China vs USA: Who Will Win the AI Race?

Thanks for listening to The Rest is Politics. To support the podcast, listen without the adverts and get early access to episodes and live show tickets, go to therestispolitics.com. That's therestispolitics.com.

This week, Matt Clifford and I are back with another episode in our AI series. So we're going to ask in this episode, who's really winning the AI race? Is it the United States? Is it China? Is the UK even in the running? And we also explore what technological rivalry could mean for global politics in the years ahead. We're joined by Tino Cuella, president of the Carnegie Endowment for International Peace.

Someone right at the heart of how governments are trying to grapple with fast-moving technology and global power shifts. Tino's a former justice on the California Supreme Court, one of the most respected voices on international cooperation. And he's really well placed to talk about how AI is reshaping the world. Here's a taster. If you like what you hear, you can listen to the full episode by signing up at the rest is politics.com.

If I look at the world between 1950 and 2023, one of the most dramatic and important changes is life expectancy went from 46 years to 73 years. Literacy has gone from like less, like 55% to closer to 80%. Infant mortality has gone down. This is all a function to my mind of science and international cooperation. And right now, those kinds of questions are channeled into...

What does this technology do? How does it work? How can people benefit from it? So my point is simply like, those are hard questions. How do you get India on board, Germany, Japan? South Korea, the US, Mexico, all these disparate countries to figure out how to use the technology smartly, but at the same time deal with the risks. And my point is simply that all the tumult in the world right now, whether domestic or global, like the invasion of Ukraine, is taking...

up a lot of oxygen and making it a little harder to focus on these questions. I think you can

One thing I'm learning as we do this podcast series is my job is to sort of be the anti-Rory. So let me give a very different framing, which I'm not sure I fully believe, but I'm going to try. I think if we all try and channel our inner Xi Jinping. I think you could tell a story that we really see over the last two years, extraordinary continuity. You know, like the idea that there's been this like...

sea change in US policy on AI, I don't think would register in Beijing. I think that at least people around Xi would say something like, the US looks at the inevitable rise of China and it says, We need to find some way to decouple the...

economic growth of this much larger country, that if it gets even close to our level of prosperity on a GDP per capital basis, will dwarf our GDP. We have to decouple that GDP from real world power. And what the successive US administration- of both stripes have done very successfully, say...

ai is one of the core ways that we can achieve this historically unprecedented hoarding of power despite demographic destiny and therefore we're going to have export controls you will not be allowed to build powerful ai and we will accelerate our companies and we will make sure that we win. What do you think?

I think there's a lot to that. And actually, I'm not sure you're the anti-Rory, you're the kind of like Rory compliment. That's a nicer way. If I reconcile the two perspectives for you, I think- This is the foundation that does peace. Right. But here's what I mean by that. You're right that- a high level of generality, at the end of the day, does the US want to be the best at Frontier AI? Absolutely. That has not changed from...

Biden to Trump. And that highlights that even with all these partisan changes and changes in style and leadership, there's like a geopolitical logic at work here. I will say just briefly on that, having negotiated with both administrations on this topic. I would say the similarities are more striking than the differences. Okay, that I want to explore even more. But I think you were rightly worried.

pointing out, like, I'm bringing in this perspective of like, oh, this technology can have a big impact on people's lives. It can be positive, like it should be shared. In a way, the hard question from my perspective is, if we're intellectually honest, how can we reconcile the reality of competition at the company level and at the geopolitical level? with the idea that those who are doing the competing will often, and I think in an authentic way, they believe what they say when they say,

I want to find some way to bring this technology to many, many people and to have it be positive in their lives. I didn't think that's what America First is about at all. I didn't think that's what Trump thinks about. I didn't get out of bed in the morning thinking, how can I share this wonderful technology?

with Africa. How can I make sure that our European... I'm talking more about the entrepreneurs that are trying to develop the technology. I'm thinking more about the people who say, so let's take the CEO of a major tech company. What... What message does that person deliver if that person were on your podcast? That person would say, look, we're in this because we think it'd be good for humanity.

And I think they mean it at some level. But how to reconcile that notion of diffusing it broadly with the fact that at the company level, obviously, they want to hold tight some of their commercial advantage. And at the geopolitical level. sharing doesn't necessarily mean let's let everybody copy our technology. Well, not just copy. I mean, at the national level, Trump has demonstrated in Ukraine that he can instruct private companies to switch off their satellites and intelligence sharing.

