Author of *The Bitter Struggle* on US-China AI competition: chip export policy is bipartisan and China is not waking up yet

Speaker 1: We have Teddy Collins, the author of the bitter struggle on his new book, US China AI competition shaping policy inside the White House. Teddy, welcome to the show.

Speaker 6: How are

Speaker 1: you doing?

Speaker 3: Doing well. Thanks so much for having me.

Speaker 1: Thank you for hopping on. First

Speaker 2: question, did you orchestrate this takedown to promote your book? We're asking all of our guests that today. I mean, It's just too convenient for you, Teddy. I don't know. It's hard to it's hard to...

Speaker 3: I, you know, I'm I I am following the news like everyone else.

Speaker 1: Yes. Yes. So, take us through the thesis of the book. Take us through the process, how long you've been working on this, and I'm sure we'll have a bunch of downstream questions, because we've been debating all show. Is China waking up? Is the relationship between The US and China going to change? And if so, how?

Speaker 3: Maybe I'll start with the the book process because, I mean, for those who haven't been been part of writing a book before, it's... The publishing process just takes very long time.

Speaker 1: Mhmm.

Speaker 3: And given the speed that AI moves at, that that poses quite a challenge. We had to lock down the text to, like, basically near the beginning of this year, which meant almost all of this was written in 2025. In most domains, it was kind of okay. Yeah. But, obviously, a ton has happened in between when we finished writing it now. So this was, like, pre mythos. This was pre hugging face, OpAI incident. Yeah. And so it's one of these things where where trying to write something like this in a forward facing way that remains current, can be quite challenging. I think that most of the sort of core themes that we had remained pretty relevant, but anyone who's picking it up today will probably be wondering, well, why isn't this incident mentioned? Why isn't that incident mentioned?

Speaker 1: Yeah.

Speaker 3: In terms of the core message of the book, in brief, you know, my coauthor and I worked in the White House during the Biden administration on a number of AI issues. And this will sound crazy today, but in 2021, when we ended the administration, AI was not a thing that most policymakers were really paying attention to.

Speaker 6: Yeah.

Speaker 3: And so it was pretty greenfield in terms of thinking through how The United States should tackle this set of issues. In the book, we try to take a lot of those considerations in terms of kind of the intersection of technical issues and policy issues Mhmm. The types of things that we previously presented to the national security adviser, the president, etcetera, and present them instead to the general public because there are, I think, very, very difficult trade offs that societies, especially democratic societies, will have to take to account. Yeah. In the book, we break it down into sort of three main struggles that are the core structure. The first is the struggle for democratic nations to establish and maintain a leading capabilities. Yeah. The second is the struggle for governments to understand and make use of this technology effectively because right now, as you talked about on your show before, there's a big gap between the private sector and the government. And the third is the struggle then to make sure that the technology doesn't become deeply unsafe and doesn't take us away from democratic values once we've melted it.

Speaker 1: Yeah. Yeah. It's it's it's very it's very... It's such a complex situation with what's going on with NVIDIA, but, there still is this, like, idea in the back of your mind where, if you're making a thing, why is the government telling you that you can't sell it to somebody? Like, that's a new thing and the conversation has completely changed. The capabilities have changed. I have updated, but I, at the same time, I do understand that if I make a soda, it's kind of like, am I... This America. It's land of the free. I should be able to freely sell it wherever I want. This is a special case. It requires extra attention. On China, I'm very interested in this question of China waking up. There's been this prediction in 2027 that China will wake up this year. And maybe that happens, maybe that doesn't, but I'm interested in the the philosophy that China brings to bear with the AI race because we've been tracing back through the five year plans. China has been investing in indigenous chip supply for thirty years more. Yeah. That's right. And when I saw Dwarkash Patel post his AGI predictions of who wins, he has these probability densities and he says, if AGI comes soon, America wins. If AI comes later, China wins. It was... It felt like it was an accurate prediction, but it was also just a reflection of the strategy of China that they think that, like, slow and steady wins the race. That's always been the the philosophy, and that's why we didn't see a Taiwan invasion in the last two years. Yeah. It's like, yeah. We'll get there.

