而茶茶去意大利时正值炎热的夏季，那时我也在考虑是否加入茶茶的行程，一起探索我尚未去过的威尼斯和弗洛伦萨。由于我在6月就已经被西班牙热到，7月又在米兰晒了几天，向热夏投降的我和茶茶分享，这个时间可能会太热，多注意防晒和补水。没想到茶茶不仅独自出发，还在游荡途中享受solo trip，结束行程没多久便发来了她的这篇游荡之旅。看完文章的我想说，别管天气热不热，别管有没有同行人，就去做你想做的事，去你还有热情和兴趣的地方，Follow your heart and enjoy yourself。
下面是茶茶的意大利solo trip游荡之旅，祝大家阅读愉快。
（游览剑桥时看到这句slogn：enjoy a big bold beautiful journey）
We have a ChinaTalk meetup this coming Thursday in SF. Sign up here if you can make it!
Uncle Sam is taking a bite out of companies left and right. Today, we’re going to focus on MP Materials — the Trump administration’s answer to China’s restrictions on rare earth material exports to America.
Jordan Schneider: Why do deals like MP Materials even need to happen in the first place?
Daleep Singh: Critical minerals markets are broken for three main reasons.
First, there’s concentrated market power. China refines 70 to 90% of most minerals that we need to power clean energy, digital infrastructure, and defense systems. They have enormous market power — not just over supply, but also pricing, standards, and logistics. No market can be resilient if one player dominates the entire market ecosystem.
Second, there’s extreme price volatility. Prices for minerals like lithium, nickel, or rare earths swing far more violently than oil and gas. For producers, this creates asymmetric risk — if you undersupply the market, you may lose some profit. If you oversupply, you may go bust. That asymmetry deters the investment we need to expand supply when quantities are low and prices are high, preventing the market from clearing.
The third problem is that we don’t really have market infrastructure for critical minerals. For oil, we have futures exchanges, benchmark prices, and deep liquidity. For most critical minerals, we don’t. Transactions are opaque, bilateral, and heavily distorted by state intervention, especially China’s. Markets don’t provide price discovery, and producers and consumers don’t have hedging tools. Investors lose confidence in these markets and walk away.
All together, we have chronic underinvestment, chronic gaps between supply and demand, and chronic vulnerability to geopolitical shocks. Those are the problems.
Jordan Schneider: Daleep, let’s dig deeper into the market infrastructure piece. What does this mean in practice — that it’s not like WTI, Brent, or something similar?
Daleep Singh: If you’re a producer, you need tools to manage price volatility. When prices fall dramatically, you need the ability to continue generating revenue to stay liquid. You need futures markets and option markets that you can use to hedge against downside price risk. Right now, if you’re a critical minerals producer for most of the minerals that matter for our economic security, you don’t have that option.
You also need price discovery — to know where prices are in the market. We really don’t have genuine price discovery from any of these markets. China can decide, just by virtue of its dominance in supply, where it wants the price to settle. If it wants that price to settle at a level that wipes out the competition, that’s its choice. That’s not a market.
Arnab Datta: One quick piece to add is that the market infrastructure problem Daleep mentioned was really an intentional strategy by China. In addition to very robust industrial policy that provided substantial subsidies to producers and refiners, they stepped into the market infrastructure gap that was retreating in the West, particularly after the global financial crisis.
When you saw liquidity leave Western markets partly because of regulations passed during that time, China seized the opportunity. They built exchanges and benchmark contracts on Chinese exchanges so they could control that market infrastructure and how these prices were constructed.
Peter Harrell: I’d add two important pieces.
First, America’s dependence on China for rare earths is actually a relatively new problem. Historically, going back several decades, the US actually produced, mined, refined, processed, and manufactured plenty of rare earths in the 1950s, 1960s, really through the 1980s and into the 1990s.
It was in the 90s and 2000s — the era of peak globalization — where China successfully expanded its rare earth refining in particular. You saw Chinese firms begin to outcompete American firms, and a real decline in US manufacturing related to this consolidation of Chinese control. This isn’t because the US never made rare earths. This is really a problem of economics that emerged in the 90s and 2000s.
Second, we saw just a couple of months ago the critical risk that dependency on China for rare earths gives us, because it became part of the trade war Trump launched with China. Back in April, China retaliated by threatening to — and then actually — cutting off its exports of rare earths to the US, which had the potential to really impact manufacturing here. It became much less of a hypothetical long-term risk and much more of an immediate threat that could actually hurt the United States in the near term because of how China responded to Trump’s trade war in April.
Arnab Datta: Just to add to the WTI comparison — if you think about how WTI is priced, it’s a physically cleared contract. You’re purchasing a barrel that will be delivered at Cushing, Oklahoma. The pricing incorporates pipeline transport, logistics, and a whole infrastructure of traders, logistics providers, and port managers — all of that goes into the price of that physically delivered barrel at Cushing.
That’s something we just don’t have in the context of many of these newer metals markets. It’s very difficult to properly price a material when the only analog you have is a Chinese benchmark that potentially has very different constraints and very different characteristics.
Strategic Resilience Reserve
Jordan Schneider: This became very acute a few months ago when Trump imposed tariffs. Something that people have been talking about in Washington for literally 20 years — China using its role in the global rare earth export market to punish countries for doing things they don’t want — finally manifested. Trump walked back, and now we have this as a central thing that China and the US are tussling over.
Peter, Daleep, you guys aren’t dumb. You knew this was an issue. People have been writing about this for a very long time. What is the activation energy required in the 21st century to do the kind of industrial policy necessary to really change the dynamics on an issue like rare earths? Why have we only seen small, half-formed efforts until spring and summer 2025?
We have a Washington that has talked about the problem for a very long time now starting to spend nine and ten figures to address it in a more direct way than the incremental efforts folks had been pursuing. Peter, talk us through this deal. What came out of the Trump administration and the DoD over the past few weeks?
Peter Harrell: As you said, this isn’t a new problem. Policymakers have been aware for more than a decade that there was US dependency on China for rare earths. The Chinese had cut off their exports of rare earths to Japan back in 2011 or 2012. We’d actually seen the Chinese execute this playbook once before on an allied country.
This isn’t a new problem, and it’s not that there were no efforts to deal with this issue prior to the deal that the Defense Department announced in July. There were some efforts — previous grants, including to MP (the company that got the deal in July) to try to restart manufacturing and processing of rare earths in California where there’d been a longtime US mine. Actually, the mine had reopened in 2017.
There had also been some grants to other companies and universities to look at other ways of mining and processing rare earths — for example, to extract them from mine tailings in West Virginia. There had been some government money to try to sponsor innovation to reduce dependencies on rare earths, maybe create magnets and other products that you need rare earths for but without actually needing the rare earth elements.
There had been some policy processes and policy money put into trying to address this problem. But there were a couple of challenges with those prior efforts. First is just the scale of the effort. Frankly, the way Washington works, until there is a very acute crisis, it can be hard to mobilize the scale of effort that is actually needed to solve it. These prior efforts were much smaller in dollar spend and scope because the crisis seemed less acute. That’s just a political reality of how Washington works.
Second, this is a very complex issue. I don’t even think this new DoD deal with MP is going to be the whole solution. It’s going to require several parts. It is, in fact, a very complex issue.
Third, related to mobilization: solving a problem like this is going to cost money. You get into big debates about who should pay for it — should US taxpayers come up with the money, or should you make the private sector bear these costs? That adds to why it takes time. It’s not that there was nothing — there was some foundation that this deal is now building on. Not that there was nothing before, but Daleep, I’d welcome you defending our work together in the Biden administration.
Daleep Singh: Jordan, I appreciate you suggesting that we’re not dumb. That’s nice — we don’t always get that. But look, there have been piecemeal efforts to funnel public money toward private sector companies that could help produce minerals we need. What we haven’t done is fix the market. That’s where we are now.
I started thinking about reimagining the Strategic Petroleum Reserve into a Strategic Resilience Reserve for 21st-century vulnerabilities.
When prices crash and China continues to flood the market, we have this recurring problem of producers going bankrupt.
A Strategic Resilience Reserve could be a buyer of last resort or provide bridge financing to companies that are solvent but illiquid. That’s what could allow producers to keep producing during downturns and keep production capacity alive.
What can we do about investors not having confidence in these markets? If you don’t have futures markets and hedging markets, and refiners can’t lock in predictable revenues, could a Strategic Resilience Reserve step in with tools like selling a put option that allows you to make money when prices fall? Could it provide a price floor or some type of demand guarantee? The point is: can you create enough certainty for private capital to keep flowing?
What do you do about concentration risk? Even with a deal like MP, no country is going to mine its way to self-sufficiency when we’re up against what China has. But we do have producers and miners in places like Canada, Australia, and Finland. They’re hesitating to expand production because they know China can tank prices tomorrow.
An SRR, if we got that authorized, could provide demand backstops and offtake agreements. Could it intervene in the market so that producers in allied countries know they’re not going to go bankrupt if Beijing floods the market? That’s the idea we’ve started to develop over time — probably with some mistakes — to change the market itself rather than a series of ad hoc transactions that don’t alter the economics.
Jordan Schneider: SRR — a Strategic Resilience Reserve — a topic we’ll get to in a few moments. I’d also like to say in defense of the last 20 years of American policymaking that this was a latent threat, and the trajectory of US-China relations that made this become an actual threat has manifested relatively recently.
The fact that the Biden administration was able to “get away” with imposing semiconductor export controls, implementing big tariffs, essentially banning Chinese electric vehicles, and a handful of other tariffs without triggering this response is important to recognize. This is only a problem in the context of the US-China diplomatic relationship. Without that relationship souring, then we just get to use some subsidized magnets and the world moves on.
Peter, what was your thinking about trying to inch forward with more and more aggressive economic tools while seeing things bubbling up in terms of new Chinese legislation but not wanting them to hit back for the efforts you were making?
Peter Harrell: When I think about how one can solve a problem like our dependency on China for rare earth elements — and then we can unpack what this deal will and will not do — you need to think about several different categories of policy tools that you need to mesh together to solve the problem.