So these companies are not sovereign. If Trump said, You cannot share the frontier large language model outside the United States. They would not be able to share the frontier large language model outside the United States. They may themselves have fantasized that their whole game was helping poor children in Africa. But if you have an administration that wants to say, for the first time, remember what we're seeing is that so many things that we took for granted that America shared.

Its nuclear umbrella was an umbrella not just for the US, it was for other people. Its amazing currency, being the world's reserve currency, was supposed to be part of a global system, right? If you suddenly have a president who wakes up one morning and says, actually, very cleverly, over 80 years, we've become a parasite inhabiting the entire body of the world and they're completely dependent on us, then you can weaponize that dependency.

But see, here's what I think is the core distinction in how we see it. I'm saying the people developing this technology say they want to share it. You respond and you say Trump. And I think that's fair because obviously he is. a pivotal figure in how my country is perceived. But I view the United States as much more than one. And I think in some ways, what's interesting about the story of American tech policy, to my mind, is that however it may look, it is not entirely.

controlled by any president or by any cabinet secretary. I mean, if anything, I would say many critiques of tech policy in America is that it's too controlled by the private sector. So to my mind, I would say The whatever Trump specifically wants to do, the reality of what ends up happening with tech policy is a sort of, I don't want to call it a negotiation exactly, but it's a pluralistic fight a little bit between contending interests in which the private sector.

for better and maybe not better, is a key player in it. I think we should ground this in what the administration said publicly, because I think there's a really interesting thing here, which is that there are... some things where the interests of the administration and the tech companies are really clearly aligned.

And then there are some things where they're not. So when J.D. Vance flew to Paris in February of 2025 for the AI Action Summit that President Macron hosted and gave a speech, it was fascinating. It was like a genuinely... On its own terms, excellent speech in that it grabbed what he wanted to grab attention. And he said four things and I remember them very clearly. And there's a lot of tensions that run through them that I think illuminate a lot of.

sort of these different views. So the first thing he said is, the future of AI is American models running on American ships. Full stop. Okay, that's interesting. But the other three fit together in a really interesting way. The second thing he said is, no one is going to regulate. these companies other than us. Other than the US. Other than the US. So it's going to be American models, American chips, American regulation. Third thing he said was, and by the way,

Some of these big companies are bad actors who are doing bad, woke things in our country. The fourth thing he said is, and by the way, Labor is going to have a seat at the table. There will be no discussion of AI policy without Labor having a seat at the table. Okay, so- American chips, American models, American regulation, but regulated with MAGA views on woke and MAGA views on labor participation. Yes. The core thing I want to draw attention to here though is that...

The role of big tech in the Trump administration worldview on AI is extremely ambiguous. On the one hand, they are the vehicle through which America will be able to dominate AI in the world. On the other, they're a somewhat...

dangerous force within the U.S. And certainly even within the MAGA coalition, even before you get to the U.S., I think Matt's point is the more effective answer to your point, Rory, about, well, what if Trump just simply said, like, you can't choose models? And I think Matt's point, as I... interpret it rightly is the vice president gave a certain speech, but it's a separate question like what would the United States do and what would the tech community do if the president said,

You have a set of financial interests and I'm just not going to care about it. Let me actually even take the NVIDIA chips as a better example because I think it's...

clearer and starker and remind listeners right about nvidia ships so so the most powerful ai runs on a very specific and special set of chips and that market though it's changing is dominated by a particularly

valuable American company. In fact, it's become by market cap for the moment, the most valuable company in the world. Remind us, how does this relate to Dutch companies and Taiwanese companies? Because even though an American company is designing the most advanced chips and pulling together the financing, the designing, the marketing that allows those chips to basically diffuse.

It is a Taiwanese company that's actually manufacturing the designs and shipping them out from Taiwan primarily. And it is a Dutch company supported by German technology as well that is designing the most extraordinarily sophisticated lithographic machines. that bend light in a particular way and let you actually turn those designs into actual chips. And yet, because it's going to be central to where we go, somehow

the Dutch and the Taiwanese don't feel fully in control of this. They don't feel that they can blackmail America. Because the reality of the- manufacturing around the tech stack is complex. It involves a ton of capital. It involves the geopolitical power of a country that also has the design control. So when you think about IP, design, marketing, brand, tech stack, capital. The actual machines that let you do that are only one piece of a larger puzzle, right?