Speaker 3: Yeah.

Speaker 1: We'll we'll... When it happens, it happens. I'm Xi Jinping. I'm gonna live till the 150. I'm not going anywhere. I'll take care of it when I take care of it. But can you give me more nuance with that? Do you agree with that? What do you think China is thinking around the AI race?

Speaker 3: Yeah. Maybe first, I'll I'll speak to the the... Your point about, like, if you make a Coke, like, you know, why should you be able to sell it anywhere, which I think is, like, a very, like, a very fair question. Yes. You know, as as as my colleagues in government will testify to, like, I honestly took some convincing on some of these policies because my instincts are also pretty laissez faire. Yeah. I do think there are some areas that are exceptions. You have some spectrum from, like, Coke on the one hand to... I think very few people would say that, let's say, during the final days of World War two or during the peak of the Cold War, The United States should have been selling technology relevant to building nuclear weapons to the Nazis or to the Soviets. And I I I think almost everyone would agree with those two points. And then the question is, like, along that spectrum

Speaker 2: Where's the

Speaker 3: exactly do you draw the line?

Speaker 1: Yeah.

Speaker 3: And that is the subject of innumerable, you know, discussions in the situation room with all of these cabinet secretaries who are, like, now trying to pour over the details of, you know, well, how many nanometers are the nodes, you know, in these chips versus those chips and so on. And and it's tough. It's tough. You know, we we did our best. We, like... I stand by kind of, like, the places that we drew lines in retrospect to, you know, maybe there were some things we did on slightly differently, but, it's it's it's a question of degree not absolute principle, I would say. Mhmm. And, yep, in the book, discussed kind of in detail, like, do we think that various types of chips fall in different categories and so on? To your to your question about, like, China waking up, I mean, this is gonna be a very unsatisfying, like, response, but, like, obviously, it's it's very hard to tell in part because China is at the highest levels of government, just, like, much much more close than United States. So they don't have their debates as much in the open. And as a result, it's it's a... There's a lot of kind of... You know, this term people used during the Cold War of a Kremlinology, trying to read the tea leaves in terms of, like, who's being invited to present things to the Politburo Standing Committee and so on. I think the short version is if you look at the statements from Chinese policymakers and from Chinese industry leaders, there is definitely, like, more interest in these sort of fantastical and sometimes terrifying future scenarios, AGI, ASI type situations than there was a couple of years ago. I think it's much less prevalent than it is in The United States, especially if you look at the tech industry. There are some people like,, CEO of DeepSeek, talks about AGI a lot, like, done since DeepSeek was founded. But, yeah, the the, like, the modal discourse is much closer to AI as a significant, but on the whole, like, somewhat normal technology. Mhmm. And so this question of, like, what does it take to get... You know, to have senior policymakers, like, really wake up

Speaker 6: Mhmm.

Speaker 3: I think it's very, very, very hard to say. Very hard Yeah. To

Speaker 1: How much of... How much do you agree with this opinion that China is more in favor of open source AI because it's commoditizing their compliments. They are an industrial powerhouse, an engineering powerhouse, manufacturing powerhouse. Their economy is not driven by services. It's not driven by software engineering, lawyering, financial work. And so this will actually make them more competitive if there is all of a sudden an infinite pool of labor, an infinite number, a country of lawyers in a data center or something that they can access. And then secondly, they have an incredible jobs program. They will just they basically have UBI if they need because they'll just build more houses and trains to nowhere. And the the average citizen in China is optimistic about AI because the government will take care of them if they lose their job.

Speaker 3: I think I'm a little bit skeptical. Okay. A couple reasons. One reason is that at the end of the day, like, someone is... Someone still has to foot the bill Mhmm. For using these systems, whether they're open source or not. Like, they still take some amount of energy to run. Right? And so this is why, like, a lot of people choose to use models, whether they're open source or closed source, like, most people aren't hosting them, like, on their own infrastructure or something. Right? So you're, like, you're still paying for, like, these multiple levels of the stack. Yeah. I think the other thing that's tricky is that the, like, the... Obviously, unemployment issues that China faces are, like, very significant. So I think the idea that these things will... And and, like, the state can only pay for so many of those cranes to nowhere. Like, eventually, the bill comes due in some capacity. The final thing I'll note is that on this point of sort of the enthusiasm for AI in China versus The US, I think that that is directionally correct, but there's some really good work, especially from Jeff Ding, trying to unpack that. I think that the gap is less significant than... It it... It's like an easy story to report on. It's like, look at this survey result, you look at that survey result. Yeah. It's often kind of apples to oranges. Sure. And I think that that makes it hard to say, like

Speaker 1: Yeah.

