The US is losing its lead in AI

0:02Hello and welcome to The Verge Cast, the flagship podcast of The Model Wars. I'm your friend David Pierce, and today on the show, we're talking about the AI race ongoing between the US and China. For really the last several years, it's been pretty apparent that most of the best models and most of the most ambitious, most impressive AI has been coming out of Silicon Valley. Google, Anthropic, OpenAI, this handful of companies has been really driving the AI revolution. But then it has increasingly become clear that China is catching up quickly. We had a big moment a year and a half ago where DeepSeek all of a sudden showed up with a vastly more efficient, vastly cheaper model that was almost as good as some of the frontier stuff.

0:44And that became a big deal. Then just this past weekend, two different companies, Moonshot and Alibaba, made very clear that they have built frontier models that they believe to be just as important as anything else coming out of Silicon Valley. There is a race happening between the U.S. and China to win at AI, whatever that means. It's becoming a huge tech story. It's becoming a huge policy story. And I want to figure out what's going to happen if and when China catches up and even passes what's happening in the labs in Silicon Valley. So that's what we're going to talk about. Lauren Feiner and Hayden Field are going to come.

1:18We're going to do it. But first, here's everything else happening on The Verge today. This is 90 Seconds on The Verge for Monday, July 20th, 2026. The Odyssey is a huge hit. It made$264.1 million worldwide this weekend, which is both Christopher Nolan's biggest opening ever and one of the biggest movies of the year so far. I haven't seen it yet. I'm hoping to wait for a 70 millimeter IMAX screening, but those are apparently sold out like forever. In general, this has been a big summer for the movies. You have obvious wins like Toy Story 5 to surprise breakouts like Obsession, both great movies by the way.

1:51It's shaping up to be the best year for movies since before the pandemic, all the way back to 2019. And if going to the movies isn't dead after all, that might be really great news for Hollywood. And maybe bad news for the streaming services that we're pretty sure they'd already won this war. Meanwhile, the European Commission just levied its biggest fine ever against the platform AliExpress for failing to prevent illegal, unsafe, or counterfeit products being sold on the platform. Honestly, if you've ever been on AliExpress, none of that should surprise you. The commission specifically mentioned things like unsafe toys and dangerous cosmetics, and said that AliExpress has until October to come up with a plan to fix it.

2:28Timu got a similar fine for similar reasons this year. I don't know if the end of these ultra-cheap, relatively unmoderated shopping platforms is coming immediately, but there is a big fight coming, and fast. And finally, a leather jacket once worn by NVIDIA's CEO Jensen Wang was bought on auction for$960 ,000. I'm not saying this is the moment to call the top on AI and tech and the economy and everything, but boy is it getting weird out there, my friends. I don't know. You can read more about all of this at TheVerge.com. That is 90 Seconds on The Verge for Monday, July 20th. This episode is brought to you by Google Chrome.

3:08You think you know a browser, but Gemini and Chrome? That's new. It can help you with practically anything on the web, like restoring a vintage motorcycle from a 50-page restoration block or finally break down that long article you've had open for weeks. Gemini and Chrome is here for it. Ready to make anything online make sense? There's no place like Chrome. Check responses set up required compatibility and availability varies 18+. When you need to build up your team to handle the growing chaos at work, use Indeed Sponsored Jobs. It gives your job post the boost it needs to be seen and helps reach people with the right skills, certifications, and more.

3:43Spend less time searching and more time actually interviewing candidates who check all your boxes. Listeners of this show will get a$75 sponsored job credit at Indeed.com slash podcast. That's Indeed.com slash podcast. Terms and conditions apply. Need a hiring hero? This is a job for Indeed sponsored jobs. Joining me now, two people deeply qualified to make this make any sense to me. The Verge is Hayden Field, our senior AI reporter. Hi, Hayden. Hey, great to be here. And Lauren Feiner, our senior policy reporter. Hello. Hi. All of us are just like deep in China is the bad guy in bizarre and confusing and unknowable ways.