We’ve had this history in American industrial policy over the past decades where we’ve focused almost exclusively on what you might think of as supply-side industrial policy. We’ve given grants to companies to build a factory or a mine to do something. In some cases, that can be sufficient because the problem we need to solve is one of startup costs. It costs more to get something off the ground in the United States, and you can provide a capex incentive to help get it off the ground.
But when you look at China’s dominance of rare earths — where they not only have already spent a lot of capex, but their operating expenses are lower than in the United States and they control the market infrastructure — if you want to break China’s control here, you can’t solve it simply with our own capex.
You also need to think about the market infrastructure, as Daleep says, and you need to think about what the demand side looks like. If US operating costs for producing rare earths are going to be higher than they are in China, you have to find some demand for that higher-cost US product. Otherwise, US companies are going to keep buying Chinese products because the Chinese products are going to be cheaper.
You need to create a market infrastructure that’s going to ensure stable demand for the US-made product. Layering these things together — these different sets of policy tools to address the different parts of the chain — is not something the US government has done in a long time. You have to get your reps in and spend some time in the gym before you can do it.
Daleep Singh: Peter and I used to sit in the part of the White House that was straddling economics and national security. For most of us, very early on in the term we understood — especially as Russia’s forces were mounting on Ukraine’s border — that we’re going to be in this incredibly contested geopolitical environment for the rest of our lives. China and Russia have now made it very clear and revealed they’re going to challenge the US-led order everywhere. Because today’s great powers are nuclear powers, our expectation became that this competition is going to play out mostly in the theater of economics, energy, and technology.
The question was, if we’re going to prevail, how can we harness the financial firepower of the world’s most dynamic financial system to advance strategic objectives? Do we have the right tools, do we have the right institutions to overcome this short-term profit motive that drives most of what’s going on on Wall Street? The answer is no. As time went on and we started to have time to breathe, we started to think about new ideas. That’s where the Strategic Resilience Reserve came up. We also started to think about whether the US should have a sovereign wealth fund. These are all ideas trying to solve the same problem: the private sector systematically underinvests in exactly the kind of projects that matter most for our economic security and for our national security.
Can the Deal Create a Market?
Jordan Schneider: What does this MP Materials deal do? What is interesting and exciting about it? And why is it not the systemic solution that Daleep craves to manifest?
Arnab Datta: One thing this deal does is treat the problem holistically. Peter mentioned that you need a mix of supply and demand side tools. The administration deserves credit for using the DPA, the Defense Production Act, in a robust way. They are applying a toolkit that includes loans, equity investments, price floors, and a guaranteed contract for offtake for the finished product. That’s just a recognition. Ultimately, if we’re going to deal with this problem over the next one year, five year, ten year, decades, we need a robust toolkit and we need a mechanism by which we can address these very challenges.
Jordan Schneider: Arnab, briefly, who did this? This is very sophisticated, impressive work. It’s a lot of puzzle pieces which haven’t been put together in a very long time.
Arnab Datta: It was done through the Defense Department. It pairs a number of different authorities. I would say the most creative, atypical interventions were through the Defense Production Act — this is Title 3 of the Defense Production Act. It has very wide authority attached to it. Peter did a recent piece in Lawfare examining this, but it basically allows you to engage in a number of different transaction types to achieve the goal of building our defense industrial base. There’s also some capital from the Office of Strategic Capital. That’s where the loan is coming from.
One thing to keep in mind is that some of these appropriations are not spoken for. Over time you could imagine funding coming from different parts of DoD from the national defense stockpile. They’re going into this with the commitment and a very clear interest and effort in continuing with this deal. But there are some risks and there’s also some structural challenges with this deal that I’d be happy to go into as well.
Jordan Schneider: Peter, give us the flip side. What doesn’t this accomplish and solve?
Peter Harrell: Let’s first walk through what this deal is, because there was some news last month when it came out. I think a lot of the news focused on the fact that the Defense Department, as a piece of this deal, was taking equity in MP Materials, which now looks like a precursor for the Trump administration going out and taking equity in Intel and maybe a whole bunch of defense companies and everything else. I think that was the piece that attracted the news. But the deal is a fairly complicated deal that has a couple of different parts.
Part one of the deal is the government gave MP Materials, this mining company, some loans and then some cash as part of the equity stake to expand its mine in California, not that far from Las Vegas — Las Vegas is the nearest big airport to this mine, but it’s in California. To expand production at the mine and then relatedly to expand and build a new facility to take some of the rare earths being produced in this mine and to manufacture them into magnets, because what we need is not raw rare earths. What you need are magnets that go into motors and turbines and all kinds of other things. There’s almost no magnet manufacturing in the US and in fact, previously this mine had been producing rare earth ore and then selling it to China to be made into magnets there.
《日月浮沉》— copperplate print by Liu Kuo-sung 劉國松. Source.
Part of this is a capital injection to MP to expand the mine and to build some magnet processing — expand some magnet manufacturing capability here in the United States. They’re doing that with both a debt and equity stake.
Another part of this deal is the Defense Department set a price floor for the raw rare earths, where the Defense Department has guaranteed that when MP is mining and doing initial processing for the raw rare earths, it now has a guaranteed minimum price, which by the way, is about twice what the current Chinese market price is.
That’s how you guarantee that it’s economical for MP to make this stuff over the next ten years. Because DoD said, “Even if the market price is $54,” which is about what I think it is today, “We’re going to guarantee a price of $110 per kilogram. We’ll pay you the difference between $54 and $110 per kilogram.” You have this price floor for the minimally processed rare earths. Then on the magnet side, DoD also said, “We’ll buy all of your magnets. You can produce these magnets for the next ten years, and we’ll buy all of them.”
There are some interesting pieces, such as if DoD and MP jointly agree that some of the magnets can be sold to buyers other than DoD, then there will be some profit sharing and other provisions. But it’s actually a pretty complicated deal with interrelated parts, which very clearly does ensure the viable business for the next decade of MP. MP gets capital injection. MP gets a guaranteed price floor for its rare earths concentrates — minimally processed rare earths. And then MP has a guaranteed buyer for its magnet.
MP is taken care of for the next decade and will be able to scale up production of both the minimally processed rare earths and probably of magnets.
But that doesn’t mean we have a market here. What we have is a market for MP.
That’s where I think there’s some interesting questions about this deal. Are we right to bet all in on MP as a national champion, or should we be thinking more systemically about the markets and less about how we guarantee the success of this particular firm? Arnab, I know you have a lot of thoughts on that piece of it.
Arnab Datta: We have a forthcoming article on the topic. We’re hoping to get it into Alphaville there, but they’re working it up the chain. We’re not fully signed off.
Jordan Schneider: In this piece, Peter and Arnab, you point out that this is similar to Chinese industrial policy circa Mao era, not the version 2.0. You’re picking one winner. And by the way, this company is probably not the best managed company in the world, as opposed to the way that China does it, where you have lots of firms fight it out to be the top dog.
Once you whittle it down to not one, but five or seven, then you start really turning on the jets and pouring on the money to secure your position in the global marketplace. As Daleep alluded to, this is also a concern with Intel.
For what it’s worth, I do think that manufacturing at the leading edge probably doesn’t support as many entrants as opposed to just building some mines and making some batteries. But, there does seem to be some tricky incentives and a lot of risk that their head of mining doesn’t go to a Coldplay concert with their head of HR or something. Daleep, where are you on this as an approach?
Daleep Singh: It makes me think of Intel a lot and I realize that we’re talking about very different markets, but I have the same take on it. Let’s actually pivot for a moment to Intel. There definitely needs to be government intervention in both of these markets. With leading edge semiconductors, we don’t produce any of them. Intel’s the only US firm capable of making them. But it has no customers and without customers, Intel can’t scale its unit cost efficiency — remains low and its competitiveness lags. Market forces aren’t going to solve that problem, nor will it solve the problem for MP.
But what gets interesting is instrument choice. What I worry about is ad hoc improvisation about what tools of industrial policy to use for particular sectors with a different context and a different kind of problem to solve. What I come back to is the systematic stuff. We do need a playbook, a governance structure, a doctrine for industrial policy. Start with the strategic objective. What problem are we trying to solve? Whether it’s MP or Intel or any other company, what is the market failure? Is it a shortfall of demand? Is it a capital constraint? Is it a cost differential? Is there a coordination problem? Is there some national security externality?
Then the third step is: pick the policy instrument that remedies the failure. Don’t default to equity injections or subsidies if the problem is demand, for example. Can you actually intervene? This goes to Peter’s analysis on MP. Does the intervention sustain competition and does it avoid a single point of failure? I would try to avoid substituting a foreign monopoly for a domestic point of failure. Can you tie the support to milestones, objective milestones, so that you can claw back the support you’re giving from taxpayers if they underperform? Can you sunset the support to avoid permanent dependence?
The last thing is how are you measuring the strategic return? What is the metric for success with this deal? It can’t just be for financial gain. How are we going to measure the benefit in terms of resilience, security, technological edge? That’s what’s missing for me. Maybe it’s out there somewhere, I just haven’t heard it.
Arnab Datta: I’d add to that a couple of things. This is a national champion that’s crowned without contest. We do have a pretty robust, vigorous competitive process folding out right now in the magnets space. There are other companies. MP Materials has the Mountain Pass Mine, but it has never produced a commercial magnet. It has not sold a rare earth magnet at commercial scale. When you think about the challenges that go into selling commercially — automotive is a major purchaser of these magnets — you need to get your production facility warranted. That’s a long process. There’s no sense right now — we don’t know they could get warranted for automotive. They might not. It’s a very challenging process.
We do have competitors that are innovating. There’s a company, Niron Magnetics, that’s based out of Minnesota, they’ve produced a rare earthless magnet. This is the best of America in my opinion. You’re innovating yourself out of this vulnerability. I don’t know if Niron can scale at this point to the commercial scale that we need. But I also don’t know about MP Materials. When you start to get into some of these policy questions about is this intervention in this single company the right one, it raises a lot of secondary thorny issues.
This is a bet on vertical integration for rare earth magnets, that’s what they’re trying to build here. With MP Materials, that might be a good thing. A lot of the Chinese champions are vertically integrated, but there’s also a world where vertical integration on its own creates its own vulnerabilities. We see this a lot in the metals space where when we need to increase production because of some challenge, it’s not the vertically integrated producers that are responding quickly to price swings. It’s the marginal producers, the independent producers. This is something very common in metals markets. It’s something very common in the oil sector as well. These are really important policy questions.
