” Wiley, 2010.
Huarong Financial Leasing (2006), Huarong Securities (2007), Huarong Trust (2008), and Huarong Real Estate (2009) were already licensed by the time of Lai’s arrival, but many more segments—particularly investment management related—were to come. Growth in assets in the years from 2014 to 2017 were 47%, 44%, 62%, and 32% respectively before decreasing in 2018, the year Lai Xiaomin was placed under investigation. From RMB315 billion in 2012, by 2017 assets peaked at RMB1.7 trillion, increasing by nearly 6x in only five years. Huarong’s assets based on its corporate filings:
Sources: Huarong’s annual reports
Wang Zhanfeng, Lai’s replacement as Huarong Chairman, stated in the long delayed 2020 annual report: “Due to the aggressive operation and disorderly expansion of former Party Secretary and Chairman Lai Xiaomin, the Company has badly deviated from its main responsibilities and core business.”
The integrated nature of these is highlighted by Moodys in its explanation of its methodology for rating LGFVs: “Given the highly integrated nature of RLGs [regional and local governments] with local SOEs and banks, a typical avenue for an RLG to support a stressed entity is to orchestrate support through local SOEs and the local banking sector by using their balance sheets to support that entity (collectively, we refer to the provincial-level government and its SOEs and banking sector as the provincial system).”
Jianchao Fan, Jing Liu and Yinggang Zhou, Investing like conglomerates: is diversification a blessing or curse for China’s local governments?” BIS Working Papers, January 2021, https://www.bis.org/publ/work920.pdf.
Chong-En Bai, Chang-Tai Hsieh, and Zheng Michael Song, “The Long Shadow of a Fiscal Expansion,” Becker Friedman Institute, November 2016, pages 24-5; IMF, Local Government Financing Vehicles Revisited, February 2022, https://www.elibrary.imf.org/view/journals/002/2022/022/article-A003-en.xml
The company made headlines in 2023 as the first LGFV pushed into the Party-state’s newest round of local government debt restructuring. Facing increasing financial difficulty, Party-state officials have also stepped into to force the LGFV’s primary bank creditors to accept a restructuring of its loans—the first policy induced restructuring of its kind. The nature of its operations, likely arising in part to circumvent regulatory tightening, ultimately led many affiliated with the company into central and provincial government cross-hairs. As of early 2024, several officials involved with the LGFV have been snared in anti-corruption investigations.
These audits, however, went with differing classification schemes for what types of LGFV debts would be included and calculated as part of implicit government debt. Many liabilities of LGFVs were determined to be market-oriented, rather than public welfare oriented, and therefore not implicit debt and not included. This makes it difficult to know the full extent of debt even with the audit.
Yi Feng, Fulong Wu, Fangzhu Zhang, “The development of local government financial vehicles in China: A case study of Jiaxing Chengtou,” Land Use Policy, 2022, page, 4, https://www.sciencedirect.com/science/article/pii/S0264837719313730.
IMF data includes 2,200 annual reports https://www.elibrary.imf.org/view/journals/002/2022/022/article-A003-en.xml#A003fn04; Rhodium data includes 2,982 LGFV annual reports https://rhg.com/research/tapped-out/; MacroPolo includes 2,500 annual reports https://macropolo.org/digital-projects/china-local-debt-hangover-map/#overview.
Chen, Z., He, Z., & Liu, C, “The financing of local government in China: Stimulus loan wanes and shadow banking waxes,” Journal of Financial Economics, 2020, page 60.
Zhiguo He and Wei Wei, “China's Financial System and Economy: A Review,” Annual Review of Economics, 2023 https://doi.org/10.1146/annurev-economics-072622-095926.
Zhiguo He and Wei Wei, “China's Financial System and Economy: A Review,” Annual Review of Economics, 2023 https://doi.org/10.1146/annurev-economics-072622-095926.
Yi Feng, Fulong Wu, Fangzhu Zhang, “The development of local government financial vehicles in China: A case study of Jiaxing Chengtou,” Land Use Policy, 2022, page, 5, https://www.sciencedirect.com/science/article/pii/S0264837719313730.
Chen, Z., He, Z., & Liu, C, “The financing of local government in China: Stimulus loan wanes and shadow banking waxes,” Journal of Financial Economics, 2020; Zhiguo He and Wei Wei, “China's Financial System and Economy: A Review,” Annual Review of Economics, 2023 https://doi.org/10.1146/annurev-economics-072622-095926.
Provincial-level LGFVs accounted for 38% of total issuance in RMB value, prefectural cities 34%, and county-levels 29% (cities and counties issue far more bonds, but the average value of provincial bond issuances is far higher).
Based on above data, as of 2015, provincials accounted for 38% of total issuance in RMB value, prefectural cities 34%, and county-levels 29%.
Yongheng Deng, Understanding the Risk of China’s Local Government Debts and Its Linkage with Property Markets, IMF & National University of Singapore, 2015. page 17.
Yongheng Deng, Understanding the Risk of China’s Local Government Debts and Its Linkage with Property Markets, IMF & National University of Singapore, 2015, https://www.imf.org/external/np/seminars/eng/2015/housingchina/pdf/Session%203_YDeng.pdf
Chen, Z., He, Z., & Liu, C, “The financing of local government in China: Stimulus loan wanes and shadow banking waxes,” Journal of Financial Economics, 2020, page 52.
Chen, Z., He, Z., & Liu, C, “The financing of local government in China: Stimulus loan wanes and shadow banking waxes,” Journal of Financial Economics, 2020, page 44; Zhiguo He and Wei Wei, “China's Financial System and Economy: A Review,” Annual Review of Economics, 2023 https://doi.org/10.1146/annurev-economics-072622-095926.
Chen, Z., He, Z., & Liu, C, “The financing of local government in China: Stimulus loan wanes and shadow banking waxes,” Journal of Financial Economics, 2020, 43.
Viral V. Acharya Jun “QJ” Qian Zhishu Yang, “In the Shadow of Banks: Wealth Management Products and Issuing Banks’ Risk in China,” February 2017, https://jrc.princeton.edu/sites/g/files/toruqf2471/files/qian_jun-shadowbank-china-aqy-10feb17-all.pdf.
Logan Wright, “Grasping Shadows: The Politics of China’s Deleveraging Campaign,” CSIS, April 2023, https://www.csis.org/analysis/grasping-shadows-politics-chinas-deleveraging-campaign.
“With increasingly pressure from LDR [loan-to-deposit ratio] regulation, big 4 banks became more aggressive in the deposit market. This put SMBs into an even more difficult situation in the deposit market, and the deposit war among banks got worse and worse. First, to circumvent the regulatory ceiling of deposit rate, banks offered extra gifts to depositors as long as they deposit certain amount of money to the bank on certain days. These gifts included cooking oil, gold bar and even cash. The closer it was to the day when LDR was censored, the more generous the gifts were. Second, the war among banks went down to individual level. Every individual employee of banks was assigned a certain amount of deposits that the employee must attract before some deadline. Failure to reach that amount would cause deduction in salaries and even loss of the job. Everyone at the bank was doing everything they could and making use of every relationship they had to attract as many deposits as possible.
The chaotic phenomenon soon caught the attention of CBRC. Concerned about the effectiveness of the interest rate policy, CBRC forbade banks from giving extra gifts in all forms to depositors. The WMPs, however, seemed to gained favor from the CBRC. Since there was no restriction on the interest rate of WMPs, and WMPs were implicitly guaranteed by the banks, WMPs were in fact deposits with no interest rate control. WMPs were therefore thought by the government to be a tool to slowly liberate the deposit rate. With acquiescence from the CBRC and the need to attract savings, issuance of WMPs soon took off.”
Story from: Viral V. Acharya Jun “QJ” Qian Zhishu Yang, “In the Shadow of Banks: Wealth Management Products and Issuing Banks’ Risk in China,” February 2017, page 17, https://jrc.princeton.edu/sites/g/files/toruqf2471/files/qian_jun-shadowbank-china-aqy-10feb17-all.pdf.
Franklin Allen, Xian Gu, C. Wei Li, Jun “QJ” Qian, Yiming Qian, “Implicit Guarantees and the Rise of Shadow Banking: the Case of Trust Products,” Forthcoming Journal of Financial Economics, April 9, 2023, page 1, https://papers.ssrn.com/sol3/papers.cfm?abstract_id=3924888
Viral V. Acharya Jun “QJ” Qian Zhishu Yang, “In the Shadow of Banks: Wealth Management Products and Issuing Banks’ Risk in China,” February 2017, pages 7-8, https://jrc.princeton.edu/sites/g/files/toruqf2471/files/qian_jun-shadowbank-china-aqy-10feb17-all.pdf.
Chen, Z., He, Z., & Liu, C, “The financing of local government in China: Stimulus loan wanes and shadow banking waxes,” Journal of Financial Economics, 2020; Zhiguo He and Wei Wei, “China's Financial System and Economy: A Review,” Annual Review of Economics, 2023 https://doi.org/10.1146/annurev-economics-072622-095926.
Yi Feng, Fulong Wu, Fangzhu Zhang, “The development of local government financial vehicles in China: A case study of Jiaxing Chengtou,” Land Use Policy, 2022, page, 6, https://www.sciencedirect.com/science/article/pii/S0264837719313730.
List of important LGFV and local debt documents from 2010-2021. Many of them are not made public. A trend that continued into 2023. Policy Documents from 2010-2021 compiled by https://www.sohu.com/a/502814673_120056596
PBOC and MUHD https://www.gov.cn/xinwen/2020-08/23/content_5536753.htm
Property developers stopped relying on banks for most of their funding years ago. When the three red lines dropped it cut developers off from marginal credit. But most importantly, the shock waves crashed household confidence and in turn developers main financing channels: pre-sales, household mortgages, and self-raised funds. New starts are now down 50% from their peaks. A massive, massive correction.
Shi Dan (史丹), “Evolution of Principal Contradiction Facing Chinese Society and the CPC Leadership over Economic Work,” Institute of Industrial Economics (IIE), Chinese Academy of Social Sciences, May 2022, http://www.chinaeconomist.com/pdf/2022/2022-5/Shi%20Dan.pdf.
Xi’s so-called Economic Thought now centers around what he and the Party call a “new development concept” (新发展理念) which prioritizes “high-quality development” (高质量发展) and “innovation-oriented development strategy” (创新驱动发展战略). These slogans (提法 in Party parlance) seem vacuous. And like an empty pill capsule in need of substance, they mostly are. But one can often divine some amount of intended directionality from them, even while the precise content remains ambiguous at best.
To discuss America’s comparative advantages in national competition and the structural forces that drive (and limit) innovation, ChinaTalk interviewed Kumar Garg.
Formerly an Obama official in the Office of Science and Technology Policy, Kumar spent several years at Schmidt Futures focusing on science and technology philanthropy. He has been a mentor and cheerleader for ChinaTalk over the years, and he is the president of the newly established Renaissance Philanthropy.
We discuss:
The inspiration behind Renaissance Philanthropy and its focus on mid-scale, field-transforming ideas
Strategies for identifying underexplored, high-impact projects — including weather forecasting, carbon sequestration, and datasets on neurocognition
Structural challenges for R&D funding at the level of government and universities
The role of focused research organizations like OpenAI in accelerating progress and understanding long-term drivers of productivity
A wide angle-view of US-China competition and strategic innovation
The underresearched importance of alliance management.
Spotify:
Apple Podcasts:
The Classical Innovation Model
Jordan Schneider: Tell us about Renaissance philanthropy — what’s the backstory and thesis?
Kumar Garg: Renaissance is a young organization. Our name and thesis harken back to the Italian Renaissance. We’re inspired by how wealthy Italian families played an outsized role in supporting innovators, scientists, and thinkers like da Vinci. We’re exploring the role wealthy patrons can play today in driving a 21st-century renaissance.
There’s enormous untapped giving potential among today’s wealthy families. A study of 2,000 families with large fortunes showed they’re giving less than 2% philanthropically.
When asked why, they often cite two reasons: they’re still actively working and plan to give later, or they haven’t come across exciting ideas. These reasons stem from the same issue — they’re busy and not encountering compelling opportunities.
Rather than waiting for donors to reach a later life stage like Bill Gates, who exited his career to focus on philanthropy, we want to engage them earlier. In science and tech, it’s challenging to be strategic by just interviewing individual researchers, who are trained to pitch their next project rather than discuss field-wide challenges.
Earlier today, I was talking to somebody who was explaining the advances in neurotechnology. We have new methods and tools which should allow us to understand the way the brain works. At a technical level, these tools didn’t even exist three to five years ago. Now, we could actually build high-dimensional data sets that could allow us to understand the whole brain in a model organism.
The dataset they want to build is just too big for individual research projects, but it’s also too small for a major university capital campaign. These ideas often struggle to get funding from research agencies like NIH because they fall outside the scope of individual investigator awards.
Renaissance aims to identify these big mid-scale ideas — three to five-year efforts that could unlock a field — and match them with donors excited about the thesis. We want to draw more people into strategic giving by presenting them with exciting hypotheses and teams to support.
Jordan Schneider: How do you source these ideas and find CEOs who can execute $10-30 million projects? What should potential applicants from the ChinaTalk audience know?
Kumar Garg: Interested individuals can reach out through our website (or email info@renphil.org). We’re looking for ideas that go beyond incremental research steps to strategies that could accelerate entire fields.
We start with a field-level view, asking what strategies could help us progress faster. This requires identifying roadblocks, bottlenecks, and areas that are less incentivized for individual researchers but highly beneficial for the field.
When interviewing researchers, we often start by discussing datasets. Creating high-quality, multidimensional datasets is valuable for the community but not always rewarded in the current system. We ask researchers to describe dream datasets that could accelerate progress in their field, then work backward to determine what’s needed to create them.
We also look for projects requiring close collaboration between researchers from different disciplines, which can be challenging to fund through traditional channels.
Another approach to locating gaps is working backward from important problems. For example, famed UChicago economist Michael Kremer made a general observation that AI is improving weather prediction.
Michael spends a lot of time working on sophisticated economic models, and one of his research areas is smallholder farmers. [Ed: Professor Kremer won a Nobel Prize for his contributions to development economics in 2019]
He considered how AI could benefit smallholder farmers by more accurately predicting monsoons. If you can predict monsoons farther in advance, that could potentially increase crop yields significantly in a whole bunch of countries, such as India. Farmers could figure out when they need to harvest their crops before the rains come. Because as an economist, he could estimate the economic value for that increase in the accuracy of monsoon prediction.
Once the value is written down, then it’s clear that somebody should fund this. It’s not just general improvement in weather prediction — this particular use case has really high social value. But those individual farmers left on their own will not finance that AI improvement. Somebody has to step in and say that aggregate value is really valuable.
To find a donor for projects like these, sometimes the gap is on the demand side — the demand is latent, and you need to define it to pull people in. Sometimes the gap is on the supply side, where certain types of organizational behaviors are undervalued, like data sets.
We are often just asking good questions — “How would we go faster, bigger, better in your field? Are those big ideas hard to fit in the current funding landscape, and if so, why?”
Writing down those details makes for a very strong application.
Jordan Schneider: This seems like a lot of work for you. Is there a way to do this besides just talking to many people? What’s the non-artisanal version of coming up with these ideas?
Kumar Garg: Creating archetypes for this way of thinking is essential. I was partly inspired the work I did with Eric Schmidt creating an organization called Convergent Research, which works with focused research organizations (FROs).
With Convergent Research and the FRO model, Adam Marblestone initially received meta-feedback on his paper about focused research organizations. Once we got some FROs funded and scoped, researchers could see concrete examples and generate ideas that fit the model.
Adam now has a list of 300 FRO ideas, not because he had 6,000 conversations, but because people are coming forward with ideas that fit this model. Our job is to assess funder demand for these ideas and potentially share them with agencies like ARPA-H.
Building out the right set of questions and showing replicable examples will be crucial as we launch various funds, enabling people to envision applications in their fields.
Jordan Schneider: Focused research organizations may sound nebulous, but OpenAI is a prime example. They started with Sam Altman declaring no clear path to profitability, focusing solely on making models smarter. With limited initial funding, they took risks that weren’t being made in academia, DeepMind, or Microsoft Research. It’s a great case study of what scientists can achieve under different institutional structures and incentives, freed from mainstream corporate R&D or traditional academic funding constraints.
Kumar Garg: It’s challenging in the modern university system. While individual researchers build labs with various related projects, the idea of collaboratively solving one mega problem requires both a structure and funding that accommodates it. This approach can have a huge impact, but it’s difficult to implement within traditional academic frameworks.
Jordan Schneider: Let’s consider the individual level. Does this require leaving the tenure track or a comfortable position? Who are the people who go from having a good idea to actually implementing it, potentially taking a lateral step in their career?
Kumar Garg: You find people at different levels of the spectrum, but a common trait is the ambition to move their entire field forward. There are several ways to approach this:
Agenda setting: Researchers can take a step back, identify transformative ideas for their field, and share them compellingly with private or public funders.
Government rotations: Top researchers can spend time at agencies like DARPA or ARPA-H, bringing their orientation to find the next generation of bets.
Alternative roles: Some researchers enjoy running labs but also want to push coordinated research programs with key results. They might become DARPA performers, FRO CEOs, fund managers, or startup founders.
The challenge is that these alternate models often feel invisible or difficult to pursue. We try to make these paths more legible by providing clear examples and explanations of different roles, such as running a focused research organization or becoming a DARPA program manager.
By making these alternate paths more visible and understandable, we can encourage more innovation in career paths, similar to how the concept of being a startup founder has been popularized and made accessible to more people.
Jordan Schneider: It’s interesting because most of the time, you still need the person to get a PhD first. That process socializes you in a very particular direction. In early science, in the late 1800s and early 1900s, not everyone in the field had to go through this seven-year process where they couldn’t see the forest for the trees due to living under all these constraints. It takes an unusual person to be in that environment for 5, 10, or 20 years and still be able to recognize that everyone’s doing X, but if someone did Y for a while, it could have an incredible multiplier effect.
Kumar Garg: You raise a fair point. The computer science community is having an outsized impact partly because of AI’s rapid development, but also because of the field’s inclination. Not everyone gets a PhD, yet they can still do cutting-edge work. These individuals are now filtering through the broader science ecosystem, bringing down the average age of researchers.
Erika DeBenedictis figured out the importance of peer networks among young researchers when working on research automation. These networks can be leveraged to promote new ideas and tools. For example, to encourage more young biotech founders, engaging with the research networks of young biotech researchers and introducing the idea of founding a company can be effective.
These emergent talent networks are probably underexplored. This relates to the Renaissance point about the social networks that drive progress. In Leonardo da Vinci’s era, many interesting people were illegitimate sons of nobility who had access to resources but lacked predetermined social obligations, allowing them to forge their own paths.
The science community will always benefit from attracting people who want to go off the beaten path, but we also need to build more paths to make it less challenging for them.
Jordan Schneider: Let’s discuss the donor side. What range of net worth is worth 30 minutes of your time? Why do you think this approach resonates more than other options available today?
Kumar Garg: My involvement in philanthropy was somewhat serendipitous. While working in the Obama White House on science and tech policy, I realized that while the government brings enormous scale to its initiatives, the early work of identifying opportunities and organizing workshops is often too early for major federal agencies to get involved. This is where funders can play an outsized role.
Philanthropy and donors play a crucial part in an ecosystem where most R&D funding comes from the government. However, donors can contribute significantly to agenda-setting, identifying major opportunities, fostering interesting collaborations, and addressing overlooked areas.
Catalytic capital can have an outsized impact when used effectively. We’re at the beginning of unlocking this capital, and most of the giving that will happen hasn’t occurred yet. We have a huge opportunity to capture imaginations and harness resources for big public goods and ideas that solve problems.
When talking to donors, donor advisors, or staff members, I often help them understand that there is important work to be done. Many people feel paralyzed when faced with large-scale issues like climate change, where the government is already spending huge sums. The key is to provide them with a mental model at a more micro level of what’s not being done that could make a significant difference.
For example, we provided early funding for enhanced rock weathering, a climate solution that wasn’t widely known two years ago.
Basically you can break up certain types of rock and actually use it as fertilizer and it actually functions as a carbon sink and as a fertilizer. I remember having this conversation with Eli Dorado a few years back where he recommended we look at enhanced rock weathering.
Those early grants from a small set of folks helped get that subfield going. Now, through the work that Stripe is supporting, this is considered a pretty important technology that might help with carbon approval. There are donors who work on smallholder farmers in Africa who feel this might be a core part of their strategy and increase farmer income.
Giving people examples — and a sense of confidence that their ideas didn’t just miss the boat entirely — shifts their mental model.
“We are figuring out new things. They’re going to be important to solving big problems, and you can help figure those things out.”
It pulls them in. My hope is that they get addicted to it, and their core motivation becomes making the world a better place. We all benefit from that.
Jordan Schneider: It’s fascinating how your career experiences have led you to this point. You needed to be part of a behemoth organization to see its blind spots. The same goes for PhDs who become frustrated when they enter government.
Your background, including working for one of America’s wealthiest individuals, presumably helps when pitching to donors. What you’re proposing is fundamentally a high-variance, unconventional idea that isn’t typical philanthropy, like donating to Massachusetts General Hospital. It’s something that other philanthropists, such as those at the Hewlett Foundation, have overlooked when it comes to climate issues. It’s an interesting combination of factors that brought you here today.
Kumar Garg: Indeed. I benefit from seeing the system from various perspectives. This is why I recommend people enter government or move between different sectors, becoming “tri-sector athletes.” It broadens your understanding of what’s possible and what’s missing.
One of the real blind spots in science and tech funding is that people often start from basic first principles, asking if an idea is good. However, if you’ve spent time in government, you’d ask why the government isn’t funding this idea if it’s so promising. There’s usually an entire agency dedicated to funding good ideas in a specific field. This line of questioning leads to interesting insights.
For example, I spoke with a venture capitalist in early healthcare who wasn’t familiar with NIH’s Small Business Innovation Research (SBIR) program. When I mentioned that SBIR’s funding matches the VC investment in early-stage healthcare (about $2 billion each), it raised questions about potential disconnects between government funding and commercialization efforts.
Being curious about why systems work the way they do and why certain opportunities are missed can lead to significant breakthroughs. You’re dealing with substantial funding flows, so understanding these dynamics is crucial.
Jordan Schneider: We’ve discussed the structural issues in government funding. Regarding established philanthropies involved in climate or science and technology, where do you see structural oversights in their programs and among their program officers?
Kumar Garg: Looking at both the public and private sectors, one significant gap becomes apparent when you step back: we allocate R&D money to some societal problems but not others. Once you realize this, it’s impossible to unsee.
As a society, we invest heavily in defense R&D, we allocate a decent amount to health and new therapies, and we spend smaller amounts on energy and space. After that, the spending drops off significantly. However, if we consider the major aspects of human flourishing — food, housing, work, learning, building strong relationships with friends, family, and community — we see a discrepancy in R&D investment.
We spend very little on R&D for criminal justice, education, workforce development, and housing. Midway through the Obama administration, around 2012, I read a paper suggesting agriculture would be a significant driver of future climate emissions. When I inquired about the R&D component for agriculture in our climate strategy, people were puzzled about what that might entail. Alternate sources of protein sounded completely random back then.
The same happened with transportation R&D early in the Obama administration. How could we make roads better? What do we need to do to prepare for the emergence of automated vehicles? What does the budget look like for impactful projects in these areas?
“Pavement durability & sustainability deserves more attention than it gets,” Sec. Pete Buttigieg opens a pavement testing facility at the Turner Fairbank Highway Research Center, 2023. Source.