But by no means are they inconsequential. I mean, if you talk to Germans about challenges with the German economy, they'll happily tell you, well, at least we play a role in that supply chain. But how can we learn from that and leverage it more? Just to give listeners a sense of how extraordinarily sophisticated those are. machines are. One expert recently told me that if you took the mirrors in these machines,

and blew them up so that the mirror was actually the size of Germany, there would be no blemish on that mirror bigger than about a centimeter or an inch. Can you imagine that? That's human ingenuity in action. That's why I kind of referred to life expectancy.

going forward. But now what do we use that technology for? And that's the question. And I think what Matt was saying is that if President Trump were to say like, okay, NVIDIA, you can't sell any chips outside the US of this kind, we're just going to have them all go to US companies.

My point is that would create an enormous backlash that would be not easy for the White House to deal with. And this is not purely hypothetical because, of course, Biden passed the Chips Act. But let's develop this a little bit because it's a very interesting case study, right?

My understanding of it is Biden basically said NVIDIA can't sell the top of the range chips to China. They can sell a sort of second band. And then when Trump came in, there was some back and forth with the CEO of NVIDIA. And then you get Dario from...

saying, oh, what have they done? You're selling chips which are far too powerful to China. Tell us a little bit about this story. So I think the export controls on these chips have been one of the, certainly the most prominent levers on trying to In fact, what I see as this sort of...

hold China off being able to develop the most advanced version of this technology, which, by the way, I think is pretty clear it's been successful. Tell us a little bit about these controls. It's exactly what you said, Rory. So, you know, NVIDIA make a range of these GPUs, these chips.

Obviously today, their core focus is on making the top end of those chips as powerful as possible to meet the needs of the largely USAI companies that are training the models on these chips. What the Biden administration said was there's a type of chip, there's going to be a specification over which you cannot...

go if you want to sell to China. You can sell them old chips, but you can't sell them new chips crudely. I think this has been very successful, by the way. I think lots of people disagree with this, but it's very interesting. In January, February of this year, there was this-

big moment the deep seek moment which you know i think we really need to talk about it's really important for this topic so what deep seek what is deep seek deep seek basically was the first chinese model that was released that I think captured the imagination of the Western public, partly because it was one of the first reasoning models, these models where you actually see the chain of thought and what it's doing, and partly because it felt almost as good.

as the top American models. And they also claimed that it was cheaper, required less energy. Some reports very strongly and plausibly argue that part of what let the DeepSeq model get trained so cheaply is to exfiltrate capabilities to distill knowledge from American models and to incorporate that into the training of the DeepSeq model. So that means that if you just look at how much DeepSeq tells you it costs to train the model.

it doesn't tell you the whole story. Tino being much more diplomatic than me is basically might be implying that they stole some of this stuff. Yeah, there are credible reports to that effect. But they continue to insist that actually running the prompts is cheaper and less energy intensive than their... Yeah, so there were two things that I think caused a sort of panic here. One was...

Nvidia stock temporarily took a nosedive because people said, well, maybe compute is not as important as we thought. If these guys didn't have the chips and they were able to do this, then do we really need Nvidia as much? It's interesting that Nvidia is worth a lot more today than it was then. But the other thing that I think is really interesting is that it casts doubt on whether the whole policy of export controls...

was the right one. And a lot of people, including very prominent people in the Trump administration said, what we should have been doing is not stopping China getting these chips, but actually getting them dependent on these chips. We should have had NVIDIA. basically build out dependence from DeepSeek and others on these chips. This is called the NATO model. This is make everybody else dependent on you and then leverage that.

Yeah, absolutely. And there's a lot of people, people like David Sacks, who's the AI czar in the White House, said, look, we just made a mistake here. Now, it's actually settled into this slightly weird Frankenstein's monster. What is the policy now? Well, the policy now is to still have some restrictions on certain kinds of memory chips that work with the semiconductors, but to be more flexible and not have these tiers that restrict chip exports as much.

of nvidia is endogenous to that like it reflects the reality that if you're able to not only make the most advanced chips but to sell them pretty much anywhere in the world that's worth even more right but i think To your point about what it's settled into is like, I don't think we've seen the last of these debates because at the end of the day, in the US right now, I would say the zeitgeist of the policymaking in the Trump administration.

tries to reconcile two points, but that requires an assumption. One point is, let's have the US be the best in the world and maintain its advantage. Two is like, we will best be able to do that if we make China dependent on these chips. And then we always have the optionality of what we do with that. But in the end, if we get the American tech stack to dominate, that is going to preserve American advantage. If you enjoyed that, you can listen