My biggest concern globally with this deal is I don’t know what that reasoning is. It’s possible there are very well thought through reasons, but these are things that need to happen with some kind of a process that has technocratic democratic legitimacy to it. That’s why Daleep talking about the systemic solution is really important because we do need to make sure that these decisions are made in that context. I am not opposed to equity investments of all kinds. I think it’s an important tool for the government to have. It lets you push the risk frontier for your investments. If you’re a program, it lets you participate in the upside. But that needs to be done in a very thoughtful way. It’s a very powerful tool and we need to think about whether we are inculcating the things that make the American system dynamic — competition, innovation, technological innovation.
Daleep Singh: Can I ask Jordan, what is the exit strategy from the MP deal? Is it tied to production capacity or profits? How is the government going to sell down its public stake if at all?
Peter Harrell: The SEC filings talk about the government taking the stake. The government has also, in addition to the price floor, the guaranteed offtake agreement for the magnets. A belt and suspenders approach also guaranteed MP an annual profit of $140 million a year, which the government will pay as a cash payment if it’s not generated from the operations of the company. Presumably the government is intending to hold its equity for at least the ten year duration of the other elements of this deal. But there’s no specific language in the SEC filings about the government’s exit plan. It’s about the equity and then the duration of these other parts of the deal, which is a decade.
Arnab Datta: It’s structured as a ten year deal. I think ultimately the expectation is that the price floor and the offtake agreement will end at that point. But there’s no protection against the dependence. How do we stop this from becoming something that’s permanently dependent on this subsidy? It’s not clear.
It also doubles down on the Chinese market infrastructure. The benchmark that they are using is the Asian metals benchmark. That brings in the risk of manipulability too. China can bleed DoD for hundreds of millions more by flooding the market. How long is Congress expected to continue appropriations for that? These are not paid for. The one thing that was very clear in the 8-K is that they don’t have appropriations for all of this. How long can we expect Congress to keep paying? I think it is a very reasonable question as well.
Maximalist Industrial Policy
Jordan Schneider: I want to have this strategic question. What are reasonable goals over a three-year, five-year, or 10-year horizon when it comes to rare earths in particular. More broadly, what types of things would you want the Strategic Resilience Reserve to touch on?
Arnab Datta: There are a couple of key objectives that we’re trying to build here.
First, can we build a governance structure that is independent, technocratic and driven by market realities and not by political exigencies or other factors?
Second, can we build that robust toolkit that we talked about earlier for different markets? Rare earths we’ve talked about have particularly unique needs. They’re smaller than some of the bigger metals markets. We can’t be sure that you need a futures market for every rare earth that is on the market. But that’s a major goal as well.
Third, I would say the explicit purpose of what we’re trying to do here is build that competitive market. Are you supporting the buildout of a market infrastructure that is tied to market dynamics that US and allied producers face? Are we doing lending with intermediaries that can engage in more trading activity because they’ve got the leverage that left the market in the 2000s and 2010s, as I described? That’s an important piece of it because over five to ten years, if we can have a more stable market infrastructure for US and allied producers that reflects the costs they face, the logistical challenges they face, ultimately you’ll have a better stable foundation in place for those producers to compete.
Jordan Schneider: Beyond solving the market plumbing for things that would fall into strategic resilience, what is the big bold version of the systemic and thoughtful way to do the sorts of things that we’ve seen over the past few months with MP and Intel and we’ve seen over the past few years with the CHIPS Act and the IRA?
Daleep Singh: The maximalist version is a sovereign wealth fund. If you believe that the private sector systematically underinvests in projects that we need most for economic security and national security, then we’re not going to invest as a country at pace and scale to build fusion plants, dozens of semiconductor fabs, next-generation lithography, 6G telephony, or advanced geothermal. We’re also not going to invest enough in old economy sectors where we need to blunt a competitive disadvantage. Think about shipbuilding, or, lagging-edge chips, or mining.
What all of these projects share in common is that they require a lot of upfront capital and they require a decade or more of patience to generate a commercially attractive return. You need a huge tolerance for risk and uncertainty. The private sector venture investors, in particular, but also corporate America, are not likely to touch these in the size that we need them to because they’ve got plenty of other opportunities to make faster, higher, less risky returns. That’s why we have this valley of death right between breakthrough research and commercial scale.
I think the maximalist way to solve this problem is to create a flagship investment vehicle that gives the US patient, flexible capital, that can step in where markets won’t and that can crowd in private investment and back projects with genuine strategic value. That’s the case for a sovereign wealth fund. It’s not about picking winners, though. It’s about picking supply chains and technologies where our national security and our economic resilience are at stake.
It’s premised on the idea that left to itself, the US’ financial system is not designed to maximally align with our national interests. We need to intervene.
Jordan Schneider: I remember first reading you and Arnab’s piece on this a few years ago and thinking that was unlikely, but now Trump is into it. I wonder if it wasn’t called a golden share if he would have been as excited about this concept. But you do enough one-off ones and then you also learn that there are mistakes in the one-off ones and that you aren’t getting a systemic solution. It can go both ways. Either you give up on the project entirely or, given that the broader strategic purpose for these things keeps rearing its ugly head, you start to think in a larger and more systematic way at attacking these problems.
Let’s go level down. How are we funding this? What’s our governance structure? How’s the democratic involvement?
Daleep Singh: Whether you’re focusing on the MP deal or the 10% stake in Intel or the 15% revenue share from Nvidia or the golden share in Nippon, the point is we have a choice. Either we can improvise and experiment or we can develop a framework. Because I think the problem with improvisation is that if we just reach for different levers — an equity stake here, a profit share there, a golden share somewhere else — if we don’t have an overarching framework for why we’re using these tools and when and how and to what extent, I worry that this has the makings of a political piggy bank and a national embarrassment.
I understand some degree of experimentation is going to be needed. We haven’t done industrial policy in 40 years, and the muscles have atrophied. I get it, let’s take small steps and learn from those steps and then recalibrate. But I’m not in favor of ad hoc capitalism with American characteristics because that’s inevitably going to pick favorites and distort incentives.
You’re asking the right question. How do you govern a sovereign wealth fund or a Strategic Resilience Reserve the right way? How do you fund it? On the sovereign wealth fund idea, my thinking is we’re asset rich as a country. The federal government owns about 30% of the land. We have extensive energy and mineral rights. We own the electromagnetic spectrum. We have infrastructure assets all over the country. We’ve got 8,000 tons of gold that’s valued at 1934 prices. We’ve got $200 billion of basically money market assets that are sitting idle. The question is, are we maximizing the strategic bang for the buck on those assets? I would say no. That’s one potential source of funding.
You could also create new revenue streams to fund the vehicle. If you think that the US has too much Wall Street and not enough Main Street, that we financialize the economy into a series of boom-bust asset cycles, then let’s raise revenues from financial activities that serve no strategic purpose. I would say high frequency trading, for example, and fund vehicles that are explicitly designed to advance our national interests.
Jordan Schneider: As long as we stay away from fixed income.
Daleep Singh: Exactly. That’s untouchable. But the most appealing approach is the most straightforward one: ask Congress, be straight up about it. Ask Congress to seed the fund, authorize its existence as an independent federally chartered corporation authority. This is too important to leave entirely to the executive branch and have Congress set a clear mandate in terms of the objectives, the metrics for success, the oversight, the democratic accountability which Arnab was pointing at earlier. It’s a shame we didn’t do this ten years ago when our cost of capital was near zero. That would have made this effort far more affordable. But this is about our long-term national competitiveness. We don’t need to try to time the market.
Arnab Datta: One model that we think about a lot at Employ America is the Federal Reserve. The Federal Reserve has an independent board still, knock on wood. But that’s a structure that is well insulated from political day-to-day activities. It is not a 51-49 majority power structure. It has staggered terms, which, in my opinion, lends itself to depoliticization that’s helpful and has served us well over time.
In terms of the congressional point that Daleep made, we have had a version of this. We’ve worked with Senator Chris Coons’ office since 2020 on his proposal to establish an Industrial Finance Corporation. This is modeled off of the Reconstruction Finance Corporation that we had in the 30s, 40s, and 50s. We had then-Senator Vance on as a co-sponsor. I don't think the political viability of something like that is small. The way we structured that was we appropriated capital to it as a backstop against the borrowing that the corporation could do itself. This corporation could go out and raise capital by raising bond capital and then deploy that capital towards these investments that Daleep mentioned.
One value add about that is you don’t need to compete with the private sector on the rate of return, but you can generate a rate of return. Ultimately that type of a structure could pay for itself. There are a lot of technical accounting rules related to how you would structure that, particularly the Federal Credit Reform Act would come into play. But that is a structure that I think could be viable over time and we have the money to do it. Ultimately because a lot of these investments would be productive over 5, 10, 20 years, I think it would pay for itself.
The Right Tools for Intel
Jordan Schneider: I can’t let you guys leave without a few more Intel takes.
Arnab Datta: I’ve seen two separate conversations happening. One is on the legality of this and another on the policy justification. Peter did an excellent piece in Lawfare that came out a couple of days ago. This is possibly legal in a very technical sense, but does probably violate the spirit of the CHIPS Act in that the CHIPS Act is intended to incentivize manufacturing investments — they are giving this money to Intel but relinquishing most of those requirements. Earlier, we talked about milestones that companies should have to meet. Intel had a bunch of milestones attached to this money. They couldn’t get it all until they reached those milestones. They now have this capital, but they don’t have to meet those milestones. I think that’s a big problem.
Separate from the legality of the policy proposal here, why was this the best way, best thing for Intel? It’s not clear. As Daleep mentioned earlier, they need customers. An equity investment is not going to help them in that sense. For all I know, the share price could go down and our investment could go down because they can’t find customers. I think it’s a big problem that we’re not approaching the question of how can we make Intel more competitive? We seem to be approaching it in an ad hoc way — how can we get the best for our dollar in the form of a deal, an equity deal.
Daleep Singh: That’s my main concern — the right tools here. I agree with the intervention, but the right tools have to come from the demand side. Procurement guarantees, offtake agreements, sourcing mandates — all of those ideas make a lot of sense to me. It’s not clear how the equity injections fill the demand gap.