Of course we should have R&D in these fields, but we’re significantly underinvesting in R&D for social sector aspects.
This oversight leads us to discover innovations accidentally through national security spending. For instance, the Department of Defense has been a major funder of alternative medicine to ensure the health of returning warfighters, which is a prety roundabout way to fund that research. Similarly, some of the most interesting education and workforce experiments have been funded by the DoD because we allocate almost no R&D dollars to the Departments of Education and Labor.
If you’re a funder working on mental health, the environment, or homelessness, consider the R&D component of your strategy.
If you can’t identify one, don’t dismiss it—this might indicate a huge opportunity to develop that component. Otherwise, you’re simply waiting for accidental positive outcomes rather than actively investing in shaping the future.
Our view used to be that R&D should be the third bullet point of any plan. If you can’t articulate what the R&D would entail, you’re in an even worse position—people haven’t even conceived of that component of the strategy. This blind spot leads people to treat science and technology as a sector rather than a crucial component of any problem-solving strategy.
R&D on Strategic Competition Mechanics
Jordan Schneider: Let’s talk about China. What are the big picture questions you’re interested in regarding how the US should compete?
Kumar Garg: Here’s my question for you, or at least the hypothesis I want you to reason out with me: If the US and China are engaging in strategic competition, to what extent should they emulate each other’s strategies? To what extent, especially on the US side, which I know more about, should we play to our advantages?
I’ll throw out a few key points:
First, talent. The US is a global magnet for talent. People, especially science and tech talent, want to move here. How much should the US lean into that?
Second, the US has deep capital markets, which have created substantial room for entrepreneurship and building emerging industries on top of our university system. How can the US leverage that?
Third, the US has been home to many platform technologies in biotech, such as mRNA and CRISPR, as well as the early computing revolution, mobile technology, and now AI. To what extent should the US frame itself as a place where platform technologies thrive? What are these platform technologies?
There’s also an interesting question regarding drones.
Finally, to what extent is the healthy relationship between research in universities, formal government research agencies, and the private sector an advantage or disadvantage for the US?
I often wonder about the IRA, which is partly treated as a set of subsidies for key industries, whether in clean tech or chips. To what extent will subsidies be a major part of the future picture versus these core US advantages?
So, what do we need to improve? What should we double down on? What should we copy?
Jordan Schneider: I’m going to approach this from a different angle. I’ve been reading a lot of Paul Kennedy recently, one of the greatest living historians. His 600-page tome, The Rise of the Anglo-German Antagonism, 1860-1914, has been particularly enlightening. It’s a deep exploration of diplomatic history, culture, religion, science, and media, examining how England and Germany interacted and ultimately became enemies, leading to World War I.
This book has challenged my thinking because many dynamics in the US-China context today are scrambled when considering Germany and England. For instance, England was the dominant power and Germany the rising power, yet Germany was the scientific powerhouse. Every consequential English inventor spent time in Germany to engage with their university system.
At the same time, Germany, despite being more advanced in some ways, still aspired to English culture. Their wealthiest children were sent to Oxford and Cambridge because that represented a life of luxury.
My takeaway, echoing Paul Kennedy, is that we need to zoom out further when considering a multi-decadal national competition strategy. It’s less about specific policies like subsidies or green cards, and more about the bigger picture. If you’re the larger economic power and you’re not mismanaging the transition from economic power into military power, you’re generally in a good position.
The US has a significant advantage: allies who comprise two-thirds of global GDP. If we don’t squander that, we can afford to make mistakes elsewhere and still be okay.
From CSET: https://cset.georgetown.edu/wp-content/uploads/CSET-System-Re-engineering.pdf
America’s growth, technology, and universities make us an attractive partner. There are enough repellent aspects of the CCP’s worldview and foreign policy choices that Japan, the EU, South Korea, and Taiwan probably won’t go align with China — even if the US doesn’t get everything right.
Kumar Garg: Wearing my policymaker hat, it’s clear that some of this won’t be under US control due to the two-sided dynamic. If other countries feel pushed by China, they might be drawn closer to the US. But it’s interesting to consider what US policymakers could proactively do to be better allies or to think about these longer-term trends.
Jordan Schneider: Absolutely. While we often focus on specific, tactical issues in ChinaTalk episodes, it’s important not to dismiss the broader view. If we’re risking 20% of global GDP going in a different direction, perhaps we need to increase our research tenfold to better understand what’s happening in areas like maintaining NATO and our alliance with Japan.
Kumar Garg: One reason I’ve been a huge fan of ChinaTalk is its focus on public goods — things that are valuable overall but often underfunded. While some aspects of foreign policy, like Middle East policy, are well-funded [Jordan: by Arab governments...Korea and Japan weirdly stingy both for think tankers and when it comes to influence operations and intel…], others that will matter significantly are not. We used to track this on the philanthropic side: how many people can actually speak the language and track work happening in key countries the US cares about? How much fresh analysis is happening?
My current view is that we’re moving in the wrong direction. The number of people I interact with who have active working relationships in China has been dropping for the past five years. This will impact our ability to understand how things are working.
Jordan Schneider: The structural problems you identified in the science and technology space absolutely apply to Asian studies. What I’m doing isn’t easily understood by government grants or even traditional US-China funders. There’s plenty of potential in both humanities and hard sciences, but it’s challenging to find support.
Regarding tactical advantages, the cleanest heuristic is to focus on long-term productivity and economic growth. If you can tie your initiative to that over a 10-15 year horizon and it outperforms other options, that’s the policy or donor idea to focus on.
Kumar Garg: One area where we need more activity in DC is understanding the long-term drivers of productivity. When I was in the White House, I noticed that while senior folks in the National Economic Council (NEC) and the Council of Economic Advisers (CEA) recognized the importance of R&D, immigration, and commercialization for long-term productivity, often one person was responsible for all of these areas plus four others. Meanwhile, substantial resources were allocated to issues like marginal tax rate debates.
If we’re going to succeed by identifying and focusing on long-term economic growth drivers, we still have room for improvement in how we staff and prioritize these areas in government.
Jordan Schneider: It’s understandable that in a democracy, politicians focus on issues that are more immediate to voters, like marginal tax rates. However, if we’re talking about long-term competition, we need to think beyond presidential cycles. The gains from high-skilled immigration, for example, may not be immediately apparent but can lead to significant innovations. For example, Elon Musk founded Tesla in America as opposed to anywhere else in the world.
This long-term perspective makes the challenges with NSF and NIH funding so frustrating. These institutions operate on longer timescales, and we would hope they can maintain their focus on long-term goals.
Kumar Garg: Indeed, politics is about balancing the important and the urgent. Policymakers must build systems to effectively balance these competing priorities.
Renaissance Book Recommendations and ChinaTalk Parent Corner
Jordan Schneider: How does coaching your children’s entrepreneurial ideas compare to advising me and ChinaTalk?
Kumar Garg: The challenge is that in the real world, people appreciate my advice. However, for my kids, it’s like an anti-signal. If I think something is good, they assume it must be outdated. This applies to fashion advice as well. Sometimes my daughters are in the car when I’m on the phone, and afterward, they question why I said certain things. I have to remind them that they only heard one side of the conversation.
Jordan Schneider: I hear your kids are trying to start an Etsy store, and you’ve been coaching them to find a comparative advantage as creative eleven-year-olds.
Kumar Garg: They signed up for a summer camp that promised to help them make art and sell it on Etsy. On the first day, they were disappointed to learn that the camp would only help with art creation, not selling on Etsy. They proposed selling slime, but I cautioned them that it might be challenging since every kid is into slime.
Jordan Schneider: What’s their annual Etsy purchasing budget?
Kumar Garg: They spend all the money they can get — from grandparents or elsewhere — on various types and textures of slime. They spend considerable time explaining to me how each type is different. We visited the Sloomoo Institute in New York, which is essentially a slime museum. Every eleven-year-old seems to know about it. I suggested they should open one in DC, as it would be an instant pop-up success.
It’s fascinating to observe future trends through my children. They recently mentioned that all their friends are into boba tea, which I now see everywhere among young people. Fifth-graders appear to be early trendsetters.
Jordan Schneider: To close, let me tell you about my favorite Renaissance books. Lauro Martines is one of the more entertaining historians I’ve encountered. He has a threebookseries on political history during the Medici era. I’ve read extensive Chinese, ancient Greek, and Roman history, but Florence offers a wealth of juicy sources. There are numerous diaries and letters detailing the conniving and intrigue, including murders in churches. It’s captivating because, unlike a monarchy, Florence had competing families and factions. With the Pope nearby and the French making appearances, it’s more thrilling than Game of Thrones. Martines’ work is essentially a historical beach read.
For primary sources, I recommend The Decameron. Written during a plague year, it’s framed as a dinner party in the hills where 15 young people attempt to outdo each other with stories while fleeing the plague. It’s incredibly relatable and alive, despite being 600 years old. It’s a funny, entertaining, and at times profound piece of literature that transported me back to that era more vividly than anything else I’ve encountered. While reading Machiavelli is one thing, imagining people having fun in the Renaissance was a unique experience.
How’s this for something deeply un-antisemitic from 600 years ago!
Michael Collins is the acting chair of the National Intelligence Council (NIC). He has spent 28 years in the intelligence community, starting as a career analyst focused on East Asia before moving into leadership roles. He served as chief of staff for the CIA deputy director and worked on modernization efforts in the agency.
We discuss:
How the intelligence community informs high-level policymaking,
Why different institutional approaches are needed to collect intelligence on non-state actors vs nation-state adversaries,
Challenges in assessing China’s technological and military capabilities,
“Narrative Intelligence” and areas where intelligence agencies have a unique edge,
Strategies for improving long-term forecasting and avoiding groupthink.
Jordan Schneider: Let’s have you introduce yourself. Can you briefly tell the audience a little bit about your journey through the intelligence community and what the NIC is?
Mike Collins: I’ve been in the intelligence community now for 28 years, and came in as a career analyst long ago after doing graduate work in international affairs.
Most of my career was as a career analyst, mostly on East Asia. I became a manager in various capacities, and worked a bit overseas to get a feel for the mission as well. At one point I was the chief of staff for the deputy director. I was privileged to work on modernization reviews in the CIA.
At one point, I was the deputy director of one of our integrated mission centers at CIA — the East Asian Pacific Mission Center — where I got a feel for bringing together all the pieces of the intelligence world for what the agency can accomplish. Previous to this job, I was the chief strategy officer at the CIA, working for Director Burns on a number of things. In particular, the initiative he put forward to review the structure and posture of the CIA for major power competition at its core and all the different derivative recommendations that came out of that. I could speak to some of that.
Then I got an opportunity to come over to the National Intelligence Council. The National Intelligence Council at its core is the lead representative and producer of the US intelligence community’s overall analytic view of issues in the world. It’s distinguished from all the individual agencies in that we speak for the entirety of the US intelligence community, not one agency in particular, and again, the lead producer and representative of that.
In the spirit of ensuring our policymakers receive the widest, most diverse, and most inclusive view across the community, we take seriously the role we have in bringing in the private sector. We are constituted right now by 18 different programs.
National Intelligence Officers make up the NIC. Those are the senior representatives in each of the regional or functional aspects that we cover in the world. Their task is to be the principal lead in their respective area.
Most importantly, our signature products are the national intelligence estimates, the long-term future anticipatory products of the US intelligence community, looking at key trends and strategic developments in the world, as well as the signature global trends product that’s produced every five years. Now we will be going out to 2045, assessing where the world will be trending then.
In recent years, this NIC has become very busy as well with direct policy support for policy meetings downtown. When there is an intelligence briefing required to start a national security conversation, the practice is to have the Intelligence Community brief an assessment pertaining to that issue. The National Intelligence Council is taking the lead on providing the intelligence on a recurring manner for that, truly at record numbers, historic numbers.
Finally, I would just say our role as the lead analytic representative for outreach and engagement. Kind of what we’re doing actually here to relate more, to inform and to support our government’s conversations with America, frankly, about the analysis we produce and the collaboration we need as well for the use of such analysis for our engagements with partners around the world.
Comparative Advantage in Asia Analysis
Jordan Schneider: You and Kirk Campbell, also a former China Talk guest, are among the few folks who’ve risen to very senior levels of government and have decades of experience engaging deeply with East Asia.
This perspective is fascinating. In the 1990s, intelligence officials who studied the Soviet Union all of a sudden ended up working on Iraq policy. Now, a lot of people who studied the Middle East are involved in East Asia policy. 10 or 20 years from now, it’s likely that we’re going to have a lot more people with deep engagement and experience in East Asia running a lot operations in different regions.
As someone who has been studying Asia since the beginning of your career, what has it been like to watch your region enter the spotlight? What makes your perspective unique compared to newcomers who are strictly focused on understanding the PLA?
Mike Collins: At a broad level, it’s important to reflect on the expertise and experience we have covering what we call, “hard target challengers” of the United States and the world — those countries that are viewed as strategic competitors of ours. We call them hard targets because they’re hard to understand and hard to collect on relative to what we call the global coverage arena. That’s sort of the rest of the world that we have to be conversant and understanding of.
Throughout my initial time thinking of East Asia, I came to appreciate all the more thinking of China as part of East Asia and of course, the dynamics that relate across that region in terms of relationships and influence and standing, understanding the governments of the region themselves and the populations themselves and the relative sort of standing our government has or what we stand for in the region relative to those of a different state, in this case, China in particular.
I would highlight three things in particular over the years that have deepened my thinking, not just in this arena, but also around the world.
1. Over the last couple of decades, we have come to appreciate more the inherent normative roots, if you will, of the conflict contestation we have with the PRC.
This is what the PRC (and under Xi Jinping in particular) defines as a geopolitical competition — they describe it as such themselves.
But when you peel the onion, at its core, it’s a difference between what we stand for, how we govern, and what we advocate in the globe — and what this authoritarian state stands for.
Especially now that Xi Jinping is increasingly doubling down.
We have to appreciate that more, and we have to understand the dynamics within China, not to think of it just as a monolith in terms of the conversations it’s having with its people, its business community, its tech community, etcetera.
Sometimes I’m asked, “When we double down, are you just focusing on China? Does that mean we don’t need to focus on anything else?”
On the contrary — it means we actually have to be better at understanding the rest of the world. What makes the rest of the world tick?Where do these populations stand, as I said earlier, in the global competition? What are the roots of and sources of unrest in the rest of the world?
For the last two decades, the intelligence community understandably focused on terrorism.
Now we’re getting back to thinking more about the study of international relations more broadly in the world.
We have to understand that increasingly all the more certainly apply that in the work we do at the NIC.
The last area is those domains that increasingly will matter for this major power competition. Technology clearly is one, and we all talk about that a lot. Obviously, we prioritize that. Director Burns certainly did when we did the reviews in the agency.
These domains that govern international affairs— we also have to be increasingly smart on, conversant on, and appreciative of the role they play, including through the gray zone. The gray zone is that area within which rivals are trying to compete with the United States without resorting to direct major power conflict, but also still right of what we consider traditional or accepted forms of statecraft.
Those areas, I’ve appreciated all the more, beginning with my works, if you will, on East Asia and thinking of East Asia itself as a region and a collective, and now thinking more broadly in this capacity, applying the same to the global landscape.
Jordan Schneider: There is an ideological competition with ISIS, but we aren’t super worried about their quantum advancements.
Reflecting on your time doing this reorganizational work in the CIA, what needs to happen at an institutional level in order to effectively analyze a country like China as opposed to the Taliban?
Mike Collins: I was asked a very similar question early on in the process of the review. In any strategic approach to reform, you try your best to parsimoniously get at what at its core will most improve your strength. When I responded to that question, and I still do hold to this today, it’s about relationships and collaborative, more integrated relationships, and leveraging capability where it exists.
I’m a career CIA officer. One thing I really appreciate when I think of the agency and its five directorates — that place is really strong in leveraging all of the unique capabilities and authorities that each of those directorates have, whether they’re as service providers or as their mission executors.
I’ll speak here for the NIC. The National Intelligence Council is not an entity unto itself. We are actually a relatively small unit. We probably only have around 100 analysts in the National Intelligence Council, give or take, covering the entire world. We succeed to the extent that we leverage the unique strengths of analysts across the entirety of the IC, as well as increasingly experts in the private sector, be they in academia or in the business or commercial sector. This is a more federated approach to taking and utilizing capability where it rightfully sits.
I say that in part because those analysts, for example, who are sitting in one particular area, bring to that analysis something very unique because they’re part of something or living in something that not everybody actually does. If you think of an analyst who works for a company whose job it is to make money selling products around the world — they’re thinking differently about their risk analysis. I could apply the same to the analysts in the State Department, the analysts in DIA, the analysts in CIA, each of which are stronger in a unique way because of where they actually sit. We have to leverage that more.
Again, I would say at its core, on the various reviews I’ve been a part of, if I had to distill it down to one thing that was probably most critical, it wasn’t necessarily about new structures or necessarily new positions, but can we maximize the strengths that come from a more, if I will, federated, but at the same time more collaborative and integrated effort, whether that’s on the responsibilities and mission execution capability of a unit or its enablement, be it substantive or be it, say, in the training or workforce or governance domain.
Jordan Schneider: I want to talk about the comparative advantage of the intelligence community, both on a domain level and a time horizon level.
Let’s do domain first. You mentioned I can totally see that IC would have an advantage in understanding some particular artillery piece in an adversary’s army, just because there’s no other body in the private sector or academia that would spend the time or money on that.
But when thinking about understanding the Chinese economy, the advantages you get from secrets insight are less obvious, to me at least. How do you think about allocating resources? What do you guys think you can and will continue to always be able to do better than other agencies? What areas benefit most from outside perspectives and voices?
Mike Collins: I love the question in a couple of ways. Let me set the table first by stressing one of the things I’ve learned in my career — the increasingly important and yet unique role that intelligence itself plays for our security as a national security asset.
It’s noteworthy that our national security strategy now includes references to intelligence as one of those categories of national power.
Heretofore, most national intelligence security strategies would reference military, economic, and diplomatic power — those normal, traditional elements of power. But now we’re talking about intelligence itself.
When I talk about intelligence, I mean that in its purest literal sense — that is, I’m more intelligent and smart about something than I was a year ago. Even in a competitive sense — am I more knowledgeable about this issue than my adversary?
One of the things we try to take advantage of to be competitive and what the intelligence community can provide are our relationships, our partnerships around the world, and increasingly our deeper partnerships with the private sector and the commercial sector. We benefit from the fact that the United States is as big as it is, and we’re as present as we are around the world. We benefit as well because we have responsibilities, therefore, unlike any in the world, to stay on top of those same issues.
In so doing, you have to have eyes on a lot of issues around the world to ensure you’re providing maximum coverage for the policymaker. But in so doing, you’re learning something beneficial that will inform the analysis we produce.
We know there will continue to be an exquisite need for the hardest of the intelligence we can collect that gives us a unique perspective on the motivations and thinking of our adversaries around the world. Increasingly, as these adversaries are clearly trying to cooperate even more together, we will need that. But even to that end, to be successful, we know we need partners and avenues to allow us to accomplish that effectively.
At the same time, we know increasingly, for all the expertise and insight that can be acquired and investigated openly in the open source arena, we would be hurting ourselves if we weren’t doubling down on that as well. We have to leverage, and we are leveraging all the more what academics are studying, but even more so what scientists are learning from what they’re doing, what commercial practitioners are learning from their experience, what the financial community is learning as we take advantage of the globalized arena.
The last point on this I would highlight for the NIC as an example. We recently launched this new external research council as a council for the council effectively, whereby we’re maintaining on a recurring way experts in all of our different portfolios, whose job it is effectively to help us stay ahead of the research and the understanding, the expertise that’s being built in all our portfolios that we may not simply have the time ourselves to stay on top of as much. But the idea there is we’re acquiring and collaborating on collective analysis. I know we’re benefiting as well those on the outside who have an interest in studying the same, whether they’re in universities or in the private sector. Again, taking advantage of what work is already being done in many areas.
Jordan Schneider: Let’s stay on this one more second. You alluded to a few things — leadership intentions, hard power, and military capabilities are probably the two that most obviously come to mind as the things that no one is ever going to be able to do outside, better than the US intelligence community. What are the other things that you think are always going to have to sit within the intelligence community?
Mike Collins: There are a number of things that will always have to sit in the intelligence community. As you well know, we don’t speak about sources or the methods and the means by which we acquire what we do. There are second or third-order derivative aspects of things we do in the intelligence community that obviously we don’t speak to for the reasons I just said.
But on top of what we might be able to uniquely understand about the motivations and intentions from centers of power around the globe, I’d probably add to that — this increasingly challenging geopolitical landscape around the globe is to some extent different than the Cold War. Big countries around the world are not necessarily taking sides, nor are we asking them to raise their hand to choose sides in this sort of geopolitical competition in that way.
Related to an earlier point about intelligence and the national security strategy, a lot of what we accomplish in the world and understand about the world does happen quietly. The disposition of foreign partners and governments around the world to work with us — not just the intelligence community, but the US government — is often gleaned uniquely from what we know our partners around the world are potentially willing to do for us in more sensitive areas.
I do think that will be an increasing source of unique expertise that we do have. Again, as we appreciate the more behind-the-scenes, sort of quiet ways in which we’re trying to form and collaborate.
It’s just not going to be as likely that people will be raising their hand and voting between one side or the other. But let me add one more point. As clearly as I know we can acquire and understand and think about our leaders in various ways from what we uniquely do, we are also reaching out to look at other ways by which we get indicators of intent — through behavior, sentiment, narratives, public conversations leaders are having with their individuals, through what they’re doing with their economy.
You know this just as much as I do — somebody on the outside can make powerful judgments about what the intentions or readiness or inclination of a country might be, based on other indices they observe in the public arena.
Those are things that are increasingly important for us as well, because they sometimes can illuminate and provide strategic warning of where something might be going, even if it can’t provide the proximate sort of indicator of what time and space that might occur within.
Jordan Schneider: What can The New York Times and The Economist not grasp that you can? Does the edge of the intelligence community hold up over time?
Mike Collins: You’re not going to find a report today that says what will the global order look like 25 years from now. But again, our job is to put together a set of variables, a set of drivers that determine where that might be going, one of which is clearly the motivations and plans of countries around the world. What are they intending to accomplish? What are their strategies gearing them toward? What do they tell themselves internally about what their metric of success is 10, 15, or 20 years from now? The motivations of countries, where do they want to be? That piece is very critical.
At the same time, of course, as another variable we look at capability, do we objectively believe they have the capability to get to where they want to go? And to the same, do we understand? Do we believe what we objectively say about their capability is what they also agree their objective sense of their capability is? The latter is something, again, we might more uniquely understand than somebody on the outside potentially might. But you can glean that as well from what they say publicly, often to their audiences.
The last variable in that same forward-looking anticipatory analysis is what’s the rest of the world doing? That permissive environment. Its motivations, its capability. But what’s that arena look like, is it vulnerable? Is it permissive? Is it resilient or not? Is the rest of the world responding differently?
Those are all areas that we have to look at when we do our forecasting. But you’re right in saying that as we get further and further ahead, the less arguably relevant, although not completely irrelevant, is the unique information we have on what countries are planning internally.
Intelligence and Long Time Horizons
Jordan Schneider: I’m interested in the longer-term perspective, particularly from the standpoint of capturing policymakers’ attention. If you drive down to 1600 Pennsylvania and say, “We think Putin’s going to invade Ukraine next week,” that’s obviously something people will wake up and pay attention to. But you have to think about the long term as well. How do you approach the challenge of selling the idea of 5, 10, and 20-year horizons to the people who are spending money and making decisions today?