4:20So we're going to try to make sense of this along one very particular axis today, and that is AI. And Hayden, I kind of want to start with you because I remember you and I talked on this show a while ago about this sort of truism in the AI community that China is six months behind. that whatever happens here in the U.S., whatever new frontier models, China and its homegrown companies are running about six months behind the state of the art in AI. Is that still sort of the generally considered idea about U.S. versus China right now? Six months is basically the best case scenario in terms of, you know, the slowest they could possibly be.

4:58It may be that they're already caught up and it may be that it takes one month, but six months is like the max. That's kind of the consensus right now. And, you know, all of those types of hypotheses were basically made around Mythos and, you know, the cybersecurity aspect of things. So everyone said, you know, they're about six months behind Max when it comes to cybersecurity capabilities that Mythos and other models may have. It seems to be that it's even less for, you know, the models we use every day. For a lot of them, they're either equal now or they will be very soon. Okay. And Lauren, give me a sense of how much politically that is a thing people are talking about.

5:39I mean, we had David Sachs a year and a half ago as the White House czar trying to figure all this stuff out, yelling about like, you know, American AI supremacy. This is a thing Trump in particular seems to care about on cybersecurity grounds like the sort of U.S. versus China AI race. Is this relevant politically at this moment in the same way that it is to the people in the AI industry? Yeah, I think the U.S. versus China race and tech, AI specifically, is perennially a big topic in Washington. I think how it ends up playing out has changed over the years a little bit. You know, we started with the Trump administration kind of being more or less in line with the Biden administration on things like export controls.

6:24Those have kind of changed over time. So I think the way that it's gone about has changed a little bit. But overall, it's always something that I think a lot of people in Washington care about and care about keeping up with and beating China in this race. And Hayden, it seems like the big question right now is about distillation. Everybody's all up in arms about distillation. Anthropic is yelling to Elizabeth Warren about distillation. OpenAI has been complaining about it for like 18 months, it turns out. Everybody's very upset about distillation. Can you just explain what this thing is? And actually, it seems like kind of the main problem all of these frontier labs in the U.S.

7:03are looking to the government to solve. What is going on here? Yeah, that's a great question, especially because, you know, it's come up a lot in high profile lawsuits recently. We found out that Elon Musk's Grok distilled on OpenAI's models, according to the lawsuit. And his response was basically like, yeah, everybody's doing this. Exactly. And he kind of tried not to say it, and then he had to because he's under oath. It was a whole thing. But yeah, so distillation is essentially kind of having a smaller AI model or a newer AI model learn or extract, knowledge from a larger, more established one, a better trained one in order to, you know, become better itself.

7:44That's kind of the simplest way to put it. So it's something that, you know, smaller or newer AI teams, sometimes with fewer resources will do because it's an easy way to kind of make your model better very quickly and use fewer resources and just, you know, kind of speed everything up. So, you know, if you see an AI company come out of the woodwork, start really soon and then really soon after that, they have a new model and it's just amazing. Everyone's talking about it. A lot of times they use distillation to kind of, you know, bridge that gap. Essentially what happens is companies will collectively generate like, you know, up to millions of exchanges with, you know, another chatbot.

8:25So one example is Anthropic put out a statement recently that it believes that three Chinese companies, DeepSeek, Moonshot AI and Minimax all distilled on Claude. And it said that, you know, collectively, all three of them generated like 16 million exchanges with Claude or something like that from tens of thousands of fraudulently created accounts. And so basically, they just, you know, collect all those responses and, you know, use them to make their own model better. And whether it's for direct training or for reinforcement learning where, you know, the AI models may like learn certain decision-making techniques through, you know, this type of process.

9:05So yeah, it's basically just kind of a get-rich-quick scheme, but instead of getting rich quick, it's learning very quickly for an AI model. Right. It's like catch-up-really-fast shortcut. It seems like you can just basically ask the model questions, take the answers, and sort of reverse engineer your way into understanding how it works. That's why a lot of people thought that DeepSeek sounded a lot like ChatGPT at the time. Interesting. Including OpenAI thought that and I think continues to think that. Speaking of DeepSeek, Lauren, I'm struck by the fact that we like what beginning of last year, there was this DeepSeek moment where all of a sudden it came out that like China's tech is is here.