When you make upfront equity investments, you are foregoing optionality. I would have liked to see warrants or options that are tied to success. In general, I think policy support should be conditional. Conditional on whether you’re reducing unit costs or diversifying customers or hitting your production capacity targets. I do like the idea of clawbacks. The government has lost a lot of optionality with an upfront common equity injection. Maybe there’s a lot in the fine print that we don’t understand, but that’s what I found lacking.
Peter Harrell: I just echo what Daleep and Arnab said. The specifics of this deal are troubling. The idea of policy support, financial support to have onshoring of US semiconductors — clearly needed, clearly broad, bipartisan support. The idea that we shouldn’t be dependent on TSMC, the Taiwanese semiconductor firm for leading edge manufacturing, I think also has bipartisan and sensible policy support. You want to have some competition and some optionality at the leading edge of semiconductor manufacturing.
But what this deal did was take a grant in which Intel was getting $11 billion in exchange for Intel investing — call it $80 billion in fabs over the next decade. Intel was going to get the $11 billion in tranches as it built the fabs. If it failed to build the fabs, there was going to be a clawback. Now Intel is getting about $9 billion of the dollars in exchange for the stock. Plus they have to complete building certain DoD specialty lines.
Most of the obligations to build fabs went poof, and they got the cash in exchange for stocks.
I get why Intel might have done it. They get cash that’s largely unrestricted. They dilute their existing shareholders, but they probably decided the cash is worth it for us to do whatever we want with it. Reasonable call from Intel.
Arnab Datta: I’m also thinking about warrants. They’re using, in all likelihood, something called other transactions authority to legally justify the use of this deal. Other transactions authority is an incredible gift to the Commerce Department to be able to design very diverse mechanisms for policy here. In my opinion, wasting it on this equity investment that has little attached to it is a real mistake. They could put some effort into something creative that did go to the root of the problem about customers, and they’re squandering it, in my opinion.
Jordan Schneider: I think what you all said makes sense under a normal presidency living in the year of our Lord 2025. The way Intel survives is it gets customers, and the way it gets customers is Trump terrifies CEOs. If 10% of the company is what Trump can do to terrify CEOs, then all right, we’ll see. When we were talking earlier about MP Materials, it’s really not rocket science. You could have a beauty pageant with five different companies all trying to mine different places and have something. There’s one horse in this race and at a certain point you have to hope that they can execute as long as the demand’s there.
My sense and hope is that having a golden share owning 10% — Trump will care and be more invested and put more of his cycles and wrath into rounding up a handful of people who are going to spend the time to deal with Intel and help them get back on track. Regardless of whether it was warrants or a grant or equity, whether or not Intel is able to catch back up to TSMC is going to be a function of execution. And a president turning the screws on US fabless customer companies to play ball with Intel. The fact that Trump is caring about this and is focused on this, I would not have priced in completely from the get-go. He was literally talking about having to fire Pat Gelsinger — probably the only man who could, the person who I trust more than anyone else on the planet to actually execute this right who doesn’t work at TSMC currently. I’m more bullish on this than you guys are.
Arnab Datta: Can I offer one pushback on that, Jordan? One thing I would say is yes, there is a tremendous focusing mechanism — companies will, you saw this with MP where just a few days after the announcement Apple signed a big deal with them, a $500 million deal. The thing I would say is at some point the market has to trust that Trump’s commitment to this company will continue. President Trump is not going to be president forever. Intel is not going to be operating only on a four-year timeline. At some point Intel is going to require commitments from other companies and at some point they might turn and say this guy’s not going to be president anymore. We’ve got someone else to please here.
Certainly I take your bullish case. But Intel can’t survive only on that. They need an outside market and they need potentially capital from external sources down the line. At some point we’re going to be in a post-Trump world and it could look very different for Intel.
2023年3月末，田纳西州纳什维尔发生校园枪击案，3名9岁的学童和3名老师遇害。一周后，Charlie Kirk在犹他州盐湖城演讲。有观众问他，每年有很多人被枪打死，这种代价是不是值得？Charlie Kirk回答说：“I think it's worth to have a cost of, unfortunately, some gun deaths every single year, so that we can have the 2nd Amendment.”——“每年都有人被枪打死，我认为，不幸的是，这个代价值得，这样我们才能拥有第二修正案。”
Charlie Kirk遇刺是个不幸的悲剧。他已经不在人世，是非功过由世人去评说。上期节目，我说了自己的看法。有人喜欢，有人不喜欢，只要不突破人性的底线，都是正常的。这里再强调一点：不管他生前说过什么，主张过什么，在言论自由的民主社会，用杀人来解决观念争端，都是不折不扣的野蛮行为。
在当代美国，持枪权经常被称为“第二修正案权利”（The Second Amendment Right）。《美国宪法》第二修正案只有一句话：“A well regulated Militia, being necessary to the security of a free State, the right of the people to keep and bear Arms, shall not be infringed.”——“一支管理良好的民兵，为保障自由州的安全所必需，人民拥有和携带武器的权利不得侵犯。”
8VC is hosting a meetup for ChinaTalk this coming Thursday. Sign up here if you can make it!
Ryan Julian is a research scientist in embodied AI. He worked on large-scale robotics foundation models at DeepMind and got his PhD in machine learning in 2021.
In our conversation today, we discuss…
What makes a robot a robot, and what makes robotics so difficult,
The promise of robotic foundation models and strategies to overcome the data bottleneck,
Why full labor replacement is far less likely than human-robot synergy,
China’s top players in the robotic industry, and what sets them apart from American companies and research institutions,
How robots will impact manufacturing, and how quickly we can expect to see robotics take off.
Robotic arms at Tesla’s factory in Fremont, California. Source.
Embodying Intelligence
Jordan Schneider: Ryan, why should we care about robotics?
Ryan Julian: Robots represent the ultimate capital good. Just as power tools, washing machines, or automated factory equipment augment human labor, robots are designed to multiply human productivity. The hypothesis is straightforward — societies that master robotics will enjoy higher labor productivity and lower costs in sectors where robots are deployed, including in logistics, manufacturing, transportation, and beyond. Citizens in these societies will benefit from increased access to goods and services.
The implications become even more profound when we consider advanced robots capable of serving in domestic, office, and service sectors. These are traditionally areas that struggle with productivity growth. Instead of just robot vacuum cleaners, imagine robot house cleaners, robot home health aides, or automated auto mechanics. While these applications remain distant, they become less far-fetched each year.
Looking at broader societal trends, declining birth rates across the developed world present a critical challenge — How do we provide labor to societies with shrinking working-age populations? Robots could offer a viable solution.
From a geopolitical perspective, robots are dual-use technology. If they can make car production cheaper, they can also reduce the cost of weapon production. There’s also the direct military application of robots as weapons, which we’re already witnessing with drones in Ukraine. From a roboticist’s perspective, current military drones represent primitive applications of robotics and AI. Companies developing more intelligent robotic weapons using state-of-the-art robotics could have enormous implications, though this isn’t my area of expertise.
Fundamentally, robots are labor-saving machines, similar to ATMs or large language models. The key differences lie in their degree of sophistication and physicality. When we call something a robot, we’re describing a machine capable of automating physical tasks previously thought impossible to automate — tasks requiring meaningful and somewhat general sensing, reasoning, and interaction with the real world.
This intelligence requirement distinguishes robots from simple machines. Waymo vehicles and Roombas are robots, but dishwashers are appliances. This distinction explains why robotics is so exciting — we’re bringing labor-saving productivity gains to economic sectors previously thought untouchable.
Jordan Schneider: We’re beginning to understand the vision of unlimited intelligence — white-collar jobs can be potentially automated because anything done on a computer might eventually be handled better, faster, and smarter by future AI systems. But robotics extends this to the physical world, requiring both brain power and physical manipulation capabilities. It’s not just automated repetitive processes, but tasks requiring genuine intelligence combined with physical dexterity.
Ryan Julian: Exactly. You need sensing, reasoning, and interaction with the world in truly non-trivial ways that require intelligence. That’s what defines an intelligent robot.
I can flip your observation — robots are becoming the physical embodiment of the advanced AI you mentioned. Current large language models and vision-language models can perform incredible digital automation — analyzing thousands of PDFs or explaining how to bake a perfect cake. But that same model cannot actually bake the cake. It lacks arms, cannot interact with the world, and doesn’t see the real world in real time.
However, if you embed that transformer-based intelligence into a machine capable of sensing and interacting with the physical world, then that intelligence could affect not just digital content but the physical world itself. The same conversations about how AI might transform legal or other white-collar professions could equally apply to physical labor.
Today’s post is brought to you by 80,000 Hours, a nonprofit that helps people find fulfilling careers that do good.80,000 Hours — named for the average length of a career — has been doing in-depth research on AI issues for over a decade, producing reports on how the US and China can manage existential risk, scenarios for potential AI catastrophe, and examining the concrete steps you can take to help ensure AI development goes well.
Their research suggests that working to reduce risks from advanced AI could be one of the most impactful ways to make a positive difference in the world.
They provide free resources to help you contribute, including:
Detailed career reviews for paths like AI safety technical research, AI governance, information security, and AI hardware,
A job board with hundreds of high-impact opportunities,
A podcast featuring deep conversations with experts like Carl Shulman, Ajeya Cotra, and Tom Davidson,
Free, one-on-one career advising to help you find your best fit.
Jordan Schneider: Ryan, why is robotics so challenging?
Ryan Julian: Several factors make robotics exceptionally difficult. First, physics is unforgiving. Any robot must exist in and correctly interpret the physical world’s incredible variation. Consider a robot designed to work in any home — it needs to understand not just the visual aspects of every home worldwide, but also the physical properties. There are countless doorknob designs globally, and the robot must know how to operate each one.
The physical world also differs fundamentally from the digital realm. Digital systems are almost entirely reversible unless intentionally designed otherwise. You can undo edits in Microsoft Word, but when a robot knocks a cup off a table and cannot retrieve it, it has made an irreversible change to the world. This makes robot failures potentially catastrophic. Anyone with a robot vacuum has experienced it consuming a cable and requiring rescue — that’s an irreversible failure.