Mike Collins: I’ve thought a lot about this. It’s a great question. I’ve been honored to brief various senior government officials and leaders. That question often comes up when considering whether to provide analysis beyond the four or eight-year cycle. My response is, to the point I just made about the strategic trajectory 20 or 25 years from now, what the world will look like in terms of the balance of power between China and the United States — to the extent you can be objectively accurate in what you say, “When historians look back 20 to 25 years from now, they will examine decisions made on this particular issue, at this particular period, to determine what was critical to the trajectory.”
When we do national intelligence estimates or global trends, we prioritize those factors in our estimates. Even if we’re estimating 10, 15, 20, or 25 years out, we owe it to every policymaker to identify the factors most influencing outcomes, and particularly, what can be done about those factors now. When you approach it that way, you get traction and a response.
The last point is that you must be clear about your metrics of analysis — the indicator set. If we see development trending in certain ways, we should constantly revisit it. Whatever the issue might be, we should establish a framework. Every year we’re seeing if it’s moving in expected directions or not, objectively suggesting whether the forecast we warned of or anticipated may be happening. This keeps you in conversation with the current group, not just writing for those who may come into power two decades from now.
Jordan Schneider: We have an upcoming interview with Kumar Garg, who spent eight years at the Office of Science and Technology Policy for the Obama administration. We discussed getting policymakers focused on drivers of long-term competition. His point was that in the White House, they had 20 people focused on some marginal tax rate policy and only one or two focused on high-skill immigration, which plays a big role in the future of American innovative capacity. My initial thought was that it makes sense because people vote on their tax rates, and politicians have to get reelected.
Broadening that insight to a systems competition level, we could have a long discussion about the extent to which China is actually long-term and strategically minded. The fact is, America has really done this in fits and starts over the past 75 years. Even looking at the Cold War, there were many different directions that governments went in defining long-term competition and what policies they thought would lead the Cold War to end positively. Do you have any thoughts or reflections on this?
Mike Collins: In the intelligence community, we don’t comment on or inform what our leaders decide from a domestic political standpoint. We recognize our elected leaders have other factors that weigh into their decisionmaking. Our responsibility is to lay out the national security implications of issues around the world and what can be done about them. It’s then up to the decision-makers to consider all factors.
If we’re not effectively drawing attention to these dynamics that we believe are truly significant for national security, we’re probably not doing our job well enough. If we really think it’s that critical, we should be doing more about it.
One of the things I admire about what Director Haynes and ODNI are doing, and we’re a big part of it, is their transparency initiative. We are more openly sharing our judgments and views of the world with the public. You see this every year during the annual threat testimony, and increasingly on our webpage where we release analysis for public consumption.
We have a responsibility to inform the larger narrative about what matters in the world if we’re going to be effective. That’s part of the reason for the transparency initiative — to open up and engage in a broader conversation with the larger American and global audience about what we see in the world. We do this hand-in-glove with policymakers.
We also do this because we need to be challenged. We need talent and insight from the private sector, and we need to avoid groupthink. We take seriously our responsibility for modeling objective critical thinking not influenced by politics, to ensure that what we write does not look like it was written specifically for a political or policy agenda.
Jordan Schneider: Staying on the domestic theme for a moment, let’s consider the concept of net assessment. Some of the CIA’s most famous accomplishments over the ’70s and ’80s with respect to the Soviet Union were doing this long-term economic analysis and grasping — way before much of the broader community — that the Soviet Union was growing a lot slower than maybe even someone in the Kremlin understood.
When doing net assessment, it’s fine for you to compare Western military capabilities to Chinese ones. But as you’re thinking about other domains of competition, like emerging technologies or international trade and investment, there’s a unique handicap in not being able to comment directly on American capabilities and policy options. How do you think about getting around that challenge?
Mike Collins: That’s a great question. Across the Intelligence Community, we’re looking more seriously at how we understand and can work with non-state actors or entities. This relates to domains that increasingly matter for raw US geopolitical national security, but also transnational security in the global commons — climate, health, and disease. We know we need to be doubling down on engaging with and understanding these arenas and the experts in them. We need their talent, expertise, and tech. We also need to think about the threats they themselves are under because they matter increasingly for national security beyond the traditional military domain.
To that end, we are starting to flex our muscles, using the net assessment concept, sometimes literally, sometimes metaphorically, to look at those other domains. We’re examining the balance of relative strength in techno-economic categories, media, education, science, and all those different areas of international affairs that reside outside of traditional state-to-state diplomacy and are in the private sector arena.
Obviously, we do not collect specifically on the United States, but to better understand the severity of a threat that would come from an adversary being more dominant in microelectronics, biomedical, media, or any other issue, we do need to understand and be conversant on the relative strengths and resilience of the United States in that same area, as well as our partners around the globe. This helps us truly understand the threat that would come from another country, an adversary, or a geopolitical challenger being more dominant in that space than they currently are.
It also informs our domain analysis, helping us step back and think about what determines success in these areas. There’s a lot of talk about AI, for example. As you can imagine, there’s a lot of work being done on the relative strengths of various ecosystems in the AI space.
This net assessment idea is increasingly what folks should start to hear more from us. We obviously have to do it in partnership with experts outside of the US government to be effective at it.
Jordan Schneider: Let’s come back to this idea of transparency. It’s one thing to write and publish more unclassified material about long-term trends and threats, but it’s another to do what we’ve seen over the past few years, particularly in the lead-up to the war in Ukraine. We’ve also seen this with the PLA purgings and the water and rockets arc that rippled through global collective consciousness over the past year or so. How do you think about those types of disclosures, maybe in the China context in particular?
Mike Collins: I’ve come to appreciate what I sometimes call “a new INT,” — NARRATIVE INT. Narrative warfare, information warfare — sometimes people refer to it as narrative contestation. There’s a lot of talk in the national security conversation lately about misinformation and disinformation in the global arena. Frankly, that’s both a supply and a demand problem.
We look at, investigate, identify, and talk about the threats our adversaries are attempting through the use of misinformation and disinformation. What President Putin did before the Ukraine operation was an act of misinformation, trying to disinform and misinform about what was going on in Ukraine. The same is happening in China. The conversation the government of China is having with its people would arguably benefit from some rigor in terms of separating fact from ideology.
When you step back and appreciate what’s happening through the use of social media and what it has enabled relative to what traditional media can provide in a more credible sense, I worry a great deal not just about the provision of misinformation and disinformation, but increasingly about the resilience of human beings to the truth - what the truth really is, what it represents, and whether people care about the truth.
One of the realities in this great power competition we have with geopolitical challengers is the difference in our relationship with the truth. Unlike us, our authoritarian rivals are often not accountable to the truth. Increasingly, as you look at what they’re doing to double down on their security architecture and what they’re doing to surveil and control narratives at home, in many cases, they fear the truth for what it represents in terms of things that have happened in their own country or what’s really happening broadly around the world.
I think this is going to be an increasingly powerful means through which we compete and succeed globally. It’s not just a one-off release of information. We need to call certain things out, but I think to the extent that we’re seen as more credible and impactful, we should be driving more open conversations about narratives that are out there that we can say, with objective, empirical evidence, are just not true.
We should think about the relationship between the state of an authoritarian competitor and its people. What do its people fundamentally believe about what’s true, for example, about what happened with COVID? We talk about vaccines and the cooperative approach or lack thereof towards sharing information pertaining to that disease. You know this as well as I do — what that doctor in Wuhan was trying to express. The state of China did not want the truth to be exposed, and we should reflect on that.
Jordan Schneider: We’re recording this on July 26, a few days after Biden gave his “I won’t run for reelection” address. One of the lines that really stuck out to me was him saying, almost as an aside, “When I came into office, the conventional wisdom was that China would inevitably surpass the United States. That’s just not the case anymore.”
When I reflect on the past five years of that narrative shift, I don’t think that was some CIA operation or psyop to convince the world of this. What ends up changing people’s minds are economic growth rates and how different countries relate to each other. You can only play so many games around that in altering people’s perceptions.
You mentioned the COVID thing, and hopefully this wasn’t your fault, but I have to bring up the article in Bloomberg which related a story, I think coming out of the Defense Department, where they paid for information operations in Tagalog saying that the Chinese vaccine was going to make your kids autistic or something.
My piece of advice to you, Michael, is that I think you’re almost underrating the power of truth, accuracy, and empiricism to win out over the long term. You can only really play games a little bit on the margin, but I don’t think the Chinese people are stupid. I don’t think the American people are stupid. I don’t think folks in the rest of the world are stupid. A strategy that bases your information operations agenda around trying to pull the wool over people’s eyes is not one that’s going to serve you well over the long term.
Mike Collins: I’m going to challenge that, Jordan. I’m not talking about pulling the wool over somebody’s eyes — to the contrary, I’m speaking about exposing the facts. I’ll be the first to admit we’re not always going to be right, but if we’re seen as trying to expose credible facts about something, the more successful we are in getting those facts out and understanding them over time, the better. We leave it to others to debate that.
I’ll just highlight the one fact I conveyed before — that poor doctor was trying to express that something was happening in Wuhan, and he was prevented from doing so. That’s a fact. That’s an example of the differences between our norms and values and freedoms of expression, as well as the liberties that we stand for relative to what the state of China stands for.
That’s an example of a fact which, not to say it was the reason millions of human beings around the world died, but it was a reason for how quickly that disease started to spread before we got our hands around it. Maybe it was a week, maybe a couple of weeks. But again, it’s a fact and a representative fact about the difference between our relationship with the truth and their relationship with the truth that we should reflect on and be conversant about.
Jordan Schneider: Reflecting on the past few years of America and strategic transparency, this was clearly not your decision to basically stop talking about the balloon. More broadly, not everyone has bought into this line of logic. Could you talk about when it isn’t necessarily in America’s interest to air the financial or extramarital dirty laundry of strategic adversaries’ leaderships, for example?
Mike Collins: In studying the impact of objective narratives around the world and what tends to stick, often it’s when you connect with something that’s important to another human being or polity around the world. If all we’re doing is being seen countering something or denigrating or criticizing something, that’s not good enough. There’s a role we can play in objectively trying to drive an understanding, as empirically founded as possible. People will always challenge it, and we all have our warts in our respective systems around the world. But truly, the benefits of a liberal democratic approach to governance and what we actually stand for, as much as we trip and stumble over ourselves as we attempt to adhere to those principles.
In many cases, a more proactive conversation about what is better for not just national security, but transnational human security is needed. One of the things I worry about in a forecasting sort of way, getting back to the China conversation — we’ve seen it all the more in recent weeks and years, is Xi Jinping doubling down on an autarkic approach to technology in the interest of self-reliance and protecting China from his clearly articulated definition of strategic industries they want to be the leader in.
Are we potentially moving into a bifurcated approach to technology and even science? One of the challenges we face is as the world is wrestling with issues for which we need cooperation, we’re at the same time contesting over the norms by which those issues are being explored. It’s about proactively driving a conversation about what is best for solving a problem, as opposed to just being seen as criticizing somebody to knock them down.
Jordan Schneider: I want to reflect more broadly about the role of intelligence. In the closing of his book Engineers of Victory, which looked at different operational campaigns over the course of World War Two, Paul Kennedy said initially he was going to have a whole chapter on how to win the intelligence war. But the more he looked at it, the more he saw over the course of World War Two that the study of intelligence over that period was like a whole lot of study of failure.
Broadly, does any of this even matter in the end? If national power is just downstream of economic growth and productivity growth and military capabilities, can a clever memo from an intelligence analyst really bend the arc of history?
Mike Collins: It’s not that easy if we just put out a formula and an algorithm to say what the future forecast of our economy will be relative to some other country.
Relationships and intellect do matter, and agency, frankly, matters. I’ve studied different theories of international relations from the most structural to the most constructivist, and the differences between what the structural forces will bring about inherently, as opposed to looking bottom up at what constitutes the agency of various actors and the decisions they make within the system.
One of the techniques we apply is the art of the alternative futures tool, whereby you begin with an outcome and then we try to talk about how you can get there. Classically in the intelligence community, we want to begin with the worst-case outcome.
But what I challenge the analysts to do is to pick an outcome where we win. The world trends in a way, or the issue trends in a way, that the United States and our partners come out on top and then work it back as diagnostically and precisely as possible for all the ways in which we could see this vector.
There are many unanticipated events that could happen. Weather events, natural disasters, tsunamis, are great case studies from what happened in the South Pacific years ago, not just the one in Japan. The impact our response to that actually had on our positioning in Southeast Asia for how the US military, for example, came to support that same be true in Japan.
We do these global trends forecasts, and we’re very honest and humble. We don’t have the magic formula, but we try to apply both. We apply what the structural data says as you apply them to the algorithm. But then you have to look back — what are those key decision points along the way? Where do individual leaders particularly matter?
Understanding Xi Jinping — he’s been arguably one of the most disruptive figures in what was happening in the US-China relationship and more broadly, of China’s place in the world.
He’s the one who said, forget about hiding capabilities and biding our time 韬光养晦. He said, now is our time to be 有所作为.
Leaders and when they come up can in fact matter.
Challenge ourselves, and certainly, be humble in the predictions we make. But at the same time, I don’t believe that it’s as easy to just say, well, look at the balance of economies would go forth. The relationships that exist among various countries and non-state actors in our countries increasingly matter. What individuals themselves think about these issues, as I said earlier, the individual will increasingly matter to a lot of issues going on in the world. As we talk about contestation in the gray zone and other areas.
Jordan Schneider: Let’s close maybe on a call to action or two. What are the biggest open analytical questions you see? What do you want to read more papers about? How can folks get in touch?
Mike Collins: Thanks for the advertisement. We, especially in the National Intelligence Council, are open for collaboration. That’s my opening pitch. We convey this through venues like this. We convey this through various addresses we have. We convey this to the External Research Council. I just described another venue we have where we’re offering up to help others.
We have a lot of talent in the United States and around the world who want to work in the intelligence business. We also have a lot of partners who work in the private sector who are studying something for some other reason that we may be able to help them as well in a collaborative way. As universities are training students who might want to get into the intelligence community business, we can help them to think through the programs they’re teaching and the things they’re studying that benefit us but also showcase our work out loud.
There are three areas, as I think about things that I particularly want to double down and think about where work needs to be done. Let me be as precise as possible:
We need to look at our geopolitical challengers, our geopolitical revisionists, trying to change the world in a different way than it currently is, in understanding them in greater detail than as a monolith. We should not just think about them as a unitary state and the leader of that state. Every country around the world has different views and pressures within them. At the end of the day, these heads of state are ultimately, in various forms, to be fair, politicians. They’re responsive to something at home that they’re probably most worried about and most focused on. We need to think more about the conversations that are happening between those leaders and their citizens and their populations in getting at potential areas of dissent and disagreement that may not show themselves as evidently as we expect in the form of, say, protest. When money moves, when people leave, when people are laying flat, that suggests some degree of disagreement or dissonance, at a minimum with what a state is providing.
Gray zone activity. This landscape is being determined. This contest is happening not necessarily as much as we warn about militaries fighting militaries, gray hull ships on gray hull ships, nor is it happening in that legitimate diplomatic, political, economic, above board, international institution governed arena. It’s happening increasingly in that area for which norms have not been established, be it in cyber, be it in foreign malign interference, technology contestation, and the private sector is in the middle of that. We have to appreciate that and study that all the more for all the reasons I said.
We in the intelligence community and with the collaboration with those on the outside, we’ve got to go beyond thinking about national security in its traditional, kinetic sense. We have to think about what determines the commercial and business success of entities in the international arena, and understand and appreciate that is the role of business and the role of commerce in international affairs. Because ultimately, if the companies of one ecosystem are thriving commercially and from a business standpoint, it’s not just from a profit standpoint, but increasingly, from a strategic standpoint, that is another area that we have to broaden our aperture and our appreciation for what national security really is.
Jordan Schneider: How should our readers reach out to ODNI if they want to chat?
Jordan Schneider: What do you think about the Cold War analogies for the situation between the US and China today?
Mike Collins: I don’t know if the entirety of the national security community has sort of come to grips with this issue in some ways, but there are parallels to understanding lessons from the Cold War confrontation when we speak to the normative and the ideological underpinnings of it and kind of what we’re facing now with China — not China itself, but the Communist Party of China. For what this leadership has been espousing and advocating for and in their own writings when they talk about what this China is trying to accomplish around the world.
The term [Cold War], broadly speaking, refers to a contestation, a challenge within which one country is using all elements of power to achieve something in a geopolitical sense at the end of the day without having to go to hot war. We see this in the descriptions and the narratives of what the leadership under Xi Jinping is trying to broadly accomplish in terms of China’s place in the world. I’m not saying they’re looking for a hot war. To the contrary, they’re trying to compete with us and challenge our standing without having to get there, but utilizing all purposes of power.
This is as much a competition over the norms, the values, the liberties we stand for relative to the norms and values that our authoritarian rivals stand for.
Today’s breakdown is authored by “Lithos Graphien,” an anonymous contributor with decades of experience in the lithography industry.
Printed Electronics and the Age of AI
65 years ago, Robert Noyce of Fairchild Semiconductor envisioned a way to make complex electronics using a printing process known as semiconductor lithography. Thus, the monolithic integrated circuit — or microchip — was born.
Not long after, Noyce co-founded Intel with Gordon Moore, who famously observed in 1965 that the number of transistors (electronic switches) on Intel’s microchips doubled every four years because of improvements in this printing process.
Moore’s Law holds to this day, with smaller and smaller circuit parts printed to pack more computational power into each new chip. In the near future, chips the size of your fingernail will contain an astronomical 100 billion transistors.
For this reason, a nation’s lithography capabilities determine the power of the semiconductors they can produce. That’s why lithography tools have been a key focus of US export controls.
Semiconductor Lithography Is at the Heart of the US-China Chip War
The US Department of Commerce — along with allies in the Netherlands and Japan — has so far issued two rounds of export controls aimed at limiting China’s lithographic capabilities. We summarized the first two rounds of export controls in this article.
Chinese companies like SMIC, however, have still managed to produce advanced semiconductors despite these controls. As another October approaches, which is Commerce’s favorite time of year to issue these updates, we can once again expect some type of action on lithography.
This week, confidential sources told Bloomberg that the Netherlands intends to let the export licenses for ASML’s scanner parts expire at the end of the year, bowing to pressure by US Commerce.
ASML is the leading provider of lithography tooling, and their most advanced scanners have already been banned from export to China. The problem, though, is that Commerce took a number of years to enact these restrictions, and a large fleet of advanced tooling is already installed in China.
The chart below summarizes the machines made by ASML that are currently allowed or banned from export to China.
The key performance metric that Commerce uses to determine the law is called overlay — the ability of the machine to overlay two circuit patterns. Overlay is a measure of alignment error, and with today’s chip parts printing features on the tens-of-nanometer scale, overlay makes the difference between printing a mature node chip (>28nm) or an advanced one (<28nm). Each new generation of ASML scanners made incremental improvements in overlay.
Another improvement metric is throughput — the number of wafers per hour that can be processed through the tool. This doesn’t enable a new chip technology, but it has a considerable impact on the value of a scanner. Commerce has thus far made restrictions only on overlap, not throughput.
So what will happen next year for a Chinese fab using these tools?
First, note that no Chinese company can run these tools in secret. Every tool is accounted for by ASML and monitored by the factory. It should be fairly straightforward for ASML to segregate the export of parts for the blocked tools and the allowed tools. Most of the spare parts will be common between the blocked and allowed tools, so knowing where the parts are going will be key for enforcing the law.
Second, ASML’s immersion scanners are known for their reliability. They can often run for months at a time without maintenance, and for a much longer period before requiring a spare part.
These banned tools will be operational well into 2025 and perhaps beyond. But on a long-enough timescale, each of these tools will eventually become an idle boat anchor without spare parts.
So Chinese companies may attempt to stockpile key parts for these tools over the next four months.
Third, there is some confusion regarding whether ASML can or will block the servicing of these tools. The Dutch issue licenses only for physical goods, not technical services, so there is technically no expiration date on equipment servicing. Even so, it’s possible that ASML will stop servicing the banned tools anyway, because it’s in the spirit of what Dutch Commerce intended.
Uncharted Territory
The Dutch government’s move — letting export licenses for ASML’s scanner parts expire — at the request of US Commerce is unprecedented, in that Chinese companies legally purchased the machines under the assumption that ASML would support them for the lifetime of the tool. ASML still supports scanners they made thirty years ago. Impacted Chinese chipmakers will definitely respond through any legal channels possible. But if that fails, what is their contingency plan? What will ASML do to preserve revenues from their lucrative business in China?
The Nvidia model
Nvidia has been the focus of US Commerce as well. To limit China’s AI capabilities, Nvidia’s GPUs are limited based on their performance specifications.
In response to each new Commerce rule, Nvidia designed a special chip for the Chinese market that met the throttle specifications. ASML can take a similar approach — ie. they can meet the requirements by throttling the overlay performance on a blocked tool inside China to the required value of 2.5nm SMO. That would be both in their interest and in the interest of Chinese chip companies.
How would scanner overlay throttling work?
Over the years, many of the overlay improvements in ASML’s scanners were driven by software. That could include a new correction algorithm or a way to take more measurements. For this reason, the most likely tool downgrade option will be a software update pushed out remotely from ASML’s factories. This accomplishes a few goals:
It’s a relatively cheap, fast way to keep China’s fleet of immersion scanners running and earning money. This will be important if the legal battles are drawn out over a long period of time.
In the event that Commerce relaxes some of the restrictions under a new administration, the performance can be updated to the new overlay targets to improve tool performance again.
The blocked tools have the added value of improved throughput — which is not under any restriction — to maximize tool value.
And the tool will retain its resale value to a customer outside of China. If SMIC wanted to sell them to TSMC, for example, a similar software update could return it to advanced performance.
What are the bigger implications for China?
These restrictions will make it harder for Chinese companies like SMIC to produce a 7nm or 5nm chip. Commerce’s intent to limit China at the 14nm node will likely be fully enforced. And without further policy intervention, China will likely accelerate production of mature chips, advancing toward becoming the world’s top producer.
ChinaTalk is a reader-supported publication. To receive new posts and support our work, consider becoming a free or paid subscriber.
Jeff Ding is a professor at GW and a leading US scholar on China and AI. He’s also been in the China tech newsletter trenches with me practically since Substack existed, putting out essential weekly translations and analysis on ChinAI since 2018.
In honor of the publication of his new book, Technology and the Rise of Great Powers, enjoy this interview with Jeff from we last year on its themes. I think it explores some of the most important topics to technology and power.
Staffers, students, think tankers, and entrepreneurs take note! If you want to make a difference on a multi-decade timeline in strategic competition, try to work on the topics that we get into in the conversation about how tech and government policy drive long term growth rates among great powers.
Jeff Ding argues in a 2023 paper which echoes his book that great powers must harness general-purpose technologies if they want to achieve global dominance. That is, diffusion capacity (not just innovation capacity) is critical to economic growth — and China actually fares much worse in diffusion capacity than mainstream narratives imply.
In this show, we discuss the historical underpinnings of that argument and apply it to AI today — drawing out policymaking lessons spanning centuries of technologically driven great power transitions. We also get into:
Why long-term productivity growth is driven by the diffusion of general-purpose technology, and what makes this so crucial for great power competition;
Historical lessons from the UK, Soviet Union, US, and Germany illustrating the cultural and policy roadblocks to tech diffusion;
The importance of decentralized systems, and how this helped America win the Cold War
Why China’s diffusion capacity lags behind its innovation capacity, and how America should avoid getting locked into any one technological trajectory.