9:44It's moving quickly. China has figured out a way to make these models in a way that is much more efficient. It's much cheaper. And all of a sudden they have like competitive technology and the fact that it wasn't the very best model actually sort of became less and less important over time. And what has the government done between then and here? Obviously, we're in this like insane mess of AI regulations that we don't have nearly enough time to try and piece apart. But it does seem to me that, oh, no, China's AI is catching up and we think that's important is a thing people in the government have been saying a very long time.

10:16Have they done anything about this in a meaningful way yet? That whole moment kind of undermined this idea that But, you know, if we just make it harder for Chinese tech firms to access advanced U.S. chips, then we can help slow down China's ability to advance in this race. And if DeepSeek was able to get to where it was without, you know, needing the most advanced chips or with, you know, so much less compute, it really kind of undermined that whole theory. So, you know, I think we've seen, you know, a different approach to export controls. The Trump administration at one point allowed for, you know, greater sales of chips to China.

11:06Now there's some bills in the NDAA that could strengthen certain export controls on chips to China. So I think there's still not a ton settled about what's the best way to go about this. But I think it also makes clear that, you know, just cracking down on chips exports might not be the only way to or might not be as effective as maybe we once thought it was to slow down China's ability to compete here. Yeah, I've enjoyed doing the research on this subject because on the one hand, you get a bunch of people who are like, well, it's China. So there's essentially nothing we can do. And this is sort of a long running thesis that actually all of our legal and regulatory fights with China really don't ever amount to anything.

11:53So just full nihilism, who knows what will happen. And then on the other side, I've heard a theory that I think kind of borders on conspiracy theory that what we've seen, like with Mythos recently, where they're actually taking steps to not release these things to the public in the same way, is actually a way of keeping them out of the hands of these companies that want to do distillation. My sense is none of that is actually what's going on. And that, in fact, what's happening is everybody is still desperately trying to figure out how to do anything here. Is that fair? Or is there is there some four dimensional chess thing in progress, do you think?

12:26I mean, it's hard to say. I think I feel like I tend to agree that we don't really it's hard to know what any of these actors are going to be doing in the next six months, one year. you know, we're really at the start of this race. And it's just becoming more and more clear how many different ways all these companies are going to have to compete. And that, you know, the path to getting to the finish line here, so to speak, could look a lot different than we once thought. So it's really hard to predict or really know, like, who really has the best strategy here or what's going to pan out. So I feel like everyone is kind of making their bets on what's going to work and it'll remain to be seen what actually does.

13:16Okay. So I have the same question for both of you that I want you both to answer from sort of very different perspectives of this question. And the question is essentially, what happens if China starts to win, right? I think at this point, Hayden, like you said at the very top, we're at this point where China is either a little ways behind or barely behind, but I don't think there's anyone who would really try to make the case that China's AI, either its products or its frontier labs or its models are ahead. It doesn't strike me as completely inconceivable that that might change. So Hayden, you spent a lot of time talking to these companies, talking to the people who make the products.

13:52We're in a weird economic moment with all of these AI companies. Like in theory, if all of a sudden the next deep seat comes out and it is it is by obvious measures the best AI product anyone makes anywhere. What happens? That's a great question. And I think that the fact that we have no idea is what's fueling a lot of these fears. I've seen a ton of OpenAI employees leave the company and write huge manifestos about their fears about what will happen if China overtakes us. No one's really been super specific about what that actually looks like besides, you know, potentially just data privacy issues and just giving China more power on the world stage.

14:37Lauren, call me crazy. This sounds like TikTok. Like we just did this with TikTok. You have been on the show talking about this with TikTok, right? Like this idea that if we let China get ahead, It will gain this economic and sort of cultural soft power that will make it take over the world in some unknowable but problematic way. Like, are we just doing this again with AI? It's a decent analogy. And I could think of it as like, what would happen if, you know, one of these Chinese AI firms had indisputably the best AI product and everyone had to use it? You know, it's like if Huawei indisputably had the best phones out there and everyone had to have them, but we can't because of restrictions here in the U.S., you know, that would pose a major issue and a major question of how should policymakers deal with a product like that.