The technological maturity gap presents another major challenge. Systems like ChatGPT, Gemini, or DeepSeek process purely digital inputs — text, images, audio. They benefit from centuries of technological development that we take for granted — monitors, cameras, microphones, and our ability to digitize the physical world.
Today’s roboticist faces a vastly more complex challenge. While AI systems process existing digital representations of the physical world, roboticists must start from scratch. It’s as if you wanted to create ChatGPT but first had to build CPUs, wind speakers, microphones, and digital cameras.
Robotics is just emerging from this foundational period, where we’re creating hardware capable of converting physical world perception into processable data. We also face the reverse challenge — translating digital intent into physical motion, action, touch, and movement in the real world. Only now is robotics hardware reaching the point where building relatively capable systems for these dual processes is both possible and economical.
Jordan Schneider: Let’s explore the brain versus body distinction in robotics — the perception and decision-making systems versus the physical mechanics of grasping, moving, and locomotion. How do these two technological tracks interact with each other? From a historical perspective, which one has been leading and which has been lagging over the past few decades?
Ryan Julian: Robotics is a fairly old field within computing. Depending on who you ask, the first robotics researchers were probably Harry Nyquist and Norbert Wiener. These researchers were interested in cybernetics in the 1950s and 60s.
Norbert Wiener, founder of cybernetics, in an MIT classroom, ~1949. Source.
Back then, cybernetics, artificial intelligence, information theory, and control theory were all one unified field of study. These disciplines eventually branched off into separate domains. Control theory evolved to enable sophisticated systems like state-of-the-art fighter plane controls. Information theory developed into data mining, databases, and the big data processing that powers companies like Google and Oracle — essentially Web 1.0 and Web 2.0 infrastructure.
Artificial intelligence famously went into the desert. It had a major revolution in the 1980s, then experienced the great AI winter from the 80s through the late 90s, before the deep learning revolution emerged. The last child of this original unified field was cybernetics, which eventually became robotics.
The original agenda was ambitious — create thinking machines that could fully supplant human existence, human thought, and human labor — that is, true artificial intelligence. The founding premise was that these computers would need physical bodies to exist in the real world.
Robotics as a field of study is now about 75 years old. From its origins through approximately 2010-2015, enormous effort was devoted to creating robotic hardware systems that could reliably interact with the physical world with sufficient power and dexterity. The fundamental questions were basic but challenging — Do we have motors powerful enough for the task? Can we assemble them in a way that enables walking?
A major milestone was the MIT Cheetah project, led by Sangbae Kim around 2008-2012. This project had two significant impacts — it established the four-legged form factor now seen in Unitree’s quadrupedal robots and Boston Dynamics’ systems, and it advanced motor technology that defines how we build motors for modern robots.
Beyond the physical components, robots require sophisticated sensing capabilities. They need to capture visual information about the world and understand three-dimensional space. Self-driving cars drove significant investment in 3D sensing technology like LiDAR, advancing our ability to perceive spatial environments.
Each of these technological components traditionally required substantial development time. Engineers had to solve fundamental questions — Can we capture high-quality images? What resolution is possible? Can we accurately sense the world’s shape and the robot’s own body position? These challenges demanded breakthroughs in electrical engineering and sensor technology.
Once you have a machine with multiple sensors and actuators, particularly sensors that generate massive amounts of data, you need robust data processing capabilities. This requires substantial onboard computation to transform physical signals into actionable information and generate appropriate motion responses — all while the machine is moving.
This is where robotics historically faced limitations. Until recently, robotics remained a fairly niche field that hadn’t attracted the massive capital investment seen in areas like self-driving cars. Robotics researchers often had to ride the waves of technological innovation happening in other industries.
A perfect example is robotic motors. A breakthrough came from cheap brushless motors originally developed for electric skateboards and power drills. With minor modifications, these motors proved excellent for robotics applications. The high-volume production for consumer applications dramatically reduced costs for robotics.
The same pattern applies to computation. Moore’s Law and GPU development have been crucial for robotics advancement. Today, robots are becoming more capable because we can pack enormous computational power into small, battery-powered packages. This enables real-time processing of cameras, LiDAR, joint sensors, proprioception, and other critical systems — performing most essential computation onboard the robot itself.
Jordan Schneider: Why does computation need to happen on the robot itself? I mean, you could theoretically have something like Elon’s approach where you have a bartender who’s actually just a robot being controlled remotely from India. That doesn’t really count as true robotics though, right?
Ryan Julian: This is a fascinating debate and trade-off that people in the field are actively grappling with right now. Certain computations absolutely need to happen on the robot for physical reasons. The key framework for thinking about this is timing — specifically, what deadlines a robot faces when making decisions.
If you have a walking robot that needs to decide where to place its foot in the next 10 milliseconds, there’s simply no time to send a query to a cloud server and wait for a response. That sensing, computation, and action must all happen within the robot because the time constraints are so tight.
The critical boundary question becomes: what’s the timescale at which off-robot computation becomes feasible? This is something that many folks working on robotics foundation models are wrestling with right now. The answer isn’t entirely clear and depends on internet connection quality, but the threshold appears to be around one second.
If you have one second to make a decision, it’s probably feasible to query a cloud system. But if you need to make a decision in less than one second — certainly less than 100 milliseconds — then that computation must happen on-board. This applies to fundamental robot movements and safety decisions. You can’t rely on an unreliable internet connection when you need to keep the robot safe and prevent it from harming itself or others.
Large portions of the robot’s fundamental motion and movement decisions must stay local. However, people are experimenting with cloud-based computation for higher-level reasoning. For instance, if you want your robot to bake a cake or pack one item from each of ten different bins, it might be acceptable for the robot to query DeepSeek or ChatGPT to break that command down into executable steps. Even if the robot gets stuck, it could call for help at this level — but it can’t afford to ask a remote server where to place its foot.
One crucial consideration for commercial deployment is that we technologists and software engineers love to think of the internet as ubiquitous, always available, and perfectly reliable. But when you deploy real systems — whether self-driving cars, factory robots, or future home robots — there will always be places and times where internet access drops out.
Given the irreversibility we discussed earlier, it’s essential that when connectivity fails, the robot doesn’t need to maintain 100% functionality for every possible feature, but it must remain safe and be able to return to a state where it can become useful again once connectivity is restored.
Jordan Schneider: You mentioned wanting robots to be safe, but there are other actors who want robots to be dangerous. This flips everything on its head in the drone context. It’s not just that Verizon has poor coverage — it’s that Russia might be directing electronic warfare at you, actively trying to break that connection.
This creates interesting questions about the balance between pressing go on twenty drones and letting them figure things out autonomously versus having humans provide dynamic guidance — orienting left or right, adjusting to circumstances. There are both upsides and downsides to having robots make these decisions independently.
Ryan Julian: Exactly right. The more autonomy you demand, the more the difficulty scales exponentially from an intelligence perspective. This is why Waymos are Level 4 self-driving cars rather than Level 5 — because Level 5 represents such a high bar. Yet you can provide incredibly useful service with positive unit economics and game-changing safety improvements with just a little bit of human assistance.
Jordan Schneider: What role do humans play in Waymo operations?
Ryan Julian: I don’t have insider information on this, but my understanding is that when a Waymo encounters trouble — when it identifies circumstances where it doesn’t know how to navigate out of a space or determine where to go next — it’s programmed to pull over at the nearest safe location. The on-board system handles finding a safe place to stop.
Then the vehicle calls home over 5G or cellular connection to Waymo’s central support center. I don’t believe humans drive the car directly because of the real-time constraints we discussed earlier — the same timing limitations that apply to robot movement also apply to cars. However, humans can provide the vehicle with high-level instructions about where it should drive and what it should do next at a high level.
Jordan Schneider: We have a sense of the possibilities and challenges — the different technological trees you have to climb. What is everyone in the field excited about? Why is there so much money and energy being poured into this space over the past few years to unlock this future?
Ryan Julian: People are excited because there’s been a fundamental shift in how we build software for robots. I mentioned that the hardware is becoming fairly mature, but even with good hardware, we previously built robots as single-purpose machines. You would either buy robot hardware off the shelf or build it yourself, but then programming the robot required employing a room full of brilliant PhDs to write highly specialized robotic software for your specific problem.
These problems were usually not very general — things like moving parts from one belt to another. Even much more advanced systems that were state-of-the-art from 2017 through 2021, like Amazon’s logistics robots, were designed to pick anything off a belt and put it into a box, or pick anything off a shelf. The only variations were where the object is located, how I position my gripper around it, what shape it is, and where I move it.
From a human perspective, that’s very low variation — this is the lowest of low-skilled work. But even handling this level of variation required centuries of collective engineering work to accomplish with robots.
A pick-and-place robot aligns wafer cookies during the packaging process. Source.
Now everyone’s excited because we’re seeing a fundamental change in how we program robots. Rather than writing specific applications for every tiny task — which obviously doesn’t scale and puts a very low ceiling on what’s economical to automate — we’re seeing robotics follow the same path as software and AI. Programming robots is transforming from an engineering problem into a data and AI problem. That’s embodied AI. That’s what robot learning represents.
The idea is that groups of people develop robot learning software — embodied AI systems primarily composed of components you’re already familiar with from the large language model and vision-language model world. Think large transformer models, data processing pipelines, and related infrastructure, plus some robot-specific additions. You build this foundation once.
Then, when you want to automate a new application, rather than hiring a big team to build a highly specialized robot system and hope it works, you simply collect data on your new application and provide it to the embodied AI system. The system learns to perform the new task based on that data.
This would be exciting enough if it worked for just one task. But we’re living in the era of LLMs and VLMs — systems that demonstrate something remarkable. When you train one system to handle thousands of purely digital tasks — summarizing books, writing poems, solving math problems, writing show notes — you get what we call a foundation model.
When you want that foundation model to tackle a new task in the digital world, you can often give it just a little bit of data, or sometimes no data at all — just a prompt describing what you want. Because the system has extensive experience across many different tasks, it can relate its existing training to the new task and accomplish it with very little additional effort. You’re automating something previously not automated with minimal effort.
The hope for robotics foundation models is achieving the same effect with robots in the physical world. If we can create a model trained on many different robotic tasks across potentially many different robots — there’s debate in the field about this — we could create the GPT of robotics, the DeepSeek of robotics.