Speaker 3: Okay. If you did, like, in the exact same conditions, you took, like, the median Chinese person and the median American, and you said, like, what do you actually think about this? What if it takes away your job? What if it poses bio risk, etcetera? I think that there would be, like, a decent amount of convergence.

Speaker 1: That's interesting. I have I have... Jordy, please.

Speaker 2: In in all your studies, have you been able to find a single critical industry where China said we're happy to let the the The United States or its allies just just handle this one?

Speaker 3: I am not aware of any.

Speaker 1: Podcasting. They don't have a Joe Rogan. That's right. That's right. They don't have a Joe Rogan.

Speaker 2: They're like, we'll let you guys keep

Speaker 1: I mean, quite ironically, it's free,

Speaker 3: like, speech. You know, the the next talent that they're gonna try and... That they're gonna try and

Speaker 1: win I don't think they can do it. I mean

Speaker 3: find the next Joe Rogan.

Speaker 1: They can't win in free speech. Like, that is

Speaker 3: That is... Yeah. That's the enduring... Like, advantage of democracies.

Speaker 1: Yeah. Yeah. Home defense maybe, civilian gun manufacturing. These are things that the Chinese economy is not excited about building up anytime soon, I suppose.

Speaker 3: That's right.

Speaker 1: Can you give me your high level understanding of the level of China hawkishness on the left and the right currently? Because when I think hawkishness, I think I think that's more of a Republican issue, but it feels like chip sanctions happened under Biden. There were... There was trade war under Trump, but Yeah. At the same time, we saw a lot of people from the Trump administration advocating for chip sales to China. And then post midterms were in this new revelation. And so is... Are are the Democrats more anti selling GPUs to China? I I haven't really been able to parse that out from just the data center debate, which is so much more narrow.

Speaker 3: Yeah. Yeah. The other thing is, honestly, the events of recent weeks

Speaker 7: Yeah.

Speaker 3: Make it even even more complicated because if I think this sort of perception of the the tech rights and what that means in terms of, like, how Democrats react to that.

Speaker 1: Totally.

Speaker 3: So I think this is all pretty fluid. I think that your characterization is has been historically accurate, and this is one thing that has kind of been shifting more recently. One thing that's worth noting is that, the the expo controls, like, on chips, actually, the sort of sanctions began in in Trump one.

Speaker 1: Oh, that's right.

Speaker 3: And and then there was kind of this sort of sort of steady, layering

Speaker 6: on top

Speaker 3: that happened during Biden. And some have been been pulled down during Trump two, but, like, the the majority of them, like, core of them have stayed on. So there are a lot of interesting dimensions here. One is that, it's an area of, relative bipartisan agreement in congress.

Speaker 1: Yeah.

Speaker 3: And then I think when you get to the executive branch, it seems to depend a lot on the individual personalities. Sure. I I honestly have trouble these days just, like, keeping up with the the news, but there seems to be a lot of coverage of, like, let's say, you know, commerce versus treasury versus the president himself in terms of those individuals' views on these questions we were talking about earlier. Like, exactly where do you draw the line? So, yeah, in in in a nutshell, would say, historically, I think that your characterization is correct. These days, I think it's it's much more mixed, and I don't think it breaks down super cleanly along party lines. When when when we were in government, we were surprised, I think pleasantly surprised, by, like, the degree most of the time of bipartisan support for this stuff as long as it was approached thoughtfully and not in, like, you know, just a reactionary nativist way.