15:32And I think it would just bolster the arguments from tech companies who say, we really need, you know, a long leash here to do what we need to do to innovate because we have to beat China. And I think if China got ahead in whatever way that looks like, that would, you know, maybe make some policymakers think twice about putting additional restrictions on U.S. tech firms. I think this is really making the case for diversification. And it's why, you know, essentially, yeah, if a Chinese AI firm indisputably has the best model, clearly that has a lot of trickle down effects. And that's what the U.S.

16:12is worried about. War wise, that has a lot of effects, surveillance, everything. And so, you know, U.S. AI companies would quickly have to catch up and the U.S. may be, you know, theoretically using a second best model. And is that going to make them second best at everything? So I think that's what they're most worried about here, like Lauren said. And it's hard because, yes, these are all real fears. But also we've seen these models leapfrog each other every week. So it's like if China was the best for a certain amount of time, would we really not? But it's just hard to predict one country coming out on top forever when we've seen so much change happen stateside every day or every week.

16:58Support for this show comes from Keeper Security. Some threats are impossible to see, like when it comes to your data. Luckily, there's Keeper. Keeper is a password manager that creates strong, unique passwords for all of your accounts, stores them securely in one place, and logs you in automatically across all your devices. So you never have to remember, guess, or worry about your login credentials again. What people really notice is how simple it feels. Keeper provides security features like autofill and face ID to make things easier. From individuals to large organizations, millions of people worldwide rely on Keeper to keep passwords organized and accounts protected.

17:37Right now, Keeper is offering our listeners 60 % off personal and family plans at keepersecurity.com slash verge. This offer is only for podcast listeners. That's keepersecurity.com slash verge for 60 % off personal and family plans. Make sure you use our link so they know we sent you. Keepersecurity.com slash verge.

18:02Support for the show comes from MongoDB. AI-assisted and agentic coding can help you build faster than ever. But if your data layer is a bottleneck, what's the point? MongoDB actually gets out of your way. MongoDB is a unified AI-ready data platform that empowers you to build scalable, generative AI applications. It eliminates the need for separate, specialized vector databases by combining a flexible document model natively with semantic vector search, full-text search, and real-time operational data. Instead of wrestling with rigid schemas or translating data formats, MongoDB's native data model mirrors the language LLMs already speak.

18:44Plus, MongoDB gives you the flexibility to ship at the speed of AI. The ACID compliance lets you sleep soundly at night and scales to handle massive Fortune 500 workloads. Developers have a word for that kind of reliability. Actually, five words. It's a great freaking database. Start building at mongodb.com slash AI. Pros trust the Home Depot for heavy-duty storage solutions for any job site or garage. Right now, get up to 15 % off select storage and organization. impact and water-resistant totes, and shelving built to hold up to 2 ,500 pounds. Storage systems have space for all your tools and protect them in the garage, on the job site, and everywhere in between.

19:27Save time and maximize efficiency with adjustable shelving customized to your business's needs. Shop and save on ProGrade Storage at The Home Depot. How pros get more done. This episode is brought to you by State Farm. Listening to this podcast instead of doom-scrolling? Smart move. Another smart move? Getting help from one of State Farm's 19 ,000 local agents when you choose to bundle home and auto. Bundling. Just another way to save with the personal price plan. Prices are based on rating plans that vary by state. Coverage options are selected by the customer. Availability, amount of discounts and savings, and eligibility vary by state.

20:07I do wonder, though. So I think it seems like at least so far there has been great pride both in Silicon Valley and in D.C. about the fact that most of the best technology here is coming from the United States. Right. That the ones leapfrogging each other, by and large, are Google, Anthropic and OpenAI. Right. They're all U.S. companies. Inside of the United States. Right. There's not even really a company that outside of kind of occasional deep-seeky blips has launched itself even into that stratosphere. In the way that like in a lot of other tech, there are roughly equivalent Chinese companies to a lot of the stuff that we have in the United States, right?

20:47So there is a sense that you can have a Chinese version of things and you can have an American version of things. this idea that the American version of AI is way ahead seems like economically and politically very important at this particular moment in time. And I guess from a from a just pure sort of American exceptionalism standpoint, Lauren, is there something to that here where it's like it's actually it is like weirdly important to America's concept of itself right now that we are winning at AI because of all of the other stuff right now that is so tied up in us winning at AI? Yeah, I mean, I definitely think it is, but I think it's not just this like theoretical, like, we just want to have like patriotism about our US AI companies.