Co-hosting is Teddy Collins, formerly of DeepMind and the White House Office of Science and Technology Policy.
Jordan Schneider: Why is long-term productivity growth all that matters for great powers?
Jeffrey Ding: Technology is going to affect a range of different things we care about in great power competition. It’s going to affect the military realm. It’s going to affect how nations perceive other nations — their prestige, their soft power.
I’m most focused on economic power. It’s the most fungible aspect of power. You can transform and convert economic strength into military strength. A country that sustains its development becomes more reputable and gains soft power because of its development model.
You can see that with China and the Chinese model of development. Economic power is the most fungible and transferable currency for measuring power. Historically, we’ve seen the rise and fall of great powers occur through a pretty regular pattern: one country sustains productivity growth and economic growth at higher levels than its rivals, becomes the preeminent economic power, and then converts that into geopolitical and military strength.
Jordan Schneider: There’s a big difference between a two-year horizon and a twenty-year horizon. At that timescale, why is productivity growth really what counts?
Jeffrey Ding: In the long run, productivity growth is what sustains economic growth. Take China as an example. Its economic growth to date has been driven by a range of factors, most notably demographic advantages — a large, young workforce willing to work for low wages — and urbanization. The transition from a rural, agriculture-based economy to a more urbanized, industrial economy brings a lot of good opportunities for growth.
But those two advantages are fading as China tries to escape the middle-income trap and become an advanced, high-income economy. Economists and development scholars have found that the way to escape that middle-income trap — to sustain economic growth in the long run — is through productivity growth and adopting new technologies.
Jordan Schneider: You argue that great power status is fundamentally based on wealth. So what drives national economic growth over the long term?
Jeffrey Ding: “Great powers” is a fuzzy concept. A great power is a country that has both a large economy and an advanced economy. Those two pieces working together are important. It’s not enough to just have a very populous country with a large economy if that economy is not efficient, because that’s what sustains growth in the long run.
The rise and fall of great powers happens because new technologies create differences in economic growth. One country can sustain growth for a longer period at a higher rate on the back of new technologies, whereas another power suffers a decline.
Jordan Schneider: The timeframe of the average policymaker is about one to five years. On a five-year basis, there are a lot of different policy decisions that can have near-term impacts on economic growth. You have fiscal policy. You can juke around with monetary policy.
But if you zoom out over ten to forty years, the changes you can make on the edge with fiscal and monetary policy end up balancing out. What remains is how you absorb and adapt to new technologies.
Jeffrey Ding: We often focus on just the initial moment of innovation and assume that will make an impact within five years. If you poll the top machine learning experts today and ask them when we’ll get high-level machine intelligence, a lot of them say it will be by 2035. Some commentators say AI will change the balance of power within the next decade. That’s our horizon.
But a lot of these revolutionary advances in the past have taken decades — often four or five decades — to diffuse throughout society and actually make the impact that we think they will make.
Jordan Schneider: From a policy perspective, focusing on a twenty- or thirty-year horizon is fundamentally different from a five-year horizon. The types of things you want to bet on and the policies that are robust to different futures are different when you’re looking so far out and the future is uncertain. What’s so interesting to me about your research is just how often this has played out over time.
Atlantic Waves: Industrial Revolutions in America and Europe
Jordan Schneider: Let’s look into some case studies to illustrate just how important technology is to long-term national greatness, or to nations being able to claim their place at the top of the global food chain. What role did technology play in helping the UK supplant its rivals in the eighteenth and nineteenth centuries during the First and Second Industrial Revolutions?
Jeffrey Ding: The UK was able to sustain productivity levels higher than their rivals because they adopted iron advances, metal-working advances, and mechanization at scale more effectively than their rivals.
Often in the case of the Industrial Revolution, we gravitate toward the fast-growing new industry of cotton textiles. My argument is that it was the more gradual, protracted diffusion of iron machine-making and mechanization throughout the entire British economy. That was the real driver of Britain’s industrializing faster than their rivals like the Netherlands and France.
Jordan Schneider: What are general-purpose technologies (GPTs), and why are they so important to the rise and fall of great powers?
Jeffrey Ding: General-purpose technologies have been deemed by economic historians as “engines of growth.”
General-purpose technologies are engines because they are fundamental advances that have the potential to transform broad swaths of the economy.
There’s a whole research agenda attached to them. There’s a lot of scope for continual improvement. Crucially, they make an impact only if there are a lot of complementary technologies across many different sectors that adapt.
Electricity is held up as a prototypical example of a general-purpose technology. Electricity is able to make an impact on productivity only if you change your factory layouts from being driven by a central steam engine and a system of shafts and belts. If you change that layout to something that’s more decentralized, each machine is driven by an individual electric generator.
Electricity was only made available to all these different applications with the rise of electric utilities, like central generating stations. This was affected by other complementary technologies, like steam turbines. These general-purpose technologies take a long time to diffuse throughout the economy. They rely on different complementary advances across a wide range of economic sectors to make their ultimate impact.
Jordan Schneider: If you’re looking on a multidecadal horizon, GPTs are going to drive productivity growth. It’s not cotton textiles. It’s not the one-offs that give you significant productivity growth in one particular segment of the economy. It’s the GPTs that do it for your whole nation.
What that requires is not necessarily having Thomas Edison invent the light bulb, but it’s all the hard work necessary to hook up the entire country to electricity. Factory layouts, industrial organization, and the creation of firms are all hard, but they end up affecting the entire economy. Getting those things right — more than whatever is the hot emerging technology of the day — is what’s going to keep your country in the lead over the long term.
Jeffrey Ding: On your point about talent, we often hold up this image of the heroic inventor, like James Watt. Britain became the leading industrial power because they had the James Watts of the world to invent the steam engine.
But when economic historians have dug deeper into Britain’s source of advantage, they’ve underscored Britain’s average level of technical literacy. They had just this higher level of average technical literacy among the early versions of mechanical engineers, machinists, and those who were able to deploy the new iron-based machines in all these different industries.
Jordan Schneider: Geniuses are nice but not necessarily essential. It’s the next level down of having the implementers that live in all the different corners of the economy. They see whatever it is that’s new and exciting and then bring it into their expertise.
But let’s jump forward to industrial revolution. How did this affect the Germany versus the US?
Jeffrey Ding: The Second Industrial Revolution was unfolding in Germany, the US, and the UK from around 1870 to 1914. The version of the story that I tell here is that the US became the preeminent economic power, not just in terms of size, but through the combination of size and overtaking Britain in economic efficiency, measured by either labor productivity or GDP per capita as a measure of the development level of the economy.
The key GPT general-purpose technology trajectory was the spread of interchangeable manufacturing. It wasn’t necessarily who was exporting the most advanced chemicals at that time (which political scientists suggest is the reason Germany became a challenger to Britain). Rather, it was the spread of interchangeable manufacturing methods — what became known as the American system of manufactures — across the making of sewing machines, bicycles, and all sorts of devices in all sorts of sectors of the economy. This led to the US gaining a productivity advantage.
Other countries were ahead in terms of inventive genius, or they had the research centers of excellence — the DeepMinds of the world at that time. The US was able to diffuse this interchangeable manufacturing method because it had stronger connections between the frontier institutions, entrepreneurs, and engineers. They were able to build up a more practice-oriented mechanical engineering discipline. Part of that is due to investments in land grant universities that built up strong mechanical engineering developments all across the country, not just in the elite universities.
General-Purpose Tech and Human Capital
Teddy Collins: It seems there are at least two different categories of this diffusion potential or diffusion capability.
One is the broad, basic literacy or “tacit knowledge” that exists throughout a population, as opposed to a handful of spiky inventors or entrepreneurs.
The second is a willingness to experiment with and implement new approaches to infrastructure. You have to redesign workflows for physical settings. That’s a big capital expense.
Those two things seem a bit different, and it seems like they would come with different policy implications.
Jeffrey Ding: That’s a fair distinction. I tend to focus on the human capital argument: how do you build a wider base of average engineers associated with general-purpose technology?
There are a lot of other factors that drive the pace and intensity of GPT adoption, including some of the things you mentioned, such as organizational restructuring and the level of vested interest or legacy institutions that exist in some countries. This is why some people say there’s a late-comer advantage. Countries that don’t have those legacy institutions are just better equipped to implement some of these structural reforms.
It’s also hard to measure and operationalize these things in terms of how many vested interests are in a particular industry. That might change across different industries. I’m comparing advanced economies that all would probably have some level of vested interests and established structures.
My preference is to look at things that cut across industries. This might be the average level of engineering talent associated with the general-purpose technology, the degree to which universities are linked and communicating effectively with industry entrepreneurs, or the strength of dissemination mechanisms for ideas about technology.
Jordan Schneider: There’s something societally disorienting about these general-purpose technologies. Your whole system and ethos need to be ready for that.
I want to read you a quote from Elting Morison’s classic work, Men, Machines, and Modern Times, which gives you a sense of what it was like for ironworkers to transition to making steel.
Morison argues that in 1857 everything was in place to transition from iron to steel. But there were all of these cultural, societal, and economic reasons not to take that leap. To explain the hesitance and the multidecadal lag that it took to adopt this new technology, Morison writes:
Would it not, by replacement of an old reagent, iron, with the new element of steel, replace also the customs, habits, procedures, and hierarchical arrangements upon which the security of life in the iron trade depended? The converter, in this context, looks less like a tool of commerce and more like some catapult leveled against a walled town.
People, institutions, and nations are that walled town. They’re looking at that catapult, and they will respond to it in very different ways.
Jeffrey Ding: Exactly. That is an example of transforming just one industry to adopt an innovation, steelmaking. Now multiply that for a general-purpose technology to any industry that a GPT would affect.
It gets even more complicated and magnified in terms of the structural changes needed to adopt GPTs. The softer stuff you mentioned — culture, status, people wanting to protect the skills they’ve already developed — all of these things come into play.
Hot Tech, Cold War: US-Soviet Competition
Jordan Schneider The Cold War is another great case for your argument. The Soviet Union had these awesome research scientists. The USSR was able to do stuff in the lab and across many fields which equaled or even exceeded what America could pull off. But when the technological grounds began to shift beneath the Soviet Union — when it moved past steel-driven development and into electronics — it didn’t adapt and diffuse the technologies nearly as effectively as the US did.
Jeffrey Ding: When we assess countries’ scientific and technological capabilities, we overweight innovation capacity. These are our typical indicators:
Who’s spending the most on R&D?
Who has the top scientists publishing the most?
Who’s getting the most cited patents out there in different fields?
These are the things that we gravitate toward — and we discount diffusion capacity once that groundbreaking paper has been published. Are those ideas being commercialized and spread across all these different industries after the advances come out of DeepMind? Are those then spreading to spreading from these frontier firms to the small and medium firms that are driving most of the productivity growth in the entire economy?
In the Soviet Union’s case, they performed so well on all these traditional measures of innovation capacity in terms of the most PhDs in STEM fields or the most spending on R&D. But in a 1969 CIA report I cite in my paper, their assessment was that the Soviet Union lacked these fast-acting, biological processes of diffusion.
The planned economy of the Soviet Union limited its ability for these new advances to permeate and spread throughout the entire economy. The Soviet Union was doing well in mission-oriented breakthroughs like Sputnik, but it was not an economy equipped to computerize at scale. That led to stagnant productivity growth and ultimately the collapse of the Soviet Union.
Jordan Schneider: A fun theme in your research is Americans freaking out about losing in the 1950s. Everyone was like, “The Soviets have more PhDs than we do! This is going to be terrible.”
No Illusions? Why Tech Diffusion Wasn’t So Big in Japan
Jordan Schneider: Then in the 1980s, the concern was that Japan would overtake the US. David Halberstam — author of The Best and Brightest and Breaks of the Game — wrote in 1983 that Japan’s industrial ascent was America’s most difficult challenge for the rest of the century and a “more intense competition than the previous political-military competition with the Soviet Union.”
There was a deep consensus within the American body politic that America was losing the technological future and long-term productivity race to Japan. What didn’t Japan get right?
Jeffrey Ding: This was a very real threat in the eyes of the US. Henry Kissinger wrote an op-ed in The Washington Post saying that Japan’s economic strength and rise in high-tech sectors would eventually convert into military power and threaten the US. A poll in the late 1980s found that more Americans were worried about Japan than the Soviet threat to US national security.
The trend that I see so clearly with all these historical examples is the US overhyping other countries’ scientific and technological capabilities. One reason we do that is because we don’t pay as much attention to diffusion capacity.
There is a case in my book manuscript about why Japan was not able to overtake the US. It got to about 90% of US productivity levels in terms of total factor productivity, but then it stalled in the 1990s. That’s due to a number of reasons, including fiscal and monetary policy. That’s all relevant here.
What I highlight is that Japan gained a lot of market share in a lot of these new, fast-growing industries like consumer electronics and key semiconductor components. Butit fell behind the US in terms of adopting computers at scale and overall computerization rates.
I highlight deficiencies in Japan’s ability to train a large number of software engineers. They built a lot of centers of excellence at certain universities, but they weren’t able to build a wider pool of institutions to train software engineers and fill in those talent gaps that held back the diffusion of computers throughout the entire economy.
Jordan Schneider: Some profound humility gets inculcated in you when you sit down and think about 2023 and what everyone agrees on. Things can end up radically different from whatever the consensus is.
Jeffrey Ding: Our takes are all shaped by who we’re talking to, the institutions we belong to, the ideas that are circling the rooms that we’re in, and the narratives of our times. But I think looking at historical examples forces you to get out of that a little bit, out of your little mini echo chamber, and understand that maybe we’re very wrong about the assumptions we have in our social groups and among all the people we’re reading and listening to. It’s an important dose of humility.
Jordan Schneider: When thinking about the example of the 1980s and 1990s, what were the ingredients to national policy that allowed the diffusion of the Information Age to happen with such dramatic success in the US?
Jeffrey Ding: One important factor behind all this is just access to a wider pool of software engineering talent. The US was tapping into so many immigrants who wanted to come to the US and study and work in these areas. Japan was relatively closed off in terms of bringing in foreign talent and even sending students out to other universities.
Secondly, a lot of times in the wake of these new GPTs, you almost have to have a new engineering discipline.
Mechanical engineering in response to mechanization,
Electrical engineering in response to electrification, and
Computer science in response to the computer.
The US university system was more decentralized, and they had the flexibility to adapt and build this new curriculum for training software engineers at scale. Japan’s system was more rigid and centralized.
Suggestions for AI Superpowers
Teddy Collins: If you were made the czar of all AI-related policy in the United States, are there specific things you would push? What will it take for AI to show up in the productivity statistics?
Jeffrey Ding: First, there’s investing in human capital. For me, that might look like widening the base of average AI engineers, people who are not necessarily training with cutting-edge models but can take an existing model and apply it to a particular scenario. Maybe they fine-tune models on a more specialized data set. Or maybe they take something that is already out there, open source, and apply it to their specific industry context. Training that talent might look like investing in community colleges and improving the capacity to train people in the general field of computer science.
Infrastructure is also in there — but for me, that’s just anything that would affect GPT diffusion, not necessarily driving cutting-edge innovation. How do you improve access to compute for a wide range of universities and even small and medium businesses?
That’s something that the national research cloud discussions have not considered much. They’ve focused more on getting high-end universities access to more computing resources and investments in institutions that encourage more technology transfer.
Some scholars advocate for voucher systems that incentivize small companies to learn and adopt new techniques from frontier firms. Subsidizing and encouraging that in some way is another step governments can take.
Jordan Schneider: One of the arguments you make is that when GPTs come online, you don’t see them in the productivity statistics until twenty years later because it takes a long time for people to wrap their heads around them.
But this time around, we already have a nationally integrated economy. We’ve figured out how to finance a lot of these institutions. We have a whole venture capital ecosystem, and everyone understands that there are enormous gains to be made from these technological innovations.
What I am worried about when it comes to diffusion is the potential policy roadblocks that could arise if change happens “too fast” and the body politic or some industries just reject it. You could end up with legislative roadblocks that make everyone worse off with lower productivity growth. The technological change which was going to come doesn’t happen. You have a poorer society because we’re not trying to make the best of these technologies that have so much positive potential.
Jeffrey Ding: This is relevant for AI because there are so many risks associated with AI systems, from misinformation to accidents. On a narrower technical dimension, there are things like misspecified reward functions that result in all these out-of-control behaviors.
There’s a case to be made for smart pragmatic regulation that can enable more sustainable development and diffusion of these GPTs. Another example is nuclear energy, which saw key accidents and safety issues derailing their adoption at scale.
Teddy Collins: I can imagine someone saying software is fundamentally different because it spreads much more easily. The marginal cost is basically zero. We’ve already seen that because we have a small number of platforms responsible for a huge proportion of software use.
Perhaps you only need a small number of people at a small number of companies who understand the cutting-edge tech and can implement something that quickly diffuses to almost everyone. Thus, broad tacit knowledge is less important.
Jeffrey Ding: Maybe my predictions about GPTs taking multiple decades to make their impact are a bit outdated in a software-based world.
One way to measure diffusion is by dating a GPT’s emergence — maybe it reaches 1% adoption in an early adopting sector. Let’s say all the big internet companies were the first to adopt AI for adjusting their search recommendation algorithms. That’s the starting date. When does it reach 50% adoption as the median across all possible adopting industries? That’s the timeframe for the specific diffusion capacity and timeline.
People have tried to measure that for electricity and information communications technologies. Nicholas Crafts has done some of this work. Other scholars have found there has been a slight decrease in diffusion time, but it’s still on the order of multiple decades.
We’ve had breakthroughs in AI now for almost a decade, with the advance of deep learning about a decade ago. The Census Bureau in 2018 asked companies across all these different sectors about the extent to which they had tested and trialed machine learning systems. The percent adoption rate was still below 3%.
There are different takes on this. Some people rightly adopt an inside view. They know a lot more about AI than me. They’re tracking everything that’s happening in AI daily. It’s exciting. They’re like, “This thing’s going to diffuse fast — let’s be ready for it. We should be thinking on a timescale within this decade.” There’s value in considering what the external view says from these past examples and historical lessons. I’m not saying one is dogmatically right, but we should have a mix. I think the balance is tilted too much toward the inside view right now, and we should have a better balance incorporating these historical insights.
The Long, Twilight Diffusion: US-China Tech Competition
Jordan Schneider: China will likely have some real challenges when it comes to getting AI diffusion right? Even if they develop roughly equal powerful base algorithms like GPT-4?
Jeffrey Ding: There’s a gap in how we measure diffusion capacity for different countries. I just went over that metric of the 1% to 50% median. That’s hard work. It’s much easier to just cite anecdotal stuff or pull a number about R&D and make broad claims.
What I did in the diffusion deficit paper is look at different global indexes on science and technology. There are hundreds of different science and technology indicators. I tried to sort the ones that trended more toward innovation capacity: top-three firms in R&D investment and top-three universities in scientific research — these are more closely tied to innovation capacity.
On the other hand, you have things like:
How fast and to what extent have information communications technologies diffused throughout the economy?
What is the adoption rate of cloud computing in the country?
How strong are the linkages between universities and companies for spreading new ideas and collaborating?
Based on that exercise, I found China’s diffusion capacity lags far behind its innovation capacity. On all the indicators that are more tied to diffusion capacity, if you average them, China’s global ranking is almost thirty places — maybe more than thirty-five places — lower than its innovation capacity ranking.
Jordan Schneider: When thinking about systems competition, I often get frustrated by American policymaking. It is so diffuse. You need to make all the different representatives happy and throw everyone a bone. You have all these irrational policies, and this state is doing something that’s going against the grain of that state.
But there is something powerful about the way America is set up: our messy political system ends up not concentrating its bets too tightly.
It is awesome that we have an OpenAI and a DeepMind and all the best universities. But something that makes America special and weird — something baked into a lot of the American system — is that it’s not a centrally planned thing.
Jeffrey Ding: Decentralization is strongly tied to high levels of diffusion capacity. That’s been borne out by a lot of different econometric research and empirical work. There are things the US can be doing. I’ve testifiedtwice in front of the US-China Economic and Security Review Commission. My first recommendation both times was that the status quo is a defensible option. The US is in a good position right now because of its decentralized science and technology system. I strongly believe that.
It’s important the US not lock into a specific technological trajectory with AI. If we were having this podcast two years ago, we might be talking more about computer vision. Now the hottest subfield is natural language processing. If we were having this podcast eighteen years ago — when Clinton announced the National Nanotechnology Initiative — we would be talking about that technology. No one talks about nanotechnology anymore, even though it might have some general-purpose technology characteristics. It’s just diffusing under the radar.
Jordan Schneider: I love this quote you have from 2003 with the Under Secretary of Technology at Commerce saying, “Nano’s potential rises to near biblical proportions.” I mean, maybe — but if America put all our eggs in that basket and wasn’t paying computer scientists to figure out AI models, then we might be at a very different place.
Jeffrey Ding: Pouring a lot of compute and resources into transformer models seems to be one bet. There are other fields of AI beyond deep learning, like reinforcement learning or causality-based thinking — models we should not ignore nor neglect in the long term.
We should not do what Japan did when they invested in their fifth-generation computing project, narrowly looking at computers as huge mainframes in a world where that was increasingly turning toward personal computing. We should avoid locking in one technological trajectory.
Jordan Schneider: How defensible is it to keep the brains or keep the IP within your national borders? Is that something that works for these general-purpose technologies?
Jeffrey Ding: No. How do you keep such a foundational technology locked up? It’s not like GPTs are pharmaceutical secrets. That’s not the model. We’re not talking about profits from one country monopolizing this super-secret innovation. It’s more about adopting these innovations at scale.
Paid subscribers get advanced access to the second part of our conversation. We discuss:
The state of play in the race to attract talent;
Why hyping China’s AI and tech prowess could lead to threat inflation;
How translating Chinese sources helped Ding understand tech diffusion;
Why war is a tempting choice for lagging great powers.
Tech Talent Tug of War
Jeffrey Ding: China’s efforts to attract talent back from overseas have helped them stay abreast of the AI research frontier and stay connected to different innovation networks. That helps facilitate the initial adoption.
But after those top Chinese scientists hear about the latest advances and learn the latest breakthroughs, how does that diffuse to the next level down and throughout the entire country? It’s hard to have indicators for that diffusion capacity in AI.
One approach I use in the book manuscript is by asking, “How many universities does a country have that meet some baseline level of quality for AI engineering education?” I look at the CS rankings database and try to figure out how many universities in China have at least one researcher who has published in the top three AI conferences.
That number is relatively low compared to the US. It’s about 100 for China, whereas the US has around 400. There’s a broader pool of US universities that meet that low bar for baseline quality in AI engineering education.
Jordan Schneider: What’s the case for allowing the free flow of talent between the US and China?
Jeffrey Ding: A lot of talent is flowing to the US and staying in the US. The flow of talent back to China is obviously helping China’s diffusion capacity. But relatively speaking, it’s arguably helping the US more.
There are a lot of other reasons why we’d want to keep those flows open, just because there are a lot of advantages to having an open economy — not discriminating on the basis of geographic origin, or nationality — when it comes to shaping immigration policy.
These things matter more than that specific policy’s effect on relative diffusion capacity levels.
Jordan Schneider: You just reached your fifth anniversary of writing a truly fantastic weekly translation roundup on Substack. What has reading all of this contemporary Chinese writing on AI given you as a researcher?
Jeffrey Ding: History forces us out of our echo chambers and preexisting biases and dispositions. Just reading what Chinese people are thinking about AI forces me to get out of the DC and academic bubble when it comes to thinking about AI and US-China competition. It’s reading a completely different set of ideas from people living in a completely different context, whether that’s blogs or government white papers.