Imagine a robot that already knows how to make coffee, sort things in a warehouse, and clean up after your kids. You ask it to assemble a piece of IKEA furniture it’s never seen before. It might look through the manual and then put the furniture together. That’s probably a fantastical vision — maybe 10 to 20 years out, though we’ll see.
But consider a softer version: a business that wants to deploy robots only needs to apprentice those robots through one week to one month of data collection, then has a reliable automation system for that business task. This could be incredibly disruptive to the cost of introducing automation across many different spaces and sectors.
That’s why people are excited. We want the foundation model for robotics because it may unlock the ability to deploy robots in many places where they’re currently impossible to use because they’re not capable enough, or where deployment is technically possible but not economical.
Jordan Schneider: Is all the excitement on the intelligence side? Are batteries basically there? Is the cost structure for building robots basically there, or are there favorable curves we’re riding on those dimensions as well?
Ryan Julian: There’s incredible excitement in the hardware world too. I mentioned earlier that robotics history, particularly robotics hardware, has been riding the wave of other industries funding the hard tech innovations necessary to make robots economical. This remains true today.
You see a huge boom in humanoid robot companies today for several reasons. I gave you this vision of robotics foundation models and general-purpose robot brains. To fully realize that vision, you still need the robot body. It doesn’t help to have a general-purpose robot brain without a general-purpose robot body — at least from the perspective of folks building humanoids.
Humanoid robots are popular today as a deep tech concept because pairing them with a general-purpose brain creates a general-purpose labor-saving machine. This entire chain of companies is riding tremendous progress in multiple areas.
Battery technology has become denser, higher power, and cheaper. Actuator technology — motors — has become more powerful and less expensive. Speed reducers, the gearing at the end of motors or integrated into them, traditionally represented very expensive components in any machine using electric motors. But there’s been significant progress making these speed reducers high-precision and much cheaper.
Sensing has become dramatically cheaper. Camera sensors that used to cost hundreds of dollars are now the same sensors in your iPhone, costing two to five dollars. That’s among the most expensive components you can imagine, yet it’s now totally economical to place them all over a robot.
Computation costs have plummeted. The GPUs in a modern robot might be worth a couple hundred dollars, which represents an unimaginably low cost for the available computational power.
Robot bodies are riding this wave of improving technologies across the broader economy — all dual-use technologies that can be integrated into robots. This explains why Tesla’s Optimus humanoid program makes sense: much of the hardware in those robots is already being developed for other parts of Tesla’s business. But this pattern extends across the entire technology economy.
Jordan Schneider: Ryan, what do you want to tell Washington? Do you have policy asks to help create a flourishing robotics ecosystem in the 21st century?
Ryan Julian: My policy ask would be for policymakers and those who inform them to really learn about the technology before worrying too much about the implications for labor. There are definitely implications for labor, and there are also implications for the military. However, the history of technology shows that most new technologies are labor-multiplying and labor-assisting. There are very few instances of pure labor replacement.
I worry that if a labor replacement narrative takes hold in this space, it could really hold back the West and the entire field. As of today, a labor replacement narrative isn’t grounded in reality.
The level of autonomy and technology required to create complete labor replacement in any of the job categories we’ve discussed is incredibly high and very far off. It’s completely theoretical at this point.
My ask is, educate yourself and think about a world where we have incredibly useful tools that make people who are already working in jobs far more productive and safer.
China’s Edge and the Data Flywheel
Jordan Schneider: On the different dimensions you outlined, what are the comparative strengths and advantages of China and the ecosystem outside China?
Ryan Julian: I’m going to separate this comparison between research and industry, because there are interesting aspects on both sides. The short version is that robotics research in China is becoming very similar to the West in quality.
Let me share an anecdote. I started my PhD in 2017, and a big part of being a PhD student — and later a research scientist — is consuming tons of research: reams of dense 20-page PDFs packed with information. You become very good at triaging what’s worth your time and what’s not. You develop heuristics for what deserves your attention, what to throw away, what to skim, and what to read deeply.
Between 2017 and 2021, a reliable heuristic was that if a robotics or AI paper came from a Chinese lab, it probably wasn’t worth your time. It might be derivative, irrelevant, or lacking novelty. In some cases, it was plainly plagiarized. This wasn’t true for everything, but during that period it was a pretty good rule of thumb.
Over the last two years, I’ve had to update my priorities completely. The robotics and AI work coming out of China improves every day. The overall caliber still isn’t quite as high as the US, EU, and other Western institutions, but the best work in China — particularly in AI and my specialization in robotics — is rapidly catching up.
Today, when I see a robotics paper from China, I make sure to read the title and abstract carefully. A good portion of the time, I save it because I need to read it thoroughly. In a couple of years, the median quality may be the same. We can discuss the trends driving this — talent returning to China, people staying rather than coming to the US, government support — but it’s all coming together to create a robust ecosystem.
Moving from research to industry, there’s an interesting contrast. Due to industry culture in China, along with government incentives and the way funding works from provinces and VC funds, the Chinese robotics industry tends to focus on hardware and scale. They emphasize physical robot production.
When I talk to Chinese robotics companies, there’s always a story about deploying intelligent AI into real-world settings. However, they typically judge success by the quantity of robots produced — a straightforward industrial definition of success. This contrasts with US companies, which usually focus on creating breakthroughs and products that nobody else could create, where the real value lies in data, software, and AI.
Chinese robotics companies do want that data, software, and AI capabilities. But it’s clear that their business model is fundamentally built around selling robots. Therefore, they focus on making robot hardware cheaper and more advanced, producing them at scale, accessing the best components, and getting them into customers’ hands. They partner with upstream or downstream companies to handle the intelligence work, creating high-volume robot sales channels.
Take Unitree as a case study — a darling of the industry that’s been covered on your channel. Unitree has excelled at this approach. Wang Xingxing and his team essentially took the open-source design for the MIT Cheetah quadruped robot and perfected it. They refined the design, made it production-ready, and likely innovated extensively on the actuators and robot morphology. Most importantly, they transformed something you could build in a research lab at low scale into something manufacturable on production lines in Shenzhen or Shanghai.
They sold these robots to anyone willing to buy, which seemed questionable at the time — around 2016 — because there wasn’t really a market for robots. Now they’re the go-to player if you want to buy off-the-shelf robots. What do they highlight in their marketing materials? Volume, advanced actuators, and superior robot bodies.
This creates an interesting duality in the industry. Most American robotics companies — even those that are vertically integrated and produce their own robots — see the core value they’re creating as intelligence or the service they deliver to end customers. They’re either trying to deliver intelligence as a service (like models, foundation models, or ChatGPT-style queryable systems where you can pay for model training) or they’re pursuing fully vertical solutions where they deploy robots to perform labor, with value measured in hours of replaced work.
On the Chinese side, companies focus on producing exceptionally good robots.
Jordan Schneider: I’ve picked up pessimistic energy from several Western robotics efforts — a sense that China already has this in the bag. Where is that coming from, Ryan?
Ryan Julian: That’s a good question. If you view AI as a race between the US and China — a winner-take-all competition — and you’re pessimistic about the United States’ or the West’s ability to maintain an edge in intelligence, then I can see how you’d become very pessimistic about the West’s ability to maintain an edge in robotics.
As we discussed, a fully deployed robot is essentially a combination of software, AI (intelligence), and a machine. The challenging components to produce are the intelligence and the machine itself. The United States and the West aren’t particularly strong at manufacturing. They excel at design but struggle to manufacture advanced machines cheaply. They can build advanced machines, but not cost-effectively.
If you project this forward to a world where millions of robots are being produced — where the marginal cost of each robot becomes critical and intelligence essentially becomes free — then I can understand why someone would believe the country capable of producing the most advanced physical robot hardware fastest and at the lowest cost would have a huge advantage.
If you believe there’s no sustainable edge in intelligence — that intelligence will eventually have zero marginal cost and become essentially free — then you face a significant problem. That’s where the pessimism originates.
Jordan Schneider: Alright, we detoured but we’re coming back to this idea of a foundation model unlocking the future. We haven’t reached the levels of excitement for robotics that we saw in October 2022 for ChatGPT. What do we need? What’s on the roadmap? What are the key inputs?
Ryan Julian: To build a great, intelligent, general-purpose robot, you need the physical robot itself. We’ve talked extensively about how robotics is riding the wave of advancements elsewhere in the tech tree, making it easier to build these robots. Of course, it’s not quite finished yet. There are excellent companies — Boston Dynamics, 1X, Figure, and many others who might be upset if I don’t mention them, plus companies like Apptronik and Unitree — all working to build great robots. But that’s fundamentally an engineering problem, and we can apply the standard playbook of scale, cost reduction, and engineering to make them better.
The key unlock, assuming we have the robot bodies, is the robot brains. We already have a method for creating robot brains — you put a bunch of PhDs in a room and they toil for years creating a fairly limited, single-purpose robot. But that approach doesn’t scale.
To achieve meaningful impact on productivity, we need a robot brain that learns and can quickly learn new tasks. This is why people are excited about robotics foundation models.
How do we create a robotics foundation model? That’s the crucial question. Everything I’m about to say is hypothetical because we haven’t created one yet, but the current thinking is that creating a robotics foundation model shouldn’t be fundamentally different from creating a purely digital foundation model. The strategy is training larger and larger models.
However, the model can’t just be large for its own sake. To train a large model effectively, you need massive amounts of data — data proportionate to the model’s size. In large language models, there appears to be a magical threshold between 5 and 7 billion parameters where intelligence begins to emerge. That’s when you start seeing GPT-2 and GPT-3 behavior. We don’t know what that number is for robotics, but those parameters imply a certain data requirement.
What do we need to create a robotics foundation model? We need vast amounts of diverse data showing robots performing many useful tasks, preferably as much as possible in real-world scenarios. In other words, we need data and diversity at scale.
This is the biggest problem for embodied AI. How does ChatGPT get its data? How do Claude or Gemini get theirs? Some they purchase, especially recently, but first they ingest essentially the entire internet — billions of images and billions of sentences of text. Most of this content is free or available for download at low cost. While they do buy valuable data, the scale of their purchases is much smaller than the massive, unstructured ingestion of internet information.