21:32I think there's also a practical element of it. And I think the reason I jumped to kind of Huawei as, you know, a potential example is like, you know, AI, like Hayden is saying, is something that's going to be built into so many different technologies and systems. And I think whose AI is built into those systems is a big, important geopolitical question. And, you know, if you think about kind of telecom infrastructure and, you know, the fight over different Chinese telecom infrastructure around the world and issues that have come up with that in the U.S., I think that's maybe a good analogy here when we think about, like, what kind of AI model is Europe going to be integrating into its health systems, weapon systems, whatever it is.

22:24And, you know, if China has the best model to use on, you know, across many different use cases, maybe they have no choice but to use that. So I think there is something there that goes beyond just kind of like the national pride of like who is going to have the best model that'll be built in across the world and, And, you know, what sort of privacy surveillance issues might that carry with it? Hayden, does this make the American frontier labs more or less powerful in this sense? Because I think about all this in the context of Anthropic and this giant fight with the U.S. government about what it can and can't do and who is and isn't in charge of the AI models.

23:07And you sort of alluded to this a minute ago that what the tech companies are saying is you need to give us a longer leash. You need to give us more room to run because that's how we win. You can't hamstring us. We have to go win. Does this maybe bode well for these companies who are like, trust us, we need to do what we need to do and only you can get out of our way? Yes, I think that's exactly what's happening here. No matter what happens with this power struggle, we've seen the Trump administration kind of try to take power back and then give them a longer leash. We've seen a lot of back and forth here.

23:38But ultimately, the tech companies do have a ton of power because the government's biggest fear is China overtaking us in AI right now. We've seen a ton of executive orders, memorandums, things coming out about how speed wins. And we must, at all costs, be the fastest and the best. And so I think that, yeah, ultimately, the AI labs have a lot of power here. They're saying, look, you know, we can work with you. We will deign to voluntarily, you know, abide by some of these frameworks you've created. But at the end of the day, you must let us do our thing and kind of just turn away until we're done because that's the only way we're going to win.

24:17And that is something the government seems to really believe and will let happen. Interesting. Yeah, it's very funny to put that next to some of the other parts of the AI business where increasingly these labs are like, please, dear God, regulate us. Because we don't want to be held accountable for all the stuff that happens on our models. We want you to tell like Demis Asabas coming out and being like, we need it. We need an AI watchdog that reviews all of the models. That's a probably a reasonable idea and be a really easy way for the companies to absolve themselves of a bunch of responsibility for the outcomes of their models.

24:49But then to point that at like the defense contractor side of these companies who are like, you know, we have to be able to run as fast as we can and do everything that we can because we have to beat China. It's like we've we've spent a lot of time, all three of us in our professional lives, listening to people explain why you have to do wild things with technology in order to beat China. And I don't know, maybe it will be more successful politically and, you know, as in product terms than it was last time. But that argument rings a little more hollow to me every single time I hear it. Lauren, what do you make of the political moment that we're in?

25:23We're heading into the midterms. Everybody's really mad about data centers. We are in a moment where these companies are hugely powerful. They're potentially about to go public. There's a lot of money about to be had. Are we headed toward U.S. versus China AI either becoming or a bunch of people trying to make it a huge story politically over the next few months? I think it's always something that tech companies are going to try to use to wedge some supporters into their camp. So I do definitely expect to hear that narrative. I think how successful that is could depend in part on like, do we see a big moment happen where China feels like it's getting ahead in some way?

26:06Because right now, you know, like you said, we have all this backlash to AI data centers. There's a lot of just fear around AI. And, you know, do we even want what we're seeing it being made into? Do we want the sort of consequences of it? So I think that's made it hard to push forward proposals that would, you know, just like shut down regulation on AI. But at the same time, I think if we had a big moment of like, oh, man, are we really behind here? I think that could put some momentum behind some policymakers who say, no, we really do need to kind of just let these companies do their thing. It is true.