The diffusion deficit paper we’ve talked about in detail here was very much inspired by ChinAI translations about companies trying to implement computer vision to improve machine quality inspection on production lines for cutting tools and knives. That’s one of my favorite translations that I’ve done. It speaks to these companies. These people are not the ones that make the news, like the company SenseTime, one of the most valuable AI startups in the world. They show how China trying to implement AI on a granular level. What I’ve talked about today has been inspired by and builds on those weekly translations.
Teddy Collins: There’s a cliche that China can copy, scale, and commercialize, but it cannot innovate. It’s interesting to see that when you dig down into the weeds, China’s diffusion capability is worse than its innovation capability and the US’s lead on diffusion is greater. A lot of people would find that result surprising.
Jeffrey Ding: It’s often informed by a few attractive examples. China is good at diffusing certain things, like high-speed rail or e-commerce (like food delivery apps). But when you look at some of these other innovations connected to general-purpose technology — information communications technologies, computers, cloud computing — those diffusion rates are pretty slow in terms of actually affecting productivity in a lot of different businesses.
Inclement Clouds on the Tech Horizon
Jordan Schneider: There are pluses and minuses that come with thinking China is going to take over the world versus a more realistic understanding of the country’s strengths and weaknesses in productivity and technology. You’ve got all this McCarthyism Cold War stuff about overshooting the gap.
There’s also the CHIPS and Science Act, the most aggressive thing that America has done in twenty years trying to pull some levers in manufacturing and science and technology investments. Maybe it’s not exactly what you would have done from a diffusion perspective, but that doesn’t happen unless you have politicians really worried about China’s rise as a technological power.
Jeffrey Ding: There are instrumental reasons to overstate and hype up China’s AI capabilities and its scientific and technological prowess. If I were someone who really believed in the CHIPS and Science Act as the most essential thing for US national security, the only way to get that across the board is to adopt a “China is going to overtake us” framing. There are reasons to do that and I see the rationale behind that.
There are also a lot of downsides, however, to overestimating someone as a threat. It could lead to more threat inflation. It could lead to more willingness to escalate conflicts.
We’ve seen that historically in the US-Soviet Union case — the illusory missile gap or the US and Soviet Union getting to the brink of world destruction many, many times. The truth matters too. Having a more accurate depiction of the scenario is a good thing.
Jordan Schneider: It’s interesting reflecting on this if you’re inside the head of Xi or some other Chinese policymaker. Everything I read about China and AI seems hyperbolic.
The self-flagellation of the Chinese internet over the past six months and watching ChatGPT explode in the West has been really interesting. All of a sudden, the discourse went from, “We’re going to be awesome and amazing” to, “We’re so pathetic as a country, and we need to get our act together for this long struggle.”
I can paint a positive upside for the dose of realism injected into Chinese discourse and what that’s doing for Chinese policymaking. But Beijing might ultimately believe China will be on the back foot technologically for years to come, which could drive a more competitive dynamic with the US. The odds of us coming out ahead on that are probably way worse than the expectations that are currently baked in.
Jeffrey Ding: We have shifted so far in the direction of US national security interests and the need to beat China in all these different forms of competition. The biggest risk is if China overtakes us on something, whether militarily, economically, or by soft power.
I’m not sure where I stand on this, but why are we not considering that the biggest national security risk for the US is a weak China and a China that can’t sustain its growth? For the longest time that was US State Department policy. A strong China is good for peace.
All this self-flagellation that’s been coming out in terms of China’s AI sector has been overhyped. China could also suffer economic stagnation. What would the national security consequences be for the US? They might not be good.
It’s not even in the Overton window. We’re not even talking about it anymore in Washington.
Jordan Schneider: That’s a tricky Goldilocks thing. The way I see this getting turned around is China slows down and America and all its allies keep growing nicely with their adoption of GPTs and whatnot.
Whoever comes after Xi might realize that, in fact, you need to be globally integrated and have happy diplomatic relations to have a place on the national stage and be respected and keep pace with other countries.
But I’ve done a lot of interviews about this idea of “temporalclaustrophobia,” where the Kaiser, Imperial Japan, and Nazi Germany all convinced themselves they were at a high watermark.
One of the ways to not play a multidecadal game is to start a war. If you roll the dice now in a really aggressive way, then massive wars are one way to short-circuit that long-term productivity contest.
You can get lucky or overperform in a narrower window because you have some edge or another. Though, most of the time, the countries that win wars are the ones that are larger and more technologically advanced. But it’s not every time. War can be your “out” if you talk yourself into believing there’s a hopeless long-term trajectory for you as a national leader.
Jeffrey Ding: That squares with my understanding as well.
Jordan Schneider: I feel like I’m ending every ChinaTalk on World War Three, which is a bummer…
ChinaTalk is a reader-supported publication. To receive new posts and support our work, consider becoming a free or paid subscriber.
Is AI espionage preventable? Are open-source AI models a threat to national security? To discuss divergent industry viewpoints, we have a special guest post by Pablo Chavez, former ChinaTalk guest, fellow at CNAS, and former VP of Google Cloud’s Public Policy.
In late July, Sam Altman and Mark Zuckerberg each wrote independently about how the United States should deploy and govern AI power. Read together, these two pieces represent a high-stakes dialogue on the geopolitics of AI, portraying sometimes competing, sometimes complementary visions of American AI leadership.
Altman advocates for a more controlled and regulated approach, while Zuckerberg champions the power of open-source collaboration.
[W]e face a strategic choice about what kind of world we are going to live in: Will it be one in which the United States and allied nations advance a global AI that spreads the technology’s benefits and opens access to it, or an authoritarian one, in which nations or movements that don’t share our values use AI to cement and expand their power? There is no third option — and it’s time to decide which path to take. … That will … mean setting out rules of the road for what sorts of chips, AI training data and other code — some of which is so sensitive that it may need to remain in the United States — can be housed in the data centers that countries around the world are racing to build to localize AI information.
The United States’ advantage is decentralized and open innovation. Some people argue that we must close our models to prevent China from gaining access to them, but my view is that this will not work and will only disadvantage the US and its allies. … I think our best strategy is to build a robust open ecosystem and have our leading companies work closely with our government and allies to ensure they can best take advantage of the latest advances and achieve a sustainable first-mover advantage over the long term.
The two visions diverge for some reasonably straightforward business reasons. Altman’s OpenAI is, at its core, a developer of AI systems that it provides to enterprises and consumers for a fee. As a consequence, protecting models — the company’s main source of revenue — is a necessary step on the path to profitability. By contrast, Meta is mainly an AI deployer that wants to leverage the technology for its products (just like other enterprises like Microsoft that integrate AI into their products). It doesn’t want to be locked into a particular supplier or live in a world where AI model suppliers become so successful that they threaten its advertising and other core businesses (as OpenAI threatens to do to Google in search, for example).
Sama interviewing Zuck in 2016. Zuck said, “I get frustrated about when people fearmonger about AI and how it could hurt people — because around diseases and self-driving cars, this is going to save people's lives and push people forward.”
Beyond their respective business interests, the two CEOs explore the larger strategic implications of the paths they propose, revealing some nuanced common ground along with some clear distinctions.
Two Competing Visions
At a high level, Altman is calling for a disciplined US-led industrial policy effort, embracing cooperation with like-minded democracies, and emphasizing the importance of coordinated action. His focus on security, infrastructure investment, and international norms leads to a controlled but generous release of AI to allies and partners — as well as co-development with these partner nations — while making significant efforts to keep frontier AI out of the hands of China and other autocratic rivals.
Conversely, Zuckerberg champions an organic, hands-off approach to democratizing AI. His call for an open-source AI ecosystem echoes a more inclusive and collaborative ethos, potentially fostering a dynamic AI landscape that empowers a broader range of actors. Like Altman, he’s concerned about China, but he sees openness as a means to stay ahead. He argues this is the only option to ensure AI technology remains broadly distributed; the alternative he describes is where the technology becomes concentrated in the hands of a select few.
While Altman’s goal is to develop and deploy AI that aligns with and upholds democratic values, Zuckerberg’s emphasis is on democratizing the development and deployment of AI itself.
These are two very different goals with potentially divergent outcomes.
In Altman’s world, both the coalition of AI countries and the technology itself should have democratic characteristics. He argues AI ought to be deployed beyond allies in the service of growing the global democratic coalition.
By contrast, Zuckerberg writes about democratizing access to AI itself through open source. In this vision, AI is water, food, healthcare, and education. With it, all countries will do better. Without it, some countries will fall behind to the detriment of all of humanity.
Ultimately, Zuckerberg argues, an open AI ecosystem should lead to a more open, safer (and perhaps more democratic) world. He points to the history of open-source software as a model for testing and ensuring the safety and stability of code.
Zuckerberg also focuses on protecting what he believes is at the core of America’s technological advantage: decentralized and open innovation. He’s not just with staying ahead in AI, but also with protecting the ecosystem that generates American technological advancement. Altman focuses on strategies for winning the AI race, rather than preserving the economic and political operating system that got the U.S. to where it is today. The advantages of AI — such as workforce training and building infrastructure — are questions for the future.
A deeper reading of their essays softens some of the contrast between the two visions.
Both Altman and Zuckerberg express concerns about the concentration of AI power. Altman worries about authoritarian regimes using AI to strengthen and broaden their control, while Zuckerberg is additionally wary of closed AI models controlled by a small number of companies.
Their disagreement over open vs. closed is a bit grayer rather than a strict dichotomy.
Altman’s coalition does not exclude open-source collaboration. Instead, it seeks to create a strong, unified front, built around closed, controlled models to promote democratic values and prevent authoritarian dominance.
He views open models as an ancillary mechanism of soft power to encourage both partnerships and self-sufficiency among third-party countries.
For its part, Zuckerberg’s open-source AI future doesn’t necessarily contemplate open model development — just open release, without specifics about how open any particular model should be.
In addition, Zuckerberg emphasizes the advantage that larger, more sophisticated institutions will have in deploying AI at scale: such institutions have more compute, and therefore an incumbent advantage over smaller players. He suggests that these larger institutions will have an interest in safety and stability and argues that America’s leading AI companies should work with the US government and allies to maintain a first-mover advantage over bad actors.
None of this is incompatible with openness, but these details are evocative of a more closed ecosystem than the top line suggests.
Similarly, while differing in emphasis, their perspectives on AI safety are not mutually exclusive. Altman’s focus on establishing international norms and protocols complements Zuckerberg’s belief in the inherent safety of transparent, scrutinizable AI. Both recognize the need for a multilayered approach to AI safety, combining technical safeguards with broader ethical and governance frameworks. Indeed, Altman’s advocacy for a multistakeholder governance model is clearly inspired by ICANN, but it’s also evocative of open-source software development communities.
On Engagement with China
Altman believes the threat of authoritarianism should be addressed by withholding technology from China. At the same time, he calls for engagement with China to cooperate on reducing catastrophic risk — a pragmatic nod to a complex geopolitical reality. In a sense, his argument is that we have no choice but to work with them given their size and influence. Zuckerberg offers an alternative method for maintaining an edge over China: fostering America’s innovation ecosystem through openness.
Fundamentally, both see China as a threat, but Altman thinks the US and its allies can still protect AI infrastructure from Chinese cyber-intrusions. Conversely, Zuckerberg assumes that Chinese actors are already in the system, and thus, the goal should be to continue to move fast and stay one step ahead.
Neither essay discusses China’s AI ecosystem — including what appears to be a fairly robust and growing open-source AI community — or how China would respond to their respective visions.
Everything in Moderation?
The dialogue between Altman and Zuckerberg underscores the complex challenges and opportunities at stake in AI advancement.
Perhaps the future of AI will likely be shaped by a combination of both approaches. Ultimately, the most successful path forward will require a delicate balance of innovation, accessibility, safety, and ethical considerations that involve a diverse set of governments, corporations, and civil society actors.
The question of who will control the future of AI remains open, but one thing is certain: the decisions we make today will have profound implications for generations to come. This must remain top of mind for American political leaders as we transition to a new administration and a new Congress in 2025.
This showdown deserved its own AI-generated track…
For more Pablo Chavez on open vs. closed AI, have a listen to the show we did together last month.
Good Job Alert
The Institute for Progress is looking for an associate editor. Your boss will be , creator of the excellent Statecraft newsletter and the think tank editor I’ve worked with who has impressed me the most. Consider applying!
ChinaTalk is a reader-supported publication. To receive new posts and support our work, consider becoming a free or paid subscriber.
To discuss the domestic and international implications of patent policy, ChinaTalk interviewed Brian Pomper. Brian was the Chief International Trade Counsel to Senate Finance Committee Chairman Max Baucus and is now a partner at Akin Gump.
We discuss:
The history of America’s innovation hegemony, from the signing of the Constitution to patent trolls and Elon Musk
Why big tech companies spent decades systematically attacking the foundations of the US patent system
The thermonuclear patent war of Apple vs Samsung
The evolution of Standard Essential Patents (SEPs) as a battleground for emerging tech competition
Why China’s approach to patent litigation is causing controversy in Europe
The intersection of patent policy and international trade agreements.
Thanks to The Innovation Alliance for sponsoring this episode. The Innovation Alliance is a coalition of research and development-based technology companies representing innovators, patent owners, and stakeholders who believe in the critical importance of maintaining a strong patent system that supports innovative enterprises of all sizes.
A Brief History of Patents in America
Jordan Schneider: Intellectual property as a concept can be traced back to the Greeks— someone in 500BC patented a food dish! Then we get our first real patents in Renaissance Italy. But fast forwarding a bit, Brian why don’t you kick us off with the Constitution?
Brian Pomper: In the original text, aside from the Bill of Rights, the only right explicitly included is that of authors and inventors to exclusive rights to their inventions.
“Congress shall have the power to promote the progress of science and useful arts, by securing for limited times to authors and inventors the exclusive right to their respective writings and discoveries.”
~ Article I, Section 8 of the US Constitution
I’ll share my own history with patents as well. My position in Congress was Chief International Trade Counsel to the Senate Finance Committee, but my experience with intellectual property goes back further.
I have a degree in mechanical engineering, of all things, and I am a member of the US Patent Bar. I took the bar exam because I was trying to avoid work for a week at my private sector law firm job.
That was a failed strategy on my part, but it left me in the unique role of a trade lobbyist with patent bar membership and an engineering background. That’s why I approach patent policy from the standpoint of geostrategic competition with China.
Jordan Schneider: Why did the founders care about this?
Brian Pomper: In England, only those who manufactured a product could have rights over their invention. It was almost like a reward system from the government — a benefit given to favorites of the court.
In America, the idea was to democratize the process. The thinking was that there was significant inventive creativity among the American populace, not just among rich people with government connections or the means to build factories.
One of the first pieces of legislation Congress passed was the Patent Act of 1790. I strongly believe that the patent system has been a foundational component in setting the American economy on the right course and developing a culture of unique, innovative creativity that we still enjoy today.
Jordan Schneider: Can you take us through the 19th century? How were patent rights respected or not respected during this period?
Brian Pomper: I’d refer to the work of my friend, Professor Adam Mossoff from George Mason University. He’s done extensive historical research on this topic. When policymakers are told about “patent trolls” taking over the American economy, Adam points out that the 19th century actually saw a lot of patent litigation activity.
Many famous inventors like Alexander Graham Bell had to litigate to secure their exclusive rights. Thomas Edison and the Wright brothers also engaged in similar practices. However, no one at the time called them patent trolls. That term is really an invention of late 20th-century America.
This brings us to an important inflection point in how we view patents. Around 1982, a commission looking at American competitiveness suggested creating a unified patent court, which led to the establishment of the Federal Circuit. The idea was to ensure uniformity in patent law interpretation across the United States, rather than having different geographical interpretations.
This focus on strengthening patent rights and ensuring predictability contrasts sharply with the current climate.
For at least the last 20 to 30 years, there’s been a sustained attack on the patent system, driven by two main factors: high-technology goods and pharmaceutical products.
Jordan Schneider: Let’s move into the 1990s and 2000s. What are some big legislative landmarks that changed how patents work in America?
Brian Pomper: The high point of support for the intellectual property system in the United States and globally was when the World Trade Organization came into existence in 1995. The Agreement on Trade-Related Aspects of Intellectual Property Rights (TRIPS) provided common standards for how countries around the world needed to treat intellectual property.
This US government initiative aimed to set a foundational floor for intellectual property enforcement in every country. The idea was that the United States, with its highly inventive economy and companies with large patent portfolios, would benefit from instituting a rule of law approach to intellectual property globally.
However, there was a backlash to the TRIPS agreement from some developing countries who felt it was too one-sided, benefiting developed countries more than those still climbing the development ladder.
Two other elements led to the erosion of support for the intellectual property system, specifically the patent system, in the United States.
First, concerns about rising drug costs led some politicians to point to patent rights as a reason for expensive medicines.
But even more importantly, the late 1990s and 2000s saw the rise of gigantic technology companies. We all know who they are — they are the largest companies the world has ever produced. These companies — while highly innovative and extraordinarily successful — aren’t innovative or successful because of reliance on their patents. Some are essentially advertising companies, while others are effectively online retail companies. Many package other people’s inventions into products they sell.
What all these companies had in common was that patents were a cost for them because they were selling products, not technologies per se. These products often incorporated many other people’s technologies.
For instance, consider a desktop computer. That has numerous technologies implicating many patents. The chances are not insignificant that you’re using somebody else’s technology without permission or a license. This situation ultimately results in a request for a license and perhaps litigation if there’s no agreement.
The rise of the high-tech industry and the business models of these companies led to a circumstance where it was in their economic interest to challenge the foundations of the patent system. This system should provide stability and predictability for patent owners. If you’re a patent owner, you want to know that your right is durable. This allows you to raise money and promise investors that this technology is yours and only you can use it. You can then create a company that implements that technology.
If a patent can be easily challenged, dragged into court, or brought before an administrative tribunal — that creates uncertainty for years. It hinders the inventor’s ability to attract investment and develop the technology as they should.
Speaking as someone who cares about patent policy, I believe this is negative for the U.S. economy. The U.S. economy should welcome inventors and ensure their rights are stable and secure so they can attract investment and become the next very large companies in the United States.
Jordan Schneider: Let’s point to one counterexample. Elon Musk — famously not a big patent advocate — has said in the past that SpaceX basically patents nothing. His argument is, “patents are for the weak,” and anything they patent regarding their rockets would just be something that Chinese competitors could copy and paste.
Clearly, there is significant innovation behind a project like Starship. Writing off companies that make products as “not technologically forward” might be pushing it too far.
However, there are clearly different ways to approach the cost-benefit calculus of patent portfolios. If Elon Musk held more patents and aggressively enforced them, would that hurt small firms or firms focused on R&D instead of productization?
Brian Pomper: First, I want to emphasize that I did not say these companies were not technology-forward. They’re highly innovative and successful companies. They just don’t use their patents to monetize or secure investment in the way that many inventors do. Different companies have different approaches.
When Elon Musk says he won’t patent his inventions because then the Chinese can copy the technology, he’s referencing the fact that patents are public documents. The patent trade-off is that you get exclusive use of that innovation for 20 years from the filing date, but you have to make your idea public. The idea is that if you make it public, others can learn from that technology and build upon it.
If Musk is saying he won’t get a patent because he’s afraid others will steal the technology, he’s really relying on trade secrets. From the standpoint of what’s best for the American economy and innovation, relying more on trade secrets than patents is negative. Trade secrets don’t teach the rest of the public what your technology is, making it difficult for people to build upon it.
Regardless, having a patent is crucial for attracting investment. Ask any venture capitalist what they inquire about when considering investing in a company. One of the first questions is whether the technology is patented. This question is common on shows like Shark Tank for a good reason. Investors need to feel assured that if they invest in Company A, Company B won’t come along and make the same exact product to compete with them.
Patents are undeniably useful for the American economy, but perhaps not for every business model — as Elon Musk has demonstrated.
Jordan Schneider: Let’s go back to the policy wars that have been fought over intellectual property in the 21st century. What have been some landmarks in the 2000s that have changed the terms of debate?
Brian Pomper: Some of the big high-tech companies have fought long and hard to change the patent system to suit their particular business model. If you’re a small company going up against a very large, well-resourced company, it’s difficult to negotiate as equals unless you have the exclusive right to a particular technology that the large company wants or needs to use. Patents level the playing field in negotiations.
This is an important aspect of our system that should encourage innovative challengers to technology incumbents. It’s how we grow and develop. Of course, big tech companies don’t want challengers; they want to preserve their markets and monopolies.
For the last 20 years, these companies have pushed hard for changes to the U.S. patent system, some of which they’ve achieved legislatively.
In 2011, Congress passed the America Invents Act, which created a process at the US Patent and Trademark Office for administrative challenges to patents. If the patent office made a mistake in issuing a patent, then they should be able to revoke it — that was the idea. At the time, the intent was to create a cheaper, quicker alternative to patent litigation for dealing with mistakes made by the Patent Office.
In practice, this has provided a tool for well-resourced accused infringers to go after patent holders.
They can file multiple challenges at the patent office on various grounds, lasting years and grinding down smaller companies that lack the resources to defend against these challenges.
This has created an imbalance, and there are now legislative efforts to bring more fairness to the process and make it more like what the original drafters intended.
While these companies succeeded in getting the America Invents Act passed in 2011, they were unsuccessful in achieving many other changes through legislation. However, they were successful in getting these changes through the courts. Their public relations campaign tarnished the patent system as a playground for abusers and patent trolls, which I believe was always overblown.
This narrative had a profound impact on judges, leading them to feel responsible for providing relief to those being challenged.
One important example is the 2006 Supreme Court decision in eBay v. MercExchange, which made it very difficult for patent owners to get injunctions against further infringement. This means that even if you prove your patent is valid and I’m infringing on it, the court may not allow you to stop me from using your technology. You’re left trying to negotiate how much I’ll pay you to use the patent, even though you can’t stop me from using it.
Thermonuclear Patent Wars
Jordan Schneider: Let’s discuss the most famous patent fight of recent vintage: Apple versus Samsung. How did the post-eBay v. MercExchange landscape affect that famous dispute?
Brian Pomper: The fight between Apple and Samsung was the patent-world equivalent of a global thermonuclear war. It played out in courts everywhere, but most prominently at the International Trade Commission (ITC), where Samsung sought to block the importation of certain iPhones into the United States, alleging they infringed Samsung’s patents.
This case is crucial because the patents Samsung accused Apple of infringing were standard-essential patents (SEPs).
SEPs are a unique type of patent. They are incorporated into industry-wide standards that all companies must use. For instance, every cell phone in the world uses the same wireless standardization.
When a company’s patented technology is incorporated into a standard, it comes with certain responsibilities.
The patent owner must agree to license their patent on fair, reasonable, and non-discriminatory (FRAND) terms to anyone who wants to use it. This responsibility exists because if every company selling a cell phone must use a standard for interoperability, they’re inevitably using the patented technology and must pay for its use. Without FRAND terms, the patent owner could charge extortionate amounts, knowing companies have no choice but to use the standard.
In the Apple-Samsung case, Samsung alleged it had offered to license Apple the patent on FRAND terms, which Apple refused. Conversely, Apple argued Samsung never offered FRAND terms. Ultimately, the U.S. Trade Representative decided there was insufficient evidence to prove Samsung had offered FRAND terms, so they didn’t allow the ITC to block Apple’s phones from being imported into the United States.