There’s no internet of robot data. Frontier models train on billions of image-text pairs, while today’s robotics foundation models with the most data train on tens of thousands of examples — requiring herculean efforts from dozens or hundreds of people.
This creates a major chicken-and-egg problem. If we had this robotics foundation model, it would be practical and economical to deploy robots in various settings, have them learn on the fly, and collect data. In robotics and AI, we call this the data flywheel: you deploy systems in the world, those systems generate data through operation, you use that data to improve your system, which gives you a better system that you can deploy more widely, generating more data and continuous improvement.
We want to spin up this flywheel, but you need to start with a system good enough to justify its existence in the world. This is robotics’ fundamental quandary.
I want to add an important note about scale. Everyone talks about big data and getting as much data as possible, but a consistent finding for both purely digital foundation models and robotics foundation models is that diversity is far more important than scale. If you give me millions of pairs of identical text or millions of demonstrations of a robot doing exactly the same thing in exactly the same place, that won’t help my system learn.
The system needs to see not only lots of data, but data covering many different scenarios. This creates another economic challenge, because while you might consider the economics of deploying 100 robots in a space to perform tasks like package picking...
Jordan Schneider: Right, if we have a robot that can fold laundry, then it can fold laundry. But will folding laundry teach it how to assemble IKEA furniture? Probably not, right?
Ryan Julian: Exactly. Economics favor scale, but we want the opposite — a few examples of many different things. This is the most expensive possible way to organize data collection.
Jordan Schneider: I have a one-year-old, and watching her build up her physics brain — understanding the different properties of things and watching her fall in various ways, but never the same way twice — has been fascinating. If you put a new object in front of her, for instance, we have a Peloton and she fell once because she put her weight on the Peloton wheel, which moved. She has never done that again.
Ryan Julian: I’m sure she’s a genius.
Jordan Schneider: Human beings are amazing. They’re really good at learning. The ability to acquire language, for example — because robots can’t do it yet. Maybe because we have ChatGPT, figuring out speech seems less of a marvel now, but the fact that evolution and our neurons enable this, particularly because you come into the world not understanding everything... watching the data ingestion happen in real time has been a real treat. Do people study toddlers for this kind of research?
Ryan Julian: Absolutely. In robot learning research, the junior professor who just had their first kid and now bases all their lectures on watching how their child learns is such a common trope. It’s not just you — but we can genuinely learn from this observation.
First, children aren’t purely blank slates. They do know some things about the world. More importantly, kids are always learning. You might think, “My kid’s only one or two years old,” but imagine one or two years of continuous, waking, HD stereo video with complete information about where your body is in space. You’re listening to your parents speak words, watching parents and other people do things, observing how the world behaves.
This was the inspiration for why, up through about 2022, myself and other researchers were fascinated with using reinforcement learning to teach robots. Reinforcement learning is a set of machine learning tools that allows machines, AIs, and robots to learn through trial and error, much like you described with your one-year-old.
What’s been popular for the last few years has been a turn toward imitation learning, which essentially means showing the robot different ways of doing things repeatedly. Imitation learning has gained favor because of the chicken-and-egg problem: if you’re not very good at tasks, most of what you try and experience won’t teach you much.
If you’re a one-year-old bumbling around the world, that’s acceptable because you have 18, 20, or 30 years to figure things out. I’m 35 and still learning new things. But we have very high expectations for robots to be immediately competent. Additionally, it’s expensive, dangerous, and difficult to allow a robot to flail around the world, breaking things, people, and itself while doing reinforcement learning in real environments. It’s simply not practical.
Having humans demonstrate tasks for robots is somewhat more practical than pure reinforcement learning. But this all comes down to solving the chicken-and-egg problem I mentioned, and nobody really knows the complete solution.
There are several approaches we can take. First, we don’t necessarily have to start from scratch. Some recent exciting results that have generated significant enthusiasm came from teams I’ve worked with, my collaborators, and other labs. We demonstrated that if we start with a state-of-the-art vision-language model and teach it robotics tasks, it can transfer knowledge from the purely digital world — like knowing “What’s the flag of Germany?” — and apply it to robotics.
Imagine you give one of these models data showing how to pick and place objects: picking things off tables, moving them to other locations, putting them down. But suppose it’s never seen a flag before, or specifically the flag of Germany, and it’s never seen a dinosaur, but it has picked up objects of similar size. You can say, “Please pick up the dinosaur and place it on the flag of Germany.” Neither the dinosaur nor the German flag were in your robotics training data, but they were part of the vision-language model’s training.
My collaborators and I, along with other researchers, showed that the system can identify “This is a dinosaur” and use its previous experience picking up objects to grab that toy dinosaur, then move it to the flag on the table that it recognizes as Germany’s flag.
One tactic — don’t start with a blank slate. Begin with something that already has knowledge.
Another approach — and this explains all those impressive dancing videos you see from China, with robots running and performing acrobatics — involves training robots in simulation using reinforcement learning, provided the physical complexity isn’t too demanding. For tasks like walking (I know I say “just” walking, but it’s actually quite complex) or general body movement, it turns out we can model the physics reasonably well on computers. We can do 99% of the training in simulation, then have robots performing those cool dance routines.
We might be able to extend this framework to much more challenging physical tasks like pouring tea, manipulating objects, and assembling things. Those physical interactions are far more complex, but you could imagine extending the simulation approach.
Jordan Schneider: Or navigating around Bakhmut or something.
Ryan Julian: Exactly, right. The second approach uses simulation. A third tactic involves getting data from sources that aren’t robots but are similar. This has been a persistent goal in robot learning for years — everyone wants robots to learn from watching YouTube videos.
There are numerous difficult challenges in achieving this, but the basic idea is extracting task information from existing video data, either from a first-person perspective (looking through the human’s eyes) or third-person perspective (watching a human perform tasks). We already have extensive video footage of people doing things.
What I’ve described represents state-of-the-art frontier research. Nobody knows exactly how to accomplish it, but these are some of our hopes. The research community tends to split into camps and companies around which strategy will ultimately succeed.
Then there’s always the “throw a giant pile of money at the problem” strategy, which represents the current gold standard. What we know works right now — and what many people are increasingly willing to fund — is building hundreds or even thousands of robots, deploying them in real environments like factories, laundries, logistics centers, and restaurants. You pay people to remotely control these robots to perform desired tasks, collect that data, and use it to train your robotics foundation model.
The hope is that you don’t run out of money before reaching that magic knee in the curve — the critical threshold we see in every other foundation model where the model becomes large enough and the data becomes sufficiently big and diverse that we suddenly have a model that learns very quickly.
There’s a whole arms race around how to deploy capital quickly enough and in the right way to find the inflection point in that curve.
Jordan Schneider: Is Waymo an example of throwing enough money at the problem to get to the solution?
Ryan Julian: Great example.
Jordan Schneider: How do we categorize that?
Ryan Julian: Waymo and other self-driving cars give people faith that this approach might work. When you step into a Waymo today, you’re being driven by what is, at its core, a robotics foundation model. There’s a single model where camera, lidar, and other sensor information from the car comes in, gets tokenized, decisions are made about what to do next, and actions emerge telling the car where to move.
That’s not the complete story. There are layers upon layers of safety systems, decision-making processes, and other checks and balances within Waymo to ensure the output is sound and won’t harm anyone. But the core process remains: collect data on the task (in this case, moving around a city in a car), use it to train a model, then use that model to produce the information you need.
Self-driving cars have been a long journey, but their success using this technique gives people significant confidence in the approach.
Let me temper your enthusiasm a bit. There’s hope, but here’s why it’s challenging. From a robotics perspective, a self-driving car is absolutely a robot. However, from that same perspective, a self-driving car has an extremely simple job — it performs only one task.
The job of a self-driving car is to transport you, Jordan, and perhaps your companions from point A to point B in a city according to a fairly limited set of traffic rules, on a relatively predictable route. The roads aren’t completely predictable, but they follow consistent patterns. The car must accomplish this without touching anything. That’s it — get from point A to point B without making contact with anything.
The general-purpose robots we’re discussing here derive their value from performing thousands of tasks, or at least hundreds, without requiring extensive training data for each one. This represents one axis of difficulty: we must handle many different tasks rather than just one.
The other challenge is that “don’t touch anything” requirement, which is incredibly convenient because every car drives essentially the same way from a physics perspective.
Jordan Schneider: Other drivers are trying to avoid you — they’re on your side and attempting to avoid collisions.
Ryan Julian: Exactly — just don’t touch anything. Whatever you do, don’t make contact. As soon as you start touching objects, the physics become far more complicated, making it much more difficult for machines to decide what to do.
The usefulness of a general-purpose robot lies in its ability to interact with objects. Unless it’s going to roam around your house or business, providing motivation and telling jokes, it needs to manipulate things to be valuable.
These are the two major leaps we need to make from the self-driving car era to the general robotics era — handling many different tasks and physically interacting with the world.
Jordan Schneider: Who are the companies in China and the rest of the world that folks should be paying attention to?
Ryan Julian: The Chinese space is gigantic, so I can only name a few companies. There are great online resources if you search for “Chinese robotics ecosystem."
In the West, particularly the US, I would divide the companies really pushing this space into two camps.
The first camp consists of hardware-forward companies that think about building and deploying robots. These tend to be vertically integrated. I call them “vertical-ish” because almost all want to build their own embodied AI, but they approach it from a “build the whole robot, integrate the AI, deploy the robot” perspective.
In this category, you have Figure AI, a vertical humanoid robot builder that also develops its own intelligence. There’s 1X Technologies, which focuses on home robots, at least currently. Boston Dynamics is the famous first mover in the space, focusing on heavy industrial robots with the Atlas platform. Apptronik has partnered with Google DeepMind and focuses on light industrial logistics applications.
Tesla Optimus is probably the most well-known entry in the space, with lots of rhetoric from Elon about how many robots they’ll make, where they’ll deploy them, and how they’ll be in homes. But it’s clear that Tesla’s first value-add will be helping automate Tesla factories. Much of the capital and many prospective customers in this space are actually automakers looking to create better automation for their future workforce.
Apple is also moving into the space with a very early effort to build humanoid robots.
