Did America just lose the AI race to China?

Hello and welcome to Regulator, a newsletter for Verge subscribers about Big Tech, Big Government, and the big paradigm shifts that result from their collision. Not a subscriber yet? It’s a new year, so why not treat yourself?

In 2022, Jake Sullivan, then national security adviser under President Joe Biden and a powerful figure in the White House’s foreign policy team, assembled an interagency planning exercise out of the Situation Room: What were all the possible circumstances and outcomes of an AI arms race between the US and China — from trade wars to real wars, possibly even the arrival of AGI — and how would the federal government respond?

The details and results of that simulation are classified, but on Sunday, I spoke to Sullivan and asked if he could at least describe one outcome: Did they ever run a scenario where the AI industry’s profit interests were driving foreign policy instead of the government?

His candid response surprised me. “I confess that we did not factor in the possibility that we would actually be rolling back our export controls.”

Sullivan, now a professor at the Harvard Kennedy School of Government, is definitively pro-artificial intelligence and pro-innovation. But he thinks about AI in a way that a former national security adviser would: as a means to gain a geopolitical advantage against adversaries like China. His decision at the time was to place tight controls on the high-end chips that American companies were allowed to sell to China, extending a policy that dates back to the Cold War: Don’t sell high-end tech to America’s foreign enemies. Despite extreme backlash from tech companies, Sullivan was confident his policies would remain in place once Biden left office, telling Bloomberg in a January 2025 interview that “by and large,” tech CEOs accepted the need for those export controls, and that he wasn’t worried about incoming President Donald Trump cutting deals with them to loosen trade restrictions.

But that is, of course, exactly what happened. Over the past year, Nvidia