Apple v. Samsung courtroom sketch. Samsung attorney Charlie Verhoeven cross-examines Susan Kare, designer of the original Macintosh icons, before Judge Lucy Koh. Source.
This decision called into question the enforceability of SEPs, which are prevalent in various technologies worldwide, especially cell phones. It raised concerns about whether companies would want to contribute their technology to standards if they couldn’t enforce their patents. This controversy continues to this day.
Jordan Schneider: Let’s take a step back and discuss SEPs. Long-time ChinaTalk listeners may remember a fun show from two years ago that covered the 200-year history of international standards [it’s a husband-and-wife team who wrote the book together—adorable!].
SEPs are the reason we have technologies like 5G. Companies worldwide contribute their engineering expertise to solve portions of the problem, then negotiate what percentage of work they should be credited for if their technology becomes part of the standard.
Once everyone agrees on what 5G looks like, companies are paid a percentage based on their contributions. This system allows us to have 5G phones that work anywhere on the planet. It’s a remarkable example of global cooperation. How has the SEP landscape evolved over the past few decades, aside from the Samsung story?
Brian Pomper: The primary question regarding SEPs is how they can be enforced. Specifically, can SEP owners obtain injunctions against infringers or exclusion orders at the ITC? This is the crux of the argument. Some argue that SEP owners should never be able to enforce their patents because they have significant market power due to their patents being included in mandatory standards.
Others, myself included, believe that the negotiation process we’ve established, where the SEP owner must offer to license the patent on FRAND terms to a willing licensee, has worked effectively. This system has been in place since its inception and has proven successful.
Those who argue that the SEP system is broken and that these patents shouldn’t be enforceable through injunctions or exclusionary relief due to anti-competitive impacts overlook a crucial point. There are currently 8 billion humans on Earth and approximately 10 billion cell phones, each of which will be replaced in the next two years. If there were anti-competitive effects in the SEP system, we would see restricted product availability and difficulty obtaining goods. However, that’s not the case with cell phones.
The system has worked daily to create the profusion of technology we see. While there are occasional instances of intransigent companies on either side, the litigation system exists to resolve these issues.
Jordan Schneider: What should patent owners do if they feel their SEP is being violated?
Brian Pomper: If you’re an SEP owner and someone is using the standard (and therefore your patent), you approach them and say, “I see you’re using this standard. I have an SEP incorporated into it, and you need to license it from me. I’m offering to license the patent on FRAND terms, in accordance with the rules of the standard-setting organization."
The other party might respond, “I don’t think that’s a FRAND offer. You’re asking for too much.” If negotiations fail, you can sue them. In court or at the ITC, you would argue that you offered to license your SEP on FRAND terms, but the other party is an unwilling licensee. The other party would counter-argue that they are a willing licensee, but you’re refusing to offer FRAND terms. The court or commission would then examine the evidence and make a determination.
Jordan Schneider: Can you define FRAND? What parameters do people argue about when one side claims FRAND and the other disagrees?
Brian Pomper: FRAND (Fair, Reasonable, and Non-Discriminatory) is intentionally not a defined term. In antitrust law, competing parties can’t agree on prices, as that would be a classic antitrust violation. Therefore, entities in a standard-setting body can’t collectively agree on a specific price for FRAND terms.
FRAND is a negotiated construct that can vary depending on the counterparty. For one party, it might involve monetary payment. For another, it could be a cross-licensing agreement. For a third, it might be a combination of payment and licensing. It’s a very fact-specific inquiry.
What’s clear is that if I agree to license you a patent on certain terms but tell someone else I won’t license it except for an exorbitant amount, the second offer is clearly not FRAND.
Jordan Schneider: Does every contract involving an SEP have to be public? How do you get to the bottom of this?
Brian Pomper: It’s a determination made by the court. If I claim my offer is FRAND and you disagree, I’m not entirely sure who has the burden of proof. If it’s me, I might need to present to the court the other licenses I’ve negotiated and demonstrate that the terms I’m offering you align with those previous agreements.
There’s no central clearinghouse for this information. You only get to that point once you’re in litigation.
Jordan Schneider: So, let’s say you win, and I’m found to be violating your patent. What are your options to penalize me?
Brian Pomper: Most patent owners, especially those with SEPs, want to license their technology rather than engage in court battles. In 99% or even 99.9% of cases, the SEP owner and the standard implementer agree on license terms.
If you were to lose, and the court ruled in favor of the patent owner, or the ITC said that they were going to issue an exclusion order for your products, then the patent owner would typically say, “Okay, I’m obligated to license to you on FRAND terms. Let’s discuss what that means now.” You could choose to stop implementing the standard, but chances are you’ll want a license.
Jordan Schneider: What’s an exclusion order?
Brian Pomper: The International Trade Commission, established in 1930, protects U.S. commerce from imports that infringe on U.S. intellectual property. If you’re a U.S. company and believe an import infringes your intellectual property, you can file a petition with the ITC requesting an exclusion order. This order instructs the customs department to prevent the infringing product from entering the United States. While this is a simplified explanation, it captures the essence of an exclusion order.
China and International Patent Law
Jordan Schneider: Let’s discuss the patent drama in the EU. What’s happening there?
Brian Pomper: The European Commission proposed, and the European Parliament recently passed, a series of changes to their intellectual property system, including modifications to how the Standard Essential Patent (SEP) system would work in Europe. These changes involve creating a government bureaucracy to determine the essentiality of patents and provide recommendations for license values.
This approach raises concerns. First, it seems to undermine the value of SEPs. Europe has some very large and successful companies that own substantial portfolios of SEPs, most notably Nokia and Ericsson. The motivation behind this regime appears to be driven by the automobile industry, particularly German auto companies, as cars are becoming essentially cell phones on wheels. These companies want to access the necessary technology as cheaply as possible, making it difficult for SEP owners to negotiate significant licensing fees.
It’s worth noting that the licenses for SEPs are relatively modest in the grand scheme of things. For example, Tesla charges users like me $99 a year for Lidar and WiFi technology in their cars, while they likely pay around $20-$25 for a lifetime license to that technology per car. This arrangement seems quite profitable for Tesla.
Another example is the dispute between Apple and Qualcomm over Qualcomm’s SEPs. Public reports suggested that Qualcomm was charging about $10 per iPhone for their license. Considering the price difference between an iPod Touch (without wireless technology) and an iPhone, this $10 fee seems reasonable for the added functionality.
These examples demonstrate that SEP owners aren’t necessarily trying to squeeze excessive profits from technology companies. Many tech companies view this as a volume game, where even small reductions in licensing costs for billions of devices can significantly impact their bottom line.
Jordan Schneider: Can you elaborate on how this situation in the EU relates to China?
Brian Pomper: The EU’s approach seems to inadvertently support China’s efforts in this area. Chinese courts have been attempting to establish authority in setting global licensing rates for SEPs. This means a court in China could dictate what a US or European company can charge for their technology worldwide, not just in China.
This approach undermines the value of SEPs and benefits China in two ways. First, it gives Chinese cell phone manufacturers like Huawei or Oppo cheaper access to necessary technology. Second, it weakens American and European technological leaders, which aids China in its geostrategic competition with the United States.
Jordan Schneider: How does jurisdiction work when there’s a dispute over whether licensing terms are FRAND?
Brian Pomper: In cases involving Chinese handset makers, we’ve seen them file declaratory judgment actions in China, claiming patent invalidity. This strategy aims to bring the case under Chinese jurisdiction, allowing Chinese courts to make global licensing determinations.
China’s Supreme People’s Court invalidates a patent held by Merck, a German biotech company, at the request of the Chinese chemical company JNC, April 2023. Source.
We need to consider the US patent system in the context of competition with China. The “Made in China 2025” plan outlines their intentions to dominate future technologies.
China’s command-and-control economy allows for massive, directed investments in specific technologies.
In contrast, the United States relies on market-based incentives to channel private sector money into innovation. The patent system is a crucial part of this policy architecture.
Unfortunately, the evolution of the US patent system over the past 10-15 years has led to a decline in early-stage innovation investment.
Many inventions that are unpatentable in the US are patentable in countries like Germany or China. This situation encourages inventors to develop their industries elsewhere, where they can better protect their technology.
Policymakers should consider the patent system in the context of our competition with China and work to create a more predictable and stable environment for patent rights.
Jordan Schneider: How are international trade agreements addressing these issues?
Brian Pomper: Trade agreements don’t typically address SEP issues directly. We’re seeing a complex situation where a Chinese court might claim the right to set global licensing rates, while the SEP owner might seek an anti-suit injunction in another jurisdiction, such as India, to say that the Chinese court can’t issue a global licensing rate. This results in competing court orders, creating a messy legal landscape.
Ultimately, companies often settle these disputes after extensive litigation. The concern is that the EU’s actions appear to support China’s efforts to become the primary forum for setting global licensing rates for SEPs, potentially undermining the SEP architecture in Europe.
Mom and Dad are Fighting Again
Jordan Schneider: I’m curious about the background of international trade lawyers who focus on technology and patents. Who ends up getting into this field and what are they like?
Brian Pomper: Most people in this field come from an IP law background and encounter international contexts in their work. I’m somewhat unusual in that I’m primarily a trade lawyer with an intellectual property background. There is a significant overlap between international trade and intellectual property, as we’ve discussed.
International agreements, coupled with the global nature of intellectual property companies, create a need for professionals who can defend IP rights overseas. These lawyers tend to be quite technical and detail-oriented.
Jordan Schneider: Litigating these “thermonuclear war” court cases must be fascinating — these companies are fighting over billions of dollars, while still doing business together.
To me, the lawyers in this dynamic are like kids watching Mom and Dad fight. It’s uncomfortable, but no one really thinks that divorce is on the table.
At the end of the day, there’s so much money to be made that these companies end up finding terms.
Any further thoughts on this dynamic?
Brian Pomper: These disputes are ultimately about licensing fees. The litigation is an attempt to gain leverage in determining the ultimate licensing fee. For example, in the Apple-Qualcomm dispute, Apple likely didn’t question whether it was using Qualcomm’s patents but was trying to negotiate a lower licensing fee.
When dealing with billions of products, even small changes in licensing fees can make a significant difference. We’re now seeing automobile companies join this fight as cars become more technologically advanced, seeking cheaper access to patented technologies.
This approach may be short-sighted for the United States and possibly even for the auto companies themselves. It often comes down to protecting bottom lines.
Policymakers need to ensure the system moves in a direction that best serves the future of the United States, considering what’s best for the system rather than any particular company.
ChinaTalk is a reader-supported publication. To receive new posts and support our work, consider becoming a subscriber.
Kweichow Moutai is best known for its 酱香 (literally “soy sauce flavored”) baijiu 白酒. But these days, the company is making the headlines with a semiconductor investment. Is this a brilliant diversification strategy, or has someone at headquarters been hitting the sauce a little too hard?
Just two months ago, after a stock downswing, an investment WeChat article urged investors to pivot from baijiu to semiconductors, claiming the market needs “real tech” 真科技, not “sauce-flavored tech” 酱香科技. Fast forward to today, and we’re seeing headlines proclaiming “sauce-flavored chips are here!” 酱香芯片来了 as Moutai crosses industry lines.
Let’s have a sober look at what’s behind these headlines.
What’s Moutai?
Moutai has high social status and is traditionally served in business or government settings. In 1972, then US President Richard Nixon and Chinese Premier Zhou Enlai toasted with Moutai. Henry Kissinger allegedly quipped, “If we drink enough Moutai, we can solve anything.”
Richard Nixon and Zhou Enlai toast each other with Moutai in the Great Hall of the People, Beijing, February 1972
Of late, Moutai has been branching out from banquet halls with crossovers like the Moutai Latte at Luckin Coffee, or Moutai-flavored ice cream.
Chip investments? Seriously?
Moutai’s latest move isn’t a new beverage but an investment in semiconductor design start-up Shanghai SmartLogic 上海思朗科技有限公司. Its founder, Wang Donglin 王东琳, once served as director of the esteemed Institute of Automation at the Chinese Academy of Sciences. Born 1956, he was already in his sixties when he founded SmartLogic in 2016.
The investment comes from two industrial development funds, both set up by Moutai in 2023. The first, Moutai Science and Technology Innovation Investment Fund 茅台科创（北京）投资基金合伙企业, is fully owned by Moutai and made a miniscule investment of 15,000 RMB in SmartLogic.
The second, Moutai Goldstone Industry Development Fund 茅台金石（贵州）产业发展基金合伙企业, is a joint fund by Moutai and Goldstone Investment 金石投资有限公司, an investment firm under government-owned conglomerate CITIC Group 中信集团. This fund invested a more substantial amount — 221,000 RMB — in SmartLogic, even though it still accounts for only 1.6% of the start-up’s investment. Together, these funds invested about 236,000 RMB (~US$33,000) in SmartLogic — a minuscule sum by semiconductor industry standards.
The two funds combined, however, have a registered capital of 750 million RMB, and have made some more sizable investments in biotech and new energy companies — for instance, a 670,00 RMB bet in Guizhou-based battery tech start-up Jiagui Energy Technology 珈硅能源科技 and a 3.7 million RMB (~US$500k) one in synthetic biotech firm Hongmo Biotechnology 虹摹生物科技.
Four investments conducted by Moutai Science and Technology Innovation Investment Fund to date.
SmartLogic is actually not Moutai’s first venture into chips. In 2017, another Moutai fund participated in a Series B round for Daoyuan Technology 稻源科技, an edge AI chip startup.
Overall, these are tiny numbers for a firm with a $250 billion market cap. But these investments are more than just a ploy for headlines — they help illustrate broader trends for SOEs today.
What is the logic behind a liquor brand venturing into high-tech investments?
Some Chinese industry analysts link it to Moutai’s own needs. Since 2013, the company has embedded chips in its bottles for anti-counterfeiting purposes, creating a demand for high-quality security chips. But this alone doesn’t justify investing in a chip-design startup.
There is little official explanation on specific investments. When establishing the two industrial development funds in 2023, they expressed hope to “enhance scientific and technological innovation capabilities” and “provide more abundant financial support for technological achievements moving towards marketization.”
How does this make any sense?
To better understand Moutai’s business logic, it is crucial to recognize that it is an SOE (state-owned enterprise), majority-owned (60.82%) by the Guizhou provincial government.
Making investments in seemingly unrelated businesses is nothing abnormal for SOEs. Moutai has set up its own investment arm in 2014, initially focusing on consumer goods.
But as the government pays increasing attention to high-tech industries, incentives for local governments are shifting. Demonstrating commitment to innovation trumps economic development. In an age where Xi Jinping bans extravagant party banquets and instead pushes a tech self-reliance narrative at every opportunity, alcohol and consumer goods don’t look good. But semiconductor design investments do.
In turn, incentives for local SOEs are shifting in the same direction as well. SOEs undergo a plethora of evaluations by their responsible government department, with science and technology playing an increasingly crucial role. Tech-focused investment funds boost scores in these assessments. While private companies are primarily driven by profits, these evaluations play a more crucial role in SOEs. For instance, managers’ promotion may hinge on strong annual evaluations.
Guess who was ranked No. 1 in tech-innovation assessments among all Guizhou provincial SOEs in 2022 and 2023? Kweichow Moutai!
While the sub-funds of Moutai’s investment arm initially focused on consumer goods, most sub-funds established in the past few years focus on high-tech industries.
Selection of Moutai investment funds set up from 2023-24. Many of their names suggest a focus on high-tech industries and pay lip service to the latest political slogans, such as the “High-Quality Growth Fund” 茅台（贵州）高质量成长股权投资合伙企业.
Such incentive structures can lead to herding behavior where the center shouts “chips” and everyone invests in chips. Look at the discourse around “new-quality productive forces” 新质生产力, a term encapsulating Beijing’s tech-powered economic ambitions. Soon after Xi first mentioned it in September 2023, every local government and SOE rushed to adopt the concept.
Everything became a new-quality productive force — including liquor. The local Industry and Commerce Bureau in the town where Moutai is located told state media that it would “promote the digital and smart upgrading of the ‘sauce-flavored’ baijiu industry to accelerate the cultivation of new quality productive forces” 积极推动酱酒产业数字化、智能化升级，加快培育新质生产力. At the Two Sessions in spring 2024, liquor SOE bosses proclaimed their supposed contributions to China’s tech self-reliance, and highlighted the need for“cultivating and strengthening Baijiu new-quality productive forces” 培育壮大白酒新质生产力.
The absurdity wasn’t lost on policymakers, and official discourse tried to dial down. Xi Jinping urged everyone to “guard against rushing in and creating bubbles” and to instead just “selectively promote the development of new industries, new models, and new growth drivers.” Party-state media published a flood of op-eds warning of irrational investments.
Is Moutai’s chip euphoria all misguided?
There is still a certain logic behind all this.
Beijing has recognized the reality that SOEs are not the most innovative firms. Some of them, however, have lots of cash. For around a decade, the new mantra is for SOEs to move from being “asset managers” 管资产 to “capital managers” 管资本.
In 2023, a high-level meeting by China’s state-owned capital administrators called to
promote the concentration of state-owned capital in forward-looking strategic emerging industries, and serve well as “long-term capital,” “patient capital,” and “strategic capital.”
推动国有资本向前瞻性战略性新兴产业集中，当好“长期资本”“耐心资本”“战略资本”
Hard tech investments require “patient capital,” and Moutai has barrels of that. It is the world’s most valuable spirits company. As of 2023, it had cash reserves of over 150 billion RMB (~US$21 billion), and its revenue is only growing. Further, most of Moutai’s tech investment funds are set up as “limited partnerships,” suggesting that the actual investment decisions are outsourced to professional investors.
Channeling some of this money to strategic industries is precisely how the center imagines SOEs to inject “patient capital” into its tech ecosystem. The government is surely hoping that Moutai orders far more rounds than its token investments we’ve discussed here.
The bigger debate
The size of Moutai’s chip investment is relatively small. Though not significant news for China’s semiconductor design industry, it plays to a deeper debate central to the future of China’s political economy: what role can “undesirable” industries (alcohol, social media, online gaming) play in an economy increasingly focused on national strategic priorities?
A 2018 state media op-ed argued that pitting these industries against each other is counterproductive. Instead, profits from “undesirable” sectors could fund strategic industries.
The development of shared bicycles, food delivery, and Moutai not only doesn’t conflict with the development of the chip industry, but can actually provide momentum for it. Using these industries as “scapegoats” for the lag in chip development is illogical.
…
Maotai is a consumer good, and economic growth is inseparable from consumption. Shared bicycles and food delivery, these new business forms and models, bring more convenience to people, stimulate more consumption, boost economic development, increase tax revenue, and only then does the state have money to support the chip industry.
Don’t expect these debates to be resolved any time soon.
Alcohol is also not the only drug whose sales revenue fuels China’s tech modernization. China Tobacco 中国烟草, controlling 96% of domestic cigarette sales, is effectively part of the Ministry of Industry and Information Technology — the same ministry overseeing China’s largest chip fund.
Perhaps Deng Xiaoping’s famous saying — “it doesn’t matter whether a cat is white or black, as long as it catches mice” — needs a modern update: “It doesn’t matter if profits come from baijiu or cigarettes, as long as they fund semiconductors”?
Time will tell if Moutai’s “sauce-flavored” investments will yield a successful blend or leave a bitter aftertaste.
ChinaTalk is a reader-supported publication. To receive new posts and support our work, consider becoming a free or paid subscriber.
Frontier AI models, what does it take to train and deploy them? How difficult is it to “fast follow”? What are the medium-term prospects for Chinese labs to catch up and surpass the likes of Anthropic, Google, and OpenAI?
To discuss, I have two guests from a podcast that has taught me a ton of engineering over the past few months, Alessio Fanelli and Shawn Wang from the Latent Space podcast. Alesio runs a venture capital firm called Decibel, and Shawn runs a firm called smol.ai. Both Dylan Patel and I agree that their show might be the best AI podcast around.
In this episode we discuss:
The secret sauce that lets frontier AI diffuses from top lab into Substacks.
How labs are managing the cultural shift from quasi-academic outfits to firms that need to turn a profit.
How open source raises the global AI standard, but why there’s likely to always be a gap between closed and open-source models.
China’s status as a “GPU-poor” nation.
Three key algorithmic innovations that could reshape the balance of power between the “GPU rich” and “GPU poor.”
Jordan Schneider: Let’s start off by talking through the ingredients that are necessary to train a frontier model.
Shawn Wang: At the very, very basic level, you need data and you need GPUs. You also need talented people to operate them.
Jordan Schneider: Let’s do the most basic. Let’s go from easy to complicated. Say all I want to do is take what’s open source and maybe tweak it a little bit for my particular firm, or use case, or language, or what have you. What’s involved in riding on the coattails of LLaMA and co.?
Shawn Wang: I would say the leading open-source models are LLaMA and Mistral, and both of them are very popular bases for creating a leading open-source model. This would not make you a frontier model, as it’s typically defined, but it can make you lead in terms of the open-source benchmarks.
Typically, what you would need is some understanding of how to fine-tune those open source-models. Those are readily available, even the mixture of experts (MoE) models are readily available. And then there are some fine-tuned data sets, whether it’s synthetic data sets or data sets that you’ve collected from some proprietary source somewhere. That’s definitely the way that you start.
Alessio Fanelli:
The biggest thing about frontier is you have to ask, what’s the frontier you’re trying to conquer?
OpenAI, DeepMind, these are all labs that are working towards AGI, I would say. That’s the end goal.
It’s one model that does everything really well and it’s amazing and all these different things, and gets closer and closer to human intelligence.
The open-source world, so far, has more been about the “GPU poors.” So if you don’t have a lot of GPUs, but you still want to get business value from AI, how can you do that? That’s a whole different set of problems than getting to AGI. A lot of times, it’s cheaper to solve those problems because you don’t need a lot of GPUs.
Sometimes, you need maybe data that is very unique to a specific domain. Or you might need a different product wrapper around the AI model that the larger labs are not interested in building.
The market is bifurcating right now. The open-source world has been really great at helping companies taking some of these models that are not as capable as GPT-4, but in a very narrow domain with very specific and unique data to yourself, you can make them better.
Data is definitely at the core of it now that LLaMA and Mistral — it’s like a GPU donation to the public. These models have been trained by Meta and by Mistral. Now you don’t have to spend the $20 million of GPU compute to do it. You can only spend a thousand dollars together or on MosaicML to do fine tuning. All of a sudden, the math really changes.
But, the data is important. But, if you want to build a model better than GPT-4, you need a lot of money, you need a lot of compute, you need a lot of data, you need a lot of smart people. You need a lot of everything.
That’s a much harder task.
Open Source vs Frontier Models
Jordan Schneider: One of the ways I’ve thought about conceptualizing the Chinese predicament — maybe not today, but in perhaps 2026/2027 — is a nation of GPU poors.
If the export controls end up playing out the way that the Biden administration hopes they do, then you may channel an entire country and multiple enormous billion-dollar startups and firms into going down these development paths.
They are not necessarily the sexiest thing from a “creating God” perspective. But they end up continuing to only lag a few months or years behind what’s happening in the leading Western labs.
A few questions follow from that. What are the mental models or frameworks you use to think about the gap between what’s available in open source plus fine-tuning as opposed to what the leading labs produce? What is driving that gap and how may you expect that to play out over time?
Shawn Wang:
The sad thing is as time passes we know less and less about what the big labs are doing because they don’t tell us, at all. We don’t know the size of GPT-4 even today.
We have some rumors and hints as to the architecture, just because people talk. And one of our podcast’s early claims to fame was having George Hotz, where he leaked the GPT-4 mixture of expert details.