The second camp focuses on robotics foundation models and software. These tend to be “horizontal-ish” — some may have bets on making their own hardware, but their core focus is foundation model AI.
My former employer, Google DeepMind, has a robotics group working on Gemini Robotics. NVIDIA also has a group doing this work, which helps them sell chips.
Among startups, there’s Physical Intelligence, founded by several of my former colleagues at Google DeepMind and based in San Francisco. Skild AI features some CMU researchers. Generalist AI includes some of my former colleagues. I recently learned that Mistral has a robotics group.
A few other notable Western companies — there’s DYNA, which is looking to automate small tasks as quickly as possible. They’re essentially saying, “You’re all getting too complicated — let’s just fold napkins, make sandwiches, and handle other simple tasks.”
There are also groups your audience should be aware of, though we don’t know exactly what they’re doing. Meta and OpenAI certainly have embodied AI efforts that are rapidly growing, but nobody knows their exact plans.
In China, partly because of the trends we discussed and due to significant funding and government encouragement (including Made in China 2025), there’s been an explosion of companies seeking to make humanoid robots specifically.
The most well-known is Unitree with their H1 and G1 robots. But there are also companies like Fourier Intelligence, AgiBot, RobotEra, UBTECH, EngineAI, and Astribot. There’s a whole ecosystem of Chinese companies trying to make excellent humanoid robots, leveraging the Shenzhen and Shanghai-centered manufacturing base and incredible supply chain to produce the hardware.
When Robots Learn
Jordan Schneider: How do people in the field of robotics discuss timelines?
Ryan Julian: It’s as diverse as any other field. Some people are really optimistic, while others are more pessimistic. Generally, it’s correlated with age or time in the field. But I know the question you’re asking: when is it coming?
Let’s ground this discussion quickly. What do robots do today? They sit in factories and do the same thing over and over again with very little variation. They might sort some packages, which requires slightly more variation. Slightly more intelligent robots rove around and inspect facilities — though they don’t touch anything, they just take pictures. Then we have consumer robots. What’s the most famous consumer robot? The Roomba. It has to move around your house in 2D and vacuum things while hopefully not smearing dog poop everywhere.
That’s robots today. What’s happening now and what we’ll see in the next three to five years falls into what I call a bucket of possibilities with current technology. There are no giant technological blockers, but it may not yet be proven economical. We’re still in pilot phases, trying to figure out how to turn this into a product.
The first place you’re going to see more general-purpose robots — maybe in humanoid form factors, maybe slightly less humanoid with wheels and arms — is in logistics, material handling, and light manufacturing roles. For instance, machine tending involves taking a part, placing it into a machine, pressing a button, letting the machine do its thing, then opening the machine and pulling the part out. You may also see some retail and hospitality back-of-house applications.
What I’m talking about here is anywhere a lot of stuff needs to be moved, organized, boxed, unboxed, or sorted. This is an easy problem, but it’s a surprisingly large part of the economy and pops up pretty much everywhere. Half or more of the labor activity in an auto plant is logistics and material feed. This involves stuff getting delivered to the auto plant, moved to the right place, and ending up at a production line where someone picks it up and places it on a new car.
More than half of car manufacturing involves this process, and it’s actually getting worse because people really want customized cars these days. Customizations are where all the profit margin is. Instead of Model T’s running down the line where every car is exactly the same, every car running down the line now requires a different set of parts. A ton of labor goes into organizing and kitting the parts for each car and making sure they end up with the right vehicle.
Ten to twelve percent of the world economy is logistics. Another fifteen to twenty percent is manufacturing. This represents a huge potential impact, and all you’re asking robots to do is move stuff — pick something up and put it somewhere else. You don’t have to assemble it or put bolts in, just move stuff.
Over the next three to five years, you’re going to see pilots starting today and many attempts, both in the West and in China, to put general-purpose robots into material handling and show that this template with robotics foundation models can work in those settings.
Now, if that works — if the capital doesn’t dry up, if researchers don’t get bored and decide to become LLM researchers because someone’s going to give them a billion dollars — then maybe in the next seven to ten years, with some more research breakthroughs, we may see these robots moving into more dexterous and complex manufacturing tasks. Think about placing bolts, assembling things, wings on 747s, putting wiring harnesses together. This is all really difficult.
You could even imagine at this point we’re starting to see maybe basic home tasks: tidying, loading and unloading a dishwasher, cleaning surfaces, vacuuming...
Jordan Schneider: When are we getting robotic massages?
Ryan Julian: Oh man, massage. I don’t know. Do you want a robot to press really hard on you?
Jordan Schneider: You know... no. Maybe that’s on a fifteen-year horizon then?
Ryan Julian: Yeah, that’s the next category. Anything that has a really high bar for safety, interaction with humans, and compliance — healthcare, massage, personal services, home health aid — will require not only orders of magnitude more intelligence than we currently have and more capable physical systems, but you also really start to dive into serious questions of trust, safety, liability, and reliability.
Having a robot roving around your house with your one-year-old kid and ensuring it doesn’t fall over requires a really high level of intelligence and trust. That’s why I say it’s a question mark. We don’t quite know when that might happen. It could be in five years — I could be totally wrong. Technology changes really fast these days, and people are more willing than I usually expect to take on risk. Autopilot and full self-driving are good examples.
One thing the current generation of robotics researchers, generalist robotics researchers, startups, and companies are trying to learn from the self-driving car era is this: maybe one reason to be optimistic is that because of this safety element, self-driving cars are moving multi-ton machines around lots of people and things they could kill or break. You have people inside who you could kill. The bar is really high — it’s almost aviation-level reliability. The system needs to be incredibly reliable with so much redundancy, and society, regulators, and governments have to have so much faith that it is safe and represents a positive cost-benefit tradeoff.
This makes it really difficult to thread the needle and make something useful. In practice, it takes you up the difficulty and autonomy curve we talked about and pushes you way up to really high levels of autonomy to be useful. It’s kind of binary — if you’re not autonomous enough, you’re not useful.
But these generalist robots we’re talking about don’t necessarily need to be that high up the autonomy difficulty curve. If they are moderately useful — if they produce more than they cost and save some labor, but not all — and you don’t need to modify your business environment, your home, or your restaurant too much to use them, and you can operate them without large amounts of safety concerns, then you have something viable.
For instance, if you’re going to have a restaurant robot, you probably shouldn’t start with cutting vegetables. Don’t put big knives in the hands of robots. There are lots of other things that happen in a restaurant that don’t involve big knives.
One of the bright spots of the current generalist robotics push and investment is that we believe there’s a much more linear utility-autonomy curve. If we can be half autonomous and only need to use fifty percent of the human labor we did before, that would make a huge difference to many different lives and businesses.
Jordan Schneider: Is that a middle-of-the-road estimate? Is it pessimistic? When will we get humanoid robot armies and machines that can change a diaper?
Ryan Julian: It’s a question of when, not if. We will see lots of general-purpose robots landing, especially in commercial spaces — logistics, manufacturing, maybe even retail back of house, possibly hospitality back of house. The trajectory of AI is very good. The machines are becoming cheaper every day, and there are many repetitive jobs in this world that are hazardous to people. We have difficulty recruiting people for jobs that are not that difficult to automate. Personally, I think that’s baked in.
If, to you, that’s a robot army — if you’re thinking about hundreds of thousands, maybe even millions of robots over the course of ten years working in factories, likely in Asia, possibly in the West — I think we will see it in the next decade.
The big question mark is how advanced we’ll be able to make the AI automation. How complicated are the jobs these machines could do? Because technology has a habit of working really well and advancing really quickly until it doesn’t. I’m not exactly sure where that stopping point will be.
If we’re on the path to AGI, then buckle up, because the robots are getting real good and the AGI is getting really good. Maybe it’ll be gay luxury space communism for everybody, or maybe it’ll be iRobot. But the truth is probably somewhere in between. That’s why I started our discussion by talking about how robots are the ultimate capital good.
If you want to think about what would happen if we had really advanced robots, just think about what would happen if your dishwasher loaded and unloaded itself or the diaper changing table could change your daughter’s diaper.
A good dividing line to think about is that home robots are very difficult because the cost needs to be very low, the capability level needs to be very diverse and very high, and the safety needs to be very high. We will require orders of magnitude more intelligence than we have now to do home robots if they do happen. We’re probably ten-plus years away from really practical home robots. But in the industrial sector — and therefore the military implications we talked about — it’s baked in at this point.
Jordan Schneider: As someone who, confession, has not worked in a warehouse or logistics before, it’s a sector of the economy that a lot of the Washington policymaking community just doesn’t have a grasp on. Automating truckers and automating cars doesn’t take many intellectual leaps, but thinking about the gradations of different types of manual labor that are more or less computationally intensive is a hard thing to wrap your head around if you haven’t seen it in action.
Ryan Julian: This is why, on research teams, we take people to these places. We go on tours of auto factories and logistics centers because your average robotics researcher has no idea what happens in an Amazon warehouse. Not really.
For your listeners who might be interested, there are also incredible resources for this provided by the US Government. O*NET has this ontology of labor with thousands of entries — every physical task that the Department of Labor has identified that anybody does in any job in the United States. It gets very detailed down to cutting vegetables or screwing a bolt.
Jordan Schneider: How can people follow this space? What would you recommend folks read or consume?
Ryan Julian: Well, of course you should subscribe to ChinaTalk. Lots of great revised coverage. The SemiAnalysis guys also seem to be getting into it a little bit. Other than that, I would join Twitter or Bluesky. That is just the rest of the AI community. That’s the best place to find original, raw content from people doing the work every day.
If you follow a couple of the right accounts and start following who they retweet over time, you will definitely build a feed where, when the coolest new embodied AI announcement comes out, you’ll know in a few minutes.
Post turtle 这个典故，我以前在微博上也讲过。来美国后有好多年，没用中文写过正经东西，听说读写能力都有点退化。在微博上，我第一次讲Post Turtle这个典故的时候，直接翻译成“桩上乌龟”，听着比较直白，但有点粗俗。一位读者重新翻译了一下，说是“桩上龟公”。这个说法让我眼前一亮，简直做到了“信达雅”，“龟公”也是“公”，是对龟的尊称。用到“伟人”身上，要用尊称，符合古典汉语的表达习惯。
Login