Speaker 1: Yeah. Yeah. It's fascinating because, there are plenty of people in artificial intelligence right now who competition with China is effectively a single issue vote for them. Like, that is the most important debate. They think this is existential. It's a it's a winner take all. It's a huge deal, much bigger than anything else that's being discussed. And yet, if I if I meet someone and that and they're just talking about AI and and China and geopolitics, I can't instantly clock them as being a Republican or Democrat. And that's the nature of our current political situation, which which could be an advantage because it's not... It hasn't been hardened onto one side and you don't have that reactionary. So I'm I'm certainly hopeful that things can move forward... Final...

Speaker 3: Yeah. And I mean, maybe one observation is like it's it's been it's been very interesting sort of for me personally having now lived on both sides of this. I I worked at DeepMind, then I was in government, now doing an AI startup. And, like, these two... The the tech world and the government world Yeah. Which, have been sort of talking to each other more and more Yeah. But just sociologically, like, really separate. And so, like, a couple years ago, think it was the case that, like, people in the tech world... I mean, it's a huge generalization, but, like, didn't think a ton about, like, geopolitical tensions and the relationship with China. And people in the government world knew very, very little about AI. Yeah. And now those things are getting closer, but, like, it's still... I I think it's hard. I think most people will be surprised if, like, you drop them in the middle of the White House, At least, you know, from my experience or to make it less political, just like any piece of government. You drop them in the of government. You drop them in the of, like, big tech company. Sort of how little mutual awareness there is in these two worlds. Sure. Jordan?

Speaker 2: Take a wild guess. What do you think happens with NVIDIA's roughly 4,000 employees based in Shanghai doing r and d and software development and a number of other things? Sales, of course. It's gotta be a weird weird Friday.

Speaker 3: Truly... Yeah. I I really have no idea. I mean, it... It's, kind of going back to this point of, like, the spectrum between, you know, selling Coca Cola to things that are, like, super sensitive. It's like these, it's really easy to focus on these aspects of very sharp bilateral competition between The United States and China. And then neglected it, like, all of these things, like supply chains, companies, etcetera, like, people are buying, selling things, of course, like, between China and The United States all the time. Lots of Chinese companies have employees in The United States. Lots of American companies have employees in China. Like, where where each of those goes as different lines get drawn, really, really hard to say. I realize that's evasive, but, yeah, I basically, have no idea. But...

Speaker 2: No. Don't think we don't think we know, but it was always it was always, you know, an interesting question there.

Speaker 1: Get out the crystal ball and answer us. Okay?

Speaker 6: Yeah. Mean, my my uncreative response is like,

Speaker 3: these days the AIs are pretty good at forecasting things.

Speaker 6: Yeah. Maybe that's... Would be

Speaker 3: my best job if I'm pricing likely a different app.

Speaker 2: Before you leave, what what's the elevator pitch for your startup? Oh, Now you're gonna have a bit more time with the book out.

Speaker 3: Yeah. Yeah. Exactly. Yeah. So, inherent, we are a quickly growing London based NEO lab, and we are interested in two questions, which is what is the future of scientific discovery

Speaker 6: Mhmm.

Speaker 3: And what is the future of human cooperation and organization? And we think that those two things are actually part of the same problem. We should say, you should approach science not as something that you're trying to automate in a box and hand over to the AIs, but as something that's like a human machine teaming problem. So they're trying to build these agents and this platform for teams to tackle the hardest problems in r and d, but in a way that is safe and that empowers humans and and where humans actually have insight into how these problems are being solved. And it also makes the most of the areas where humans retain a competitive advantage Yeah. Taste, creativity, and so on.

Speaker 1: So going after the the specifically unverifiable rewards, things that can't Exactly. Just be Exactly.

Speaker 3: Makes perfect sense. Domains that are sort of underspecified, nonverifiable. This is an area where, like, taste really matters. Yep. Humans have, like, in general, the best humans have, like, a better sense of taste in these systems. How can you structure an institution around AI capabilities such that you're getting the best of both the the people and the machines?

Speaker 1: I love it. Well, thanks so much for coming on the show. Yeah. Great when you have more news, from the company and progress. I'm really excited.

Speaker 3: Absolutely. Would love to. Thanks for the time.

Speaker 2: Chat, Teddy.