But it’s very hard to compare Gemini versus GPT-4 versus Claude just because we don’t know the architecture of any of those things. And it’s all sort of closed-door research now, as these things become more and more valuable.
That said, I do think that the large labs are all pursuing step-change differences in model architecture that are going to really make a difference. Whereas, the GPU poors are typically pursuing more incremental changes based on techniques that are known to work, that would improve the state-of-the-art open-source models a moderate amount.
To date, even though GPT-4 finished training in August 2022, there is still no open-source model that even comes close to the original GPT-4, much less the November 6th GPT-4 Turbo that was released. That is even better than GPT-4. The closed models are well ahead of the open-source models and the gap is widening.
We can talk about speculations about what the big model labs are doing. We can also talk about what some of the Chinese companies are doing as well, which are pretty interesting from my point of view. But those seem more incremental versus what the big labs are likely to do in terms of the big leaps in AI progress that we’re going to likely see this year.
Alessio Fanelli: Yeah. And I think the other big thing about open source is retaining momentum. So a lot of open-source work is things that you can get out quickly that get interest and get more people looped into contributing to them versus a lot of the labs do work that is maybe less applicable in the short term that hopefully turns into a breakthrough later on.
So you can have different incentives. Therefore, it’s going to be hard to get open source to build a better model than GPT-4, just because there’s so many things that go into it. You can only figure those things out if you take a long time just experimenting and trying out.
“Oh, a talking [open-source] llama? Do tell.”
Loose Lips, Nerd Parties, & Frontier AI
Jordan Schneider: This idea of architecture innovation in a world in which people don’t publish their findings is a really interesting one.
One of the key questions is to what extent that knowledge will end up staying secret, both at a Western firm competition level, as well as a China versus the rest of the world’s labs level.
I’m curious, before we go into the architectures themselves. How does the knowledge of what the frontier labs are doing — even though they’re not publishing — end up leaking out into the broader ether?
Shawn Wang:
Yeah, honestly, just San Francisco parties. People just get together and talk because they went to school together or they worked together.
OpenAI does layoffs. I don’t know if people know that. They just did a fairly big one in January, where some people left.
Just through that natural attrition — people leave all the time, whether it’s by choice or not by choice, and then they talk.
They do take knowledge with them and, California is a non-compete state. You can’t violate IP, but you can take with you the knowledge that you gained working at a company. That does diffuse knowledge quite a bit between all the big labs — between Google, OpenAI, Anthropic, whatever. And so, I expect that is informally how things diffuse.
More formally, people do publish some papers. OpenAI has provided some detail on DALL-E 3 and GPT-4 Vision. That was surprising because they’re not as open on the language model stuff. DeepMind continues to publish quite a lot of papers on everything they do, except they don’t publish the models, so you can’t really try them out. You can go down the list in terms of Anthropic publishing a lot of interpretability research, but nothing on Claude.
You can go down the list and bet on the diffusion of knowledge through humans — natural attrition. There’s a fair amount of discussion.
You can see these ideas pop up in open source where they try to — if people hear about a good idea, they try to whitewash it and then brand it as their own. There’s a very prominent example with Upstage AI last December, where they took an idea that had been in the air, applied their own name on it, and then published it on paper, claiming that idea as their own. And there’s just a little bit of a hoo-ha around attribution and stuff.
But, if an idea is valuable, it’ll find its way out just because everyone’s going to be talking about it in that really small community. Sometimes it will be in its original form, and sometimes it will be in a different new form.
Jordan Schneider: Is that directional knowledge enough to get you most of the way there? Where does the know-how and the experience of actually having worked on these models in the past play into being able to unlock the benefits of whatever architectural innovation is coming down the pipeline or seems promising within one of the major labs?
Alessio Fanelli: I would say, a lot. Also, when we talk about some of these innovations, you need to actually have a model running. So if you think about mixture of experts, if you look at the Mistral MoE model, which is 8x7 billion parameters, heads, you need about 80 gigabytes of VRAM to run it, which is the largest H100 out there. If you’re trying to do that on GPT-4, which is a 220 billion heads, you need 3.5 terabytes of VRAM, which is 43 H100s.
You need people that are algorithm experts, but then you also need people that are system engineering experts.
You need people that are hardware experts to actually run these clusters.
The know-how is across a lot of things. You might even have people living at OpenAI that have unique ideas, but don’t actually have the rest of the stack to help them put it into use. Because they can’t actually get some of these clusters to run it at that scale.
The other example that you can think of is Anthropic. The founders of Anthropic used to work at OpenAI and, if you look at Claude, Claude is definitely on GPT-3.5 level as far as performance, but they couldn’t get to GPT-4. There’s already a gap there and they hadn’t been away from OpenAI for that long before. This learning is really quick.
Versus if you look at Mistral, the Mistral team came out of Meta and they were some of the authors on the LLaMA paper. Their model is better than LLaMA on a parameter-by-parameter basis. They had obviously some unique knowledge to themselves that they brought with them. It’s on a case-to-case basis depending on where your impact was at the previous firm.
“This past year, organizations released 149 foundation models, more than double the number released in 2022. Of these newly released models, 65.7% were open-source (meaning they can be freely used and modified by anyone), compared with only 44.4% in 2022 and 33.3% in 2021.” | Stanford HAI
Seeing Open Source Like a State
Jordan Schneider: This is the big question. Say a state actor hacks the GPT-4 weights and gets to read all of OpenAI’s emails for a few months. Is that all you need? To what extent is there also tacit knowledge, and the architecture already running, and this, that, and the other thing, in order to be able to run as fast as them?
Shawn Wang: Oh, for sure, a bunch of architecture that’s encoded in there that’s not going to be in the emails. It depends on what degree opponent you’re assuming. If talking about weights, weights you can publish right away. Just weights alone doesn’t do it. You have to have the code that matches it up and sometimes you can reconstruct it from the weights. Sometimes, you cannot.
But let’s just assume that you can steal GPT-4 right away. Then, going to the level of communication. Then, going to the level of tacit knowledge and infrastructure that is running. And I do think that the level of infrastructure for training extremely large models, like we’re likely to be talking trillion-parameter models this year.
Those extremely large models are going to be very proprietary and a collection of hard-won expertise to do with managing distributed GPU clusters.
Particularly that might be very specific to their setup, like what OpenAI has with Microsoft. That Microsoft effectively built an entire data center, out in Austin, for OpenAI.
I’m not sure how much of that you can steal without also stealing the infrastructure.
Alessio Fanelli: I think, in a way, you’ve seen some of this discussion with the semiconductor boom and the USSR and Zelenograd.
You can obviously copy a lot of the end product, but it’s hard to copy the process that takes you to it. Then, once you’re done with the process, you very quickly fall behind again.
If you got the GPT-4 weights, again like Shawn Wang said, the model was trained two years ago. So you’re already two years behind once you’ve figured out how to run it, which is not even that easy. So that’s really the hard part about it.
And software moves so quickly that in a way it’s good because you don’t have all the machinery to construct. But, at the same time, this is the first time when software has actually been really bound by hardware probably in the last 20–30 years.
Even getting GPT-4, you probably couldn’t serve more than 50,000 customers, I don’t know, 30,000 customers? There’s just not that many GPUs available for you to buy. That’s the other part. It’s like, academically, you could maybe run it, but you cannot compete with OpenAI because you cannot serve it at the same rate.
Jordan Schneider: It’s really interesting, thinking about the challenges from an industrial espionage perspective comparing across different industries. Because you’ve seen a fair amount of success with Huawei and routers back in the ’90s and 2000s.
But you had more mixed success when it comes to stuff like jet engines and aerospace where there’s a lot of tacit knowledge in there and building out everything that goes into manufacturing something that’s as fine-tuned as a jet engine.
It’s a really interesting contrast between on the one hand, it’s software, you can just download it, but also you can’t just download it because you’re training these new models and you have to deploy them to be able to end up having the models have any economic utility at the end of the day.
Alessio Fanelli: I was going to say, Jordan, another way to think about it, just in terms of open source and not as similar yet to the AI world where some countries, and even China in a way, were maybe our place is not to be at the cutting edge of this. It’s to actually have very massive manufacturing in NAND or not as cutting edge production.
I think open source is going to go in a similar way, where open source is going to be great at doing models in the 7, 15, 70-billion-parameters-range; and they’re going to be great models. They’re going to be very good for a lot of applications, but is AGI going to come from a few open-source people working on a model? I find that unlikely.
I think you’ll see maybe more concentration in the new year of, okay, let’s not actually worry about getting AGI here. Let’s just focus on getting a great model to do code generation, to do summarization, to do all these smaller tasks.
You may not like it, but this is what peak Soviet semiconductor manufacturing performance looks like… at least in Zelenograd. | Wikimedia Commons
There’s No Such Thing as a Free Model
Jordan Schneider: Well, what is the rationale for a Mistral or a Meta to spend, I don’t know, a hundred billion dollars training something and then just put it out for free? Does that make sense going forward? Or has the thing underpinning step-change increases in open source ultimately going to be cannibalized by capitalism?
Alessio Fanelli: Meta burns a lot more money than VR and AR, and they don’t get a lot out of it. I think the ROI on getting LLaMA was probably much higher, especially in terms of brand. I would say that helped them.
There’s obviously the good old VC-subsidized lifestyle, that in the United States we first had with ride-sharing and food delivery, where everything was free. I think now the same thing is happening with AI.
We have a lot of money flowing into these companies to train a model, do fine-tunes, offer very cheap AI imprints.
At some point, you got to make money.There’s not an endless amount of it.
But I think today, as you said, you need talent to do these things too. To get talent, you need to be able to attract it, to know that they’re going to do good work.
If you have a lot of money and you have a lot of GPUs, you can go to the best people and say, “Hey, why would you go work at a company that really cannot give you the infrastructure you need to do the work you need to do? Why don’t you work at Meta? Why don’t you work at Together AI?” You can work at Mistral or any of these companies.
So that’s another angle. It’s almost like the winners keep on winning. It’s like, okay, you’re already ahead because you have more GPUs. Now, you also got the best people. And because more people use you, you get more data. And it’s kind of like a self-fulfilling prophecy in a way.
So I think you’ll see more of that this year because LLaMA 3 is going to come out at some point. I’m sure Mistral is working on something else. OpenAI should release GPT-5, I think Sam said, “soon,” which I don’t know what that means in his mind. But he said, “You cannot out-accelerate me.” So it must be in the short term.
So yeah, there’s a lot coming up there.
Shawn Wang: There is a little bit of co-opting by capitalism, as you put it. Mistral only put out their 7B and 8x7B models, but their Mistral Medium model is effectively closed source, just like OpenAI’s.
In a way, you can start to see the open-source models as free-tier marketing for the closed-source versions of those open-source models.
If this Mistral playbook is what’s going on for some of the other companies as well, the perplexity ones.
There is some amount of that, which is open source can be a recruiting tool, which it is for Meta, or it can be marketing, which it is for Mistral. And there is some incentive to continue putting things out in open source, but it will obviously become increasingly competitive as the cost of these things goes up.
Jordan Schneider: One of the very dramatic things that my eyes were opened to at NeurIPS was that it’s one thing to see the charts about the nationality of researchers, and it’s another thing to be like, “Oh, man, everyone’s speaking Mandarin here, that’s cool.”
And if by 2025/2026, Huawei hasn’t gotten its act together and there just aren’t a lot of top-of-the-line AI accelerators for you to play with if you work at Baidu or Tencent, then there’s a relative trade-off.
Staying in the US versus taking a trip back to China and joining some startup that’s raised $500 million or whatever, ends up being another factor where the top engineers really end up wanting to spend their professional careers.
ChinaTalk is a reader-supported publication. To receive new posts and support our work, consider becoming a free or paid subscriber.
Open-Source Models & Chinese Labs
Shawn Wang: There is some draw. Yi, Qwen-VL/Alibaba, and DeepSeek all are very well-performing, respectable Chinese labs effectively that have secured their GPUs and have secured their reputation as research destinations. I would consider all of them on par with the major US ones.
Jordan Schneider: Let’s talk about those labs and those models. I’ve played around a fair amount with them and have come away just impressed with the performance. Any broader takes on what you’re seeing out of these companies?
Shawn Wang: DeepSeek is surprisingly good. All of the three that I mentioned are the leading ones. There are other attempts that are not as prominent, like Zhipu and all that. But I would say each of them have their own claim as to open-source models that have stood the test of time, at least in this very short AI cycle that everyone else outside of China is still using.
Usually, in the olden days, the pitch for Chinese models would be, “It does Chinese and English.” And then that would be the main source of differentiation.
But now, they’re just standing alone as really good coding models, really good general language models, really good bases for fine tuning. And I think that’s great.
Jordan Schneider: What’s interesting is you’ve seen a similar dynamic where the established firms have struggled relative to the startups where we had a Google was sitting on their hands for a while, and the same thing with Baidu of just not quite getting to where the independent labs were.
Sam: It’s interesting that Baidu seems to be the Google of China in many ways. I know they hate the Google-China comparison, but even Baidu’s AI launch was also uninspired. They announced ERNIE 4.0, and they were like, “Trust us. It’s better than everyone else.” And no one’s able to verify that.
OpenAI’s Secret Sauce?
Jordan Schneider: Yeah, it’s been an interesting ride for them, betting the house on this, only to be upstaged by a handful of startups that have raised like a hundred million dollars.
I want to come back to what makes OpenAI so special. You guys alluded to Anthropic seemingly not being able to capture the magic.
What from an organizational design perspective has really allowed them to pop relative to the other labs you guys think?
Alessio Fanelli: It’s always hard to say from the outside because they’re so secretive. Like Shawn Wang and I were at a hackathon at OpenAI maybe a year and a half ago, and they would host an event in their office. I think today you need DHS and security clearance to get into the OpenAI office. It’s hard to get a glimpse today into how they work.
Roon, who’s famous on Twitter, had this tweet saying all the people at OpenAI that make eye contact started working here in the last six months. The type of people that work in the company have changed.
I would say they’ve been early to the space, in relative terms. OpenAI is now, I would say, five maybe six years old, something like that. A lot of the labs and other new companies that start today that just want to do what they do, they cannot get equally great talent because a lot of the people that were great — Ilia and Karpathy and folks like that — are already there. They are passionate about the mission, and they’re already there.
Going back to the talent loop. It’s like, “Oh, I want to go work with Andrej Karpathy. I should go work at OpenAI.” “I want to go work with Sam Altman. I should go work at OpenAI.” That has been really, really helpful.
The other thing, they’ve done a lot more work trying to draw people in that are not researchers with some of their product launches. I actually don’t think they’re really great at product on an absolute scale compared to product companies. The GPTs and the plug-in store, they’re kind of half-baked.
But it inspires people that don’t just want to be limited to research to go there. That’s what then helps them capture more of the broader mindshare of product engineers and AI engineers. And they’re more in touch with the OpenAI brand because they get to play with it.
Also, for example, with Claude — I don’t think many people use Claude, but I use it. I use Claude API, but I don’t really go on the Claude Chat. Like there’s really not — it’s just really a simple text box. It makes that it is hard for exploration.
I would say that’s a lot of it. One of my friends left OpenAI recently. He was like a software engineer. He said Sam Altman called him personally and he was a fan of his work.
I don’t think in a lot of companies, you have the CEO of — probably the most important AI company in the world — call you on a Saturday, as an individual contributor saying, “Oh, I really appreciated your work and it’s sad to see you go.” That doesn’t happen often.
That kind of gives you a glimpse into the culture. If you look at Greg Brockman on Twitter — he’s just like an hardcore engineer — he’s not somebody that is just saying buzzwords and whatnot, and that attracts that kind of people.
How they got to the best results with GPT-4 — I don’t think it’s some secret scientific breakthrough. I think it’s more like sound engineering and a lot of it compounding together.
That’s what the other labs need to catch up on. They probably have similar PhD-level talent, but they might not have the same type of talent to get the infrastructure and the product around that.
Shawn Wang: There have been a few comments from Sam over the years that I do keep in mind whenever thinking about the building of OpenAI. He actually had a blog post maybe about two months ago called, “What I Wish Someone Had Told Me,” which is probably the closest you’ll ever get to an honest, direct reflection from Sam on how he thinks about building OpenAI.
A lot of it is fighting bureaucracy, spending time on recruiting, focusing on outcomes and not process.
For me, the more interesting reflection for Sam on ChatGPT was that he realized that you cannot just be a research-only company. You have to be sort of a full-stack research and product company. Now with, his venture into CHIPS, which he has strenuously denied commenting on, he’s going even more full stack than most people consider full stack.
That seems to be working quite a bit in AI — not being too narrow in your domain and being general in terms of the entire stack, thinking in first principles and what you need to happen, then hiring the people to get that going. It seems to be working for them really well.
Jordan Schneider: I felt a little bad for Sam. Those CHIPS Act applications have closed. I don’t think he’ll be able to get in on that gravy train. But it was funny seeing him talk, being on the one hand, “Yeah, I want to raise $7 trillion,” and “Chat with Raimondo about it,” just to get her take.
Engineering Talent at the Frontier
Jordan Schneider: Alessio, I want to come back to one of the things you said about this breakdown between having these research researchers and the engineers who are more on the system side doing the actual implementation.
There’s a long tradition in these lab-type organizations. The culture you want to create should be welcoming and exciting enough for researchers to give up academic careers without being all about production.
We’ve heard lots of stories — probably personally as well as reported in the news — about the challenges DeepMind has had in changing modes from “we’re just researching and doing stuff we think is cool” to Sundar saying, “Come on, I’m under the gun here. Everyone wants to fire me. Let’s ship some stuff already.”
Would you expand on the tension in these these organizations? They have to walk and chew gum at the same time.
Alessio Fanelli: I see a lot of this as what we do at Decibel. We invest in early-stage software infrastructure. Usually we’re working with the founders to build companies. I’ve seen a lot about how the talent evolves at different stages of it. I just mentioned this with OpenAI. It is not that old. It’s only five, six years old.
If you think about Google, you have a lot of talent depth. As with tech depth in code, talent is similar.
You’re trying to reorganize yourself in a new area. You have a lot of people already there. They might not be ready for what’s next.
If you think about AI five years ago, AlphaGo was the pinnacle of AI. It’s a research project. You’re playing Go against a person. But they’re bringing the computers to the place. They’re all sitting there running the algorithm in front of them. It’s not a product.
Now, all of a sudden, it’s like, “Oh, OpenAI has 100 million users, and we need to build Bard and Gemini to compete with them.” That’s a completely different ballpark to be in.
Some people might not want to do it. They might not be built for it. But then again, they’re your most senior people because they’ve been there this whole time, spearheading DeepMind and building their organization. It takes a bit of time to recalibrate that.
We see that in definitely a lot of our founders. They are people who were previously at large companies and felt like the company could not move themselves in a way that is going to be on track with the new technology wave. They end up starting new companies.
I think what has maybe stopped more of that from happening today is the companies are still doing well, especially OpenAI. OpenAI is an amazing business. I don’t really see a lot of founders leaving OpenAI to start something new because I think the consensus within the company is that they are by far the best. They have, by far, the best model, by far, the best access to capital and GPUs, and they have the best people.
There’s not leaving OpenAI and saying, “I’m going to start a company and dethrone them.” It’s kind of crazy.
You see maybe more of that in vertical applications — where people say OpenAI wants to be.
I use this analogy of synchronous versus asynchronous AI. OpenAI is very synchronous. You go on ChatGPT and it’s one-on-one. You use their chat completion API. You do one-on-one. And then there’s the whole asynchronous part, which is AI agents, copilots that work for you in the background. Things like that. That is not really in the OpenAI DNA so far in product.
You see a company — people leaving to start those kinds of companies — but outside of that it’s hard to convince founders to leave. We tried. We had some ideas that we wanted people to leave those companies and start and it’s really hard to get them out of it.
But I’m curious to see how OpenAI in the next two, three, four years changes. Because it will change by nature of the work that they’re doing. And maybe more OpenAI founders will pop up.
Jordan Schneider: Part of the reason you keep your brilliant PhD’s happy is to be the first to find and exploit architectural advancements.
In part 2, we get into:
OpenAI's secretive Q* project and what it means for GPT-5.
How a revolution in AI search could transform algorithmic problem-solving and shape how companies fast follow.
How AI diffusion could unlock massive productivity growth in GPU “rich” and “poor” nations alike in the coming decades.
Why the CCP’s biggest obstacle in AI development is… probably the CCP.
Chinese labs’ innovation on Mixture of Experts models, and how they’re outpacing Western labs in this domain.
Why coding will still be worth doing despite the rise of prompt engineering.
In his book Engineers of Victory, Paul Kennedy argues that the most important historical variable for transforming resources into hard power is a culture of innovation.
How does one achieve one’s strategic aims when one possesses considerable resources but does not, or at least not yet, have the instruments and organizations at hand?
[The answer is] the creation of war-making systems that contained impressive feedback loops, flexibility, a capacity to learn from mistakes, and a ‘culture of encouragement’…that permitted the middlemen in this grinding conflict the freedom to experiment, to offer ideas and opinions, and to cross traditional institutional boundaries.
To discuss the Defense Department’s culture of encouragement, ChinaTalk interviewed Chris Kirchhoff.
Chris served on Obama’s NSC and was a founding member of the Defense Innovation Unit (DIU). He recently published a book called Unit X: How the Pentagon and Silicon Valley Are Transforming the Future of War.
In our interview, we discuss:
The origin story of DIU and its early struggles to break Pentagon bureaucracy;
How DIU leveraged “waiver authority” to circumvent red tape under Defense Secretary Ash Carter;
Why the defense industrial base is ill-equipped to keep pace with technological change;
The case for shifting more DoD spending to non-traditional tech companies;
Lessons from commercial spaceflight for future AI governance, including potential issues with a “Manhattan project for AI.”
Secretary Ash Carter
Jordan Schneider: Let’s start with the origin of the DIU. What was Ash Carter’s vision and what was the pre-Ash version of DIU like?
Chris Kirchhoff: The story begins before Ash became Secretary of Defense. At the time, Pentagon leaders were wowed by the revolution in commercial technology in the 2010s. When I worked for General Marty Dempsey, Chairman of the Joint Chiefs of Staff, one of my first jobs was to organize a trip to Silicon Valley.
Marty Dempsey is an extraordinary leader, and he had what he called his “campaign of learning.” At our first event, a breakfast with Vint Cerf and other tech luminaries, General Dempsey announced: “I’m here on my campaign of learning. If I don’t learn something, Chris gets fired.”
President Obama meets with his Secretary of Defense Ash Carter (left) and Army Gen. Martin E. Dempsey, chairman of the Joint Chiefs of Staff (right), 2015. Source.
But what’s most memorable about the trip was our visit to Google, where we met with Astro Teller, head of Google X. Teller briefed us on their self-driving car project, which is now Waymo. One of the test cars, a Prius with a lidar dome, was parked next to our motorcade.
As we were walking back, General Dempsey said he wanted to go for a ride - which was absolutely not the plan given security protocols. But Dempsey got in the backseat and drove off, leaving me thinking I was definitely going to be fired.
He came back after 15 minutes, having watched the Prius navigate autonomously and maneuver around a cyclist. Dempsey asked the key question: “How much does all this hardware cost?” The engineers answered, “About $3,000.”
You could see the lightbulb go off in Dempsey’s head. This revolution in consumer electronics wasn’t just about smartphones — it would impact the military as well.
Jordan Schneider: When you think back about various revolutions in military affairs across history, where some people “got it” and others didn’t, what separates those who understand the future from those stuck in the current paradigm?
Chris Kirchhoff: You have to be modest — it’s easy to view the current push to link consumer and military tech ecosystems as unique. But the more military history you read, the more you realize the deep continuity with the past.
What’s different now is the relative scale. Today’s consumer tech market is $25 trillion. The Pentagon buys a lot of stuff, but we’re talking billions, not trillions. Never before has the commercial innovation ecosystem been so much larger than the defense ecosystem. That’s the generational shift Ash Carter spotted and acted on.
Jordan Schneider: If you think about precision targeting in the 80s and 90s, it wasn’t the DoD pushing shrinking nodes, but the relative balance between commercial and defense R&D was much closer than today.
Chris Kirchhoff: Exactly. That brings us to Ash Carter. In a 2001 article as a professor, he hypothesized that growth in commercial tech would quickly make that ecosystem so much larger than defense that it couldn’t be ignored.
With post-Cold War drawdowns in federal R&D, Ash realized DoD needed to shift to being a fast follower rather than always being generations ahead. He saw this clearly in 2001 and shortly after becoming Secretary moved to create the Defense Innovation Board and DIU.
Jordan Schneider: You have great stories about the outdated tech in military facilities circa 2016. To paraphrase some anecdotes, your co-author Raj Shah visited a command center that “was as if the military had resigned itself to becoming a display in the Museum of Computer History.” Trae Stephens, who co-founded Anduril, said of his early days in intelligence: “20% of my time was merging database files. I thought I’d be doing James Bond stuff with supercomputers. Instead, it was a joke.”
The most interesting part of DIU version 1.0 was that Congress tried to pull the plug and defund the program, which motivated the reset. What drove you and Raj to ultimately get on board for the reboot?
Chris Kirchhoff: It goes back to leadership. Ash Carter is a serious guy who doesn’t miss much. When he heard DIU wasn’t succeeding, he sent Todd Park to investigate firsthand. That says something about how Ash approached his job — not leaving anything unnoticed.
I credit Ash for realizing DIU’s initial incarnation wasn’t set up to succeed, going back to the drawing board, and thinking hard about how to make it work.
“To the extent present military and civilian leadership is articulating its strategy, it is one built, for the most part, on a continuation of previous programmatic and budgetary trendlines.
If there is a strategy for losing a future war in China, this is it.”
- Chris Kirchhoff in Unit X
Hacking the Pentagon Bureaucracy
Jordan Schneider: Can you tell the story of the “term sheet” you crafted at the beginning to set DIU on a path where it could potentially succeed?
Chris Kirchhoff: When Secretary Carter approached us about potentially leading the DIU reboot, we’d already watched the first team struggle. During an earlier visit to their office, I remember literally sitting on folding chairs for our briefing, thinking “I’m glad I didn’t come work here."
Raj and I carefully considered what would be needed for success in the reboot.
First, we needed personal commitment from the Secretary of Defense. We needed his personal buy-in to our unit because we knew how difficult the internal struggles would be. Working closely with the Secretary and his staff was critical to avoid being buried under layers of bureaucracy.
The second key item on our term sheet was “waiver authority.” This was a new idea in the department. The concept was that if you hit a policy or procedure (not a law) squarely blocking your mission, you could request it be waived.
Under the policy we set up, the owner of that policy had 14 days to handle a waiver request. If they declined, it would go immediately to Secretary Carter for review. This was an incredible bureaucratic weapon because now anyone who wanted to block us would have to answer to the Secretary of Defense.
That waiver authority ended up being the most important early tool we had at DIU.
Jordan Schneider: It’s striking that waiver authority and Other Transaction Authorities (OTAs) are basically ways to avoid doing things the way everyone else does. Thinking back, to what extent would you keep any of these rules if you could start from scratch?
Chris Kirchhoff: The challenge in any public sector institution is that you can’t really start from scratch. Our government doesn’t allow experimentation easily by design.
But at DIU, we tried to look afresh at what was possible. We had this phrase among our group of change agents: “Just legal enough.” ‘
There’s often an enormous distance between the strict letter of the law and where the practice is.
“Can we do this acquisition faster? Is that legal?” “Just legal enough.”
The most incredible example is Lauren Dailey, an acquisition dynamo. Her father was a tank driver, and her way of serving was to become a civilian acquisition expert in the Pentagon.
She stayed up late reading the text of the just-passed National Defense Authorization Act and found a single sentence that would change how OTA contracting could be done. It allowed the immediate conversion of a successful OTA pilot into a production contract.
This was a game-changer. Before, if you were a firm and got an OTA, there was no real path to go from a one-off technology pilot to something at scale. Lauren brought this to our attention, and we took it all the way up the chain.
In just over two weeks, we got everyone to agree on a whole new process for OTAs which we called the Commercial Solutions Opening (CSO). That allowed DIU to crack the hardest problem we faced - how to let contracts at Silicon Valley speed.
Jordan Schneider: We’re recording this July 3rd. Everyone across the executive branch is wringing their hands about the Supreme Court’s Chevron decision. Have you put any thought into this? Will this change help the DoD to shed some of its shackles?
Chris Kirchhoff: Well, I’m no lawyer, but I think Chevron will be much more disruptive for regulatory agencies. I’d be interested to learn what the exact implications are for the Department of Defense.
But it’s a general reminder that institutions tend to accumulate rules and procedures over time — and those might not serve their evolving goals. Leaders should consider what they can discard when trying new things.
Jordan Schneider: In the book, you point to an internal vetocracy in the DoD. If you want to do something, you actually need 20 people to not be upset with you. We see similar vetocracies across government with permitting for data centers and so on. There are many veto points where someone can block progress if you haven’t sufficiently convinced them of your vision.
Your theory of success required the buy-in from the person at the very top. What are your recommendations for bureaucrats at the other levels, who don’t necessarily have the Secretary backing them in every knife fight?
Chris Kirchhoff: I’ll explain with an anecdote. In the first couple of months at DIU, Ash Carter became aware of how little time he had left as Secretary. He wanted to make progress every day, which meant he sometimes moved incredibly fast.
Just after Raj and I were appointed, we got a call saying the Secretary was visiting Boston in three weeks and wanted to open a DIU office there during the trip. They also wanted us to have our official charter ratified by then.
Suddenly I found myself in the basement of the Pentagon with the office that writes charters for OSD entities. They have an elaborate process involving circulating drafts to hundreds of offices, any of which can object.
I asked, “What’s the fastest you’ve ever put a charter together.” They said 18 months for JIEDDO during the Iraq War, which they viewed as a land speed record.
A Bradley armored vehicle set ablaze by an improvised explosive device in Baghdad, 2004. “[A]fter five years of work, hundreds of projects, and a blizzard of cash paid to some of America’s biggest defense contractors, JIEDDO has not found a high-tech way to detect or defeat [IEDs].” Source: JIEDDO: The Manhattan Project that bombed
I had to uncomfortably push back and say, “The Secretary wants this in 3 weeks.”
One person raised their voice, insisting it was impossible. I had to respond, “I’d be happy to personally tell the Secretary that you said it couldn’t be done. Is that the message you wanted me to deliver?”
Ruffling feathers like this is uncomfortable but necessary.
We didn’t have the charter ready for Boston, but we had it not too many weeks after. That shows what leadership can do.
In the years after Secretary Carter departed, DIU quickly began to struggle, despite the fact that it did have a lot of support at the top, particularly under Secretary Mattis.
In public institutions, leadership is absolutely essential.
Artificially Unintelligent Institutions
Jordan Schneider: How much can AI reduce paperwork?
Chris Kirchhoff: An OSD AI official recently said to me, “Can you imagine if we had ChatGPT at our desktops? For OSD policy offices that spend most of their time writing briefing memos, reports, and template-compliant staff update emails — can you imagine how much faster they could work with LLM support?"
Even though we now have an entirely new office dedicated to this (the CDAO), it’s still remarkably difficult to make AI work in the public sector.
Any employee of a Fortune 100 company likely has a suite of AI tools at their desktop. In theory, people in OSD policy do as well, because the Pentagon decided to build a bespoke custom AI interface just for the DoD. Rather than buying something off the shelf like every Fortune 100 company in the country, we’ve made the problem harder for ourselves.
We still have a long way to go.
Jordan Schneider: I’m going to quote a show I did a few months ago with Garrett Berntsen of the CDAO:
Garrett Berntsen: In my perspective, having done this for a bunch of years now — that group of folks in the department is so stretched just trying to keep the lights on and get paper out the door. When faced with the prospect of taking on a new thing that’s going to be innovative and cool they’re like, “I don't know if I've got time to do that,” even if it would save them time in the long run.
… They’ve seen a ton of waves of this before. They’ve seen a lot of these, “we're going to modernize everything with X new tool,” 15 years ago. When e-mail rolled out, they were told that it was going to make everything easier. Instead, it’s made their lives an endless cycle of responding to e-mails.
Jordan Schneider: Let’s go a little sci-fi for a second. Beyond just having ChatGPT give you four paragraphs on what’s going on in Pakistan, are there other back-office applications of AI that you’re particularly excited about?
Chris Kirchhoff: Well, I’m sure there are many, and my greatest hope is just that the Department of Defense seizes the opportunity to experiment with AI, increasing their productivity, and making it easier for their employees to do knowledge work.
This is an incredible opportunity for the department, but again, DIU was created for a reason. There’s a beautiful quote in the opening of our book from Chris Brose, the longtime staff director of the Senate Armed Services Committee and now head of strategy for Anduril:
[T]he second decade of the twenty-first century was one of colossal missed opportunities for the US military. DoD, on the whole, had missed the advent of modern software development, the move to cloud computing, the commercial space revolution, the centrality of data, and the rise of artificial intelligence (AI) and machine learning. This was the case despite having funded the fundamental research that led to many of these advances. DIUx and DDS were to change this by placing DoD personnel directly in the commercial technology ecosystem.
The track record isn’t great. That should be foremost in mind of not only the current Secretary and Deputy Secretary, but the next several.
Jordan Schneider: Mike Horowitz, who is now working on emerging technology in the Office of the Secretary of Defense, previously wrote a book where he describes a two-by-two matrix for the level of organizational capital required to implement a major military revolution, versus the level of financial intensity.
Basically, if an innovation is easy and expensive, then this is something that the US is particularly well-suited to working with. But innovations in quadrant 4 are difficult even though they don’t require a ton of money, but rather require you to reimagine your current bureaucratic infrastructure. Those types of changes are much more difficult for a leading global hegemon to internalize.
Chris Kirchhoff: I admire Michael Horowitz’s scholarship. His work on innovation adoption is essential reading, and in fact we cite it in DIUx. The reality today, as we’re learning so brutally in Ukraine, is that inexpensive technology is quite potent and powerful.
Footage from drone strikes on Russian warships. Source: Euromaidan Press.
But today, what’s perhaps more relevant than the size of the investment, is whether it needs to happen through traditional budgeting channels or through new channels with which the department is not as comfortable.
In other words, it’s really easy for the Navy to buy another aircraft carrier. They know how to do that. They’ve been doing that for a long time. There are very few suppliers, there’s a very set schedule for getting something like that through Congress.
But at smaller levels, what if you want to buy 1,000 drones and experiment with them? Which category of funding and which program office does that?
It could be that there isn’t one and you actually have to create it.
Technology shifts have thrown us into a place where a lot of the department’s traditional processes are simply not applicable. That’s very uncomfortable for the institution. But of course, if you ignore that, you do so at your peril, because for every day you’re not fixing that and addressing that, you’re falling further behind the technology power curve.
DoD doesn’t have an innovation problem; it has an innovation adoption problem.
There’s plenty of incredible technology floating around the department at places like DARPA and now at DIU. But getting it to work at scale is not something the department does well.
Jordan Schneider: I want to throw two more quotes at you. First is from Bill LaPlante, who’s the most important acquisitions leader in the defense department:
Bill LaPlante: Ukraine is not holding their own against Russia with quantum. They’re not they’re not holding their own with AI. … It’s hardcore production of really serious weaponry. That’s what matters. Not to say that we shouldn't invest in quantum or we shouldn't invest in AI. I'm not at all saying that. … If somebody gives you a really cool liquored up story about a DIU or an OTA, ask them when it’s going into production, ask them how many numbers, ask them what the unit cost is going to be, ask them how it will work against China. … And don’t tell me it’s got AI and quantum in it. I don’t care.
The second quote is from Alex Wang, CEO of Scale AI:
Alex Wang: It doesn’t take a lot of imagination to realize that powerful AI systems, if they existed, would be a definitive military technology.
You could imagine that nuclear deterrence doesn’t even really matter in a world where you have advanced AI systems.
They can do superhuman hacking. They enable autonomous drone swarms. They can give you full situational awareness, they can develop biological weapons.
You can diffuse and perhaps neutralize an enemy’s nuclear arsenal.
Everyone within the national security community in the United States needs to understand, grok, and pay attention. This is a Manhattan project-level of technological capability.
Chris, thoughts on what these two mindsets tell us about the future?
Chris Kirchhoff: Well, I think there’s clearly a tension between those two views. Traditional weapon systems still have incredible relevance in Ukraine and elsewhere. At the same time, when we visited Ukraine, it was just astonishing to see the digital overlay that has been built around artillery, tanks, and trenches - things that have been around warfare for a very long time.
It really is going to be the synthesis of these two things where you’re going to see new combat power materialize. Ukraine, as horrific as the human disaster has been, has been a real wake-up call.
Jordan Schneider: The thing that Bill LaPlante sort of misses in that quote specifically is the idea that all that technological overlay is almost like table stakes. You have to have it in order to play the more traditional game of attrition with guns and old dumb artillery shells.
We’re clearly not in the world of Alex Wang’s future, where all you need is great technology and you’ll be able to achieve costless victory. But to completely dismiss the role that the tech bros are playing, I think, is also really missing what’s happening in the present and the future.
Chris Kirchhoff: That takes us back to the cover of the book. Even though the F-35 didn’t become fully operational until 2016, it’s still operating off of PowerPC processors. By 2016, we already had fifth-generation iPhones. It’s a stark reminder of how quickly technology moves compared to traditional defense acquisition cycles.
Lessons from Commercial Space for AI Governance
Jordan Schneider: We’ve seen an interesting arc around national security and AI labs over the past 18 months. More of the world, and the labs themselves, are waking up to the fact that AI is perhaps one of the most important dual-use technologies that will matter in coming years.
There’s increasing awareness around lab security issues and a steady bifurcation of the Chinese and US AI ecosystems. The current discussion among those who claim to be serious about national security assumes we’ll end up in a world where AI development is nationalized like the Manhattan Project.
The counterargument is that the private sector will end up doing whatever it wants. An interesting middle ground is the arc of the US space industry. What lessons do 20th and 21st-century commercial space hold for the future relationship between AI labs and government?
Chris Kirchhoff: My former colleague at Anthropic, Michael Page, made an interesting observation that commercial space could provide a model for how the US government handles increasingly sophisticated AI.
I’ll tell a quick story. My first job out of college was working on the Space Shuttle Columbia accident investigation. Space shuttles were an incredibly dangerous experimental technology - there are 5,100 critical failure points.
When I was working on that investigation, the idea that a private company could successfully launch people into space seemed far-fetched, maybe even insane. It seemed like you needed the full expertise of government to underwrite that kind of dangerous experiment.
For the next few years, I looked skeptically at the arrival of commercial space companies like SpaceX. I thought they were doomed to fail - that there was no way to improve on the engineering baselines that NASA engineers had labored for decades to achieve.
I turned out to be wrong.
SpaceX is wildly more successful at safely launching things into space — and at far lower cost than any government launch — with more or less the same tools NASA had all along.
This becomes relevant to AI, which unlike many advanced technologies has grown almost exclusively in the private sector. That’s after decades of government R&D contributions, but in terms of who’s operationalizing it now, it’s largely private companies.
Could NASA’s model of outsourcing launch to the private sector, with minimal engineering oversight of spacecraft operations, be a model for harnessing the nation’s AI capabilities? Could the government bolt onto itself private AI companies, rather than needing in-house AI development through national labs or another vehicle?
Jordan Schneider: The OpenAI and SpaceX stories are fascinating parables of organizational design. OpenAI didn’t have more compute than anyone else in 2017-2018, but they had the institutional freedom to take different bets than Google. SpaceX could innovate in ways Boeing couldn’t.
Chris Kirchhoff: Yes, and this is important to point out to those who hold up the national labs as a potential model for AI.
The national labs do incredibly important work in many areas. But that said, who produces more peer-reviewed publications, Google or the national labs? My money would be on Google.
The message here is that the private sector has proven to be an incredible national advantage for America. The free market has allowed a flourishing of research and development, particularly in the last two decades, that is just unparalleled.
I mean, DIU had to struggle so hard to pay advanced AI engineers half the salary they’d receive at a tech company.
We broke or nearly broke all the rules in the book, and we still had to convince people to take a 50% pay cut to come and work on the Department of Defense’s AI mission.
That makes me very cautious about those who view the national labs as a kind of turnkey solution.
Jordan Schneider: Chris, you wrote, “Unlike China, we have a system where information flows freely and where competition ensures that the best ideas and products rise to the top.”
Do we though?
Chris Kirchhoff: Well, the whole point of writing the book was to substantiate that claim. My co-author, Raj Shah, now runs Shield Capital, the first venture capital firm dedicated to companies that can contribute to national security.
The defense industrial complex is termed a “complex” because it’s not a free market in the sense that none of the major primes are product companies. They respond to a different set of incentives. They are saddled by the Department of Defense and the federal acquisition regulations and a different set of requirements for auditing and accounting.
They cannot operate like Google, Microsoft, or OpenAI, and they most certainly cannot operate like a startup. The debate we need to be having at some level is — how much of our defense spending do we want to go into institutions in that complex, and how much money do we want to shift into the free market innovation economy?
Jordan Schneider: Ultimately, this debate is going to get hashed out in Congress. What is it going to take for Congress to see the world in your way?
Chris Kirchhoff: On day two of our jobs, Raj Shah and I were saddled with the knowledge that Congress had just cut our entire budget, the whole thing, and that unless we hustled and really pulled a rabbit out of a hat, DIU would cease to exist.
The good news is, since then, there’s been a real swing of support towards the idea that the right thing for the Department of Defense to do is to access Silicon Valley and more technology from it.
In fact, in last year’s NDAA, we actually saw the Defense Innovation Unit get codified in law. Not only that, Congress bestowed upon it a budget of just under a billion dollars, which is really exceptional.
Even with such tremendous progress, only a very small portion of the Department of Defense’s budget goes to the new defense economy. Unless we change that, we’re increasingly going to find ourselves surprised.
We’ll be surprised by what’s happening in Ukraine, surprised by Hezbollah’s use of loitering munitions to depopulate northern Israel, surprised by Hamas using quadcopters to knock down Israeli border towers to enable 1000 fighters to pour over a kind of modern-day Maginot line.
We kind of ignore this debate, I think, very much at our own peril.
The Human Toll of Public Service
Jordan Schneider: One of the things I appreciated you writing about is the toll these jobs take on the people who want to do them well. Would you like to talk a bit about your experience?
Chris Kirchhoff: I left DIU in 2017 when the Obama administration stepped down. I was pretty sure I was done with full-time public service. I had given thirteen exceptional but also exhausting years.
The semester after I left, I taught undergraduates at Harvard. Many took my class because they aspired to work in national security. From their perspective, seeing pictures of me with President Obama in my office, it must have looked amazing.
But during office hours I had to tell them: It is wonderful. You get to work on incredibly important problems with real-world impact. It’s gratifying to use your political science training to make the world better.
But there are things you can’t unsee. I used to get an email before every US drone strike.
Whether in uniform or as a civilian, when that’s your day-to-day, it takes a toll. That weighed on me when I left DIU. I was almost positive I’d never go back into government. But you never say never.
The day I was most tempted was when Ron Klain, a former boss I really admire, was announced as the next president’s chief of staff. I thought hard about going back in. I called my oldest mentor, and she asked me three questions. The conversation went like this:
What do you actually remember about the last time you were working at the White House?
Driving home at 2 am. I was so tired that all I could think about was trying not to crash my car into Rock Creek Park.
How long ago was your divorce finalized?
About a year. I was with my first husband for 13 years. Not coincidentally, we had separated not long after my time in public service ended.
Are you ready to leave San Francisco? Haven’t you just met someone new there?
In three minutes, she helped me realize that. I share that story with purpose, because it’s important for people to realize the joys but also the burdens of serving in national security.
Jordan Schneider: You can’t completely remove people from involvement with life-or-death decisions at the Department of Defense. But then there are also stories of George Marshall playing afternoon golf during World War II.
I’m curious, what percentage of those 12-16 hour days actually needed to be worked? If the whole place ran better, might people be able to sustain themselves longer?
Chris Kirchhoff: You’re part of a system and a team, working collectively to help leaders make the best decisions possible.
That’s why it’s hard to go home at 5:30 pm in these jobs - you simply don’t want to do anything but your best.
But in addition to the tough days, there are also some incredible ones.
As the science nerd on General Dempsey’s staff, he asked me to closely track Ebola back when it was still a small outbreak in West Africa. The WHO maintained it was under control, but Dempsey wanted someone watching in case it became an emergency requiring military capabilities.
Fast forward a few months, and we went from a small outbreak to one threatening global security. Thanks to the pre-planning that summer, we had options ready for the Secretary and President on how the military could help.
That led to Operation United Assistance, which deployed 3,500 US service members to support 100,000 medical first responders in West Africa. That operation ultimately ended the Ebola epidemic.
Late in the outbreak, I traveled to West Africa with a White House delegation. I got to meet people who survived Ebola because of what DoD did. All you could do was hug people and cry because you had done something incredible for them and for America by helping stave off a global threat.
The job is full of sorrow and joy, and it’s all incredibly meaningful. With any luck, you do a lot of good even on the worst days.
Jordan Schneider: Chris, you just finished a policy residency at Anthropic. Why were you more interested in working at a lab?
Chris Kirchhoff: I really enjoyed my time at Anthropic, and I actually got to work there on a number of national security issues, including policy issues and issues of practice with Anthropics AI safety team, which they call their Frontier Red team.
I’m really excited to keep working in this area because I think generative AI will have a tremendous impact on American military capability. There are many opportunities but also many risks that need to be managed.
Jordan Schneider: Can you give us some book recommendations to take us out?
Chris Kirchhoff: I’ll give you three.
The first is called Tracers in the Dark: The Global Hunt for the Crime Lords of Cryptocurrency by Andy Greenberg. It’s an absolute page-turning barn burner of a story about crypto lords trying to hide their tracks from law enforcement, and the FBI agents who are hot pursuing them.
It’s a story about state sovereignty in the technological age, exploring whether states have the capacity to enforce the rule of law when criminals are backed by such sophisticated technology. You will tear through it. You won't be able to go to bed.
The second book is Say Nothing: A True Story of Murder and Memory in Northern Ireland by Patrick Radden Keefe. It’s an incredible history of The Troubles in Ireland, based on oral history and a set of interviews that journalists had never accessed before.
It paints this heartbreaking picture of violence between two groups who were sorting out their political institutions. In a strange way, I felt like the United Kingdom went through its own version of the Iraq War in Ireland — Ireland was kind of like Al Anbar, and the British military came in to try to pacify things, using violence that turned even more people against them. In terms of the sociology of violence and the sociology of conflict, it’s a tremendous read.
The last book is Malcolm Gladwell’s Bomber Mafia: A Dream, a Temptation, and the Longest Night of the Second World War, which is best consumed as an audiobook.
Gladwell paints these vivid portraits of characters during WWII. You see history through the eyes of soldiers and air force bombers, in both the Atlantic and the Pacific theaters.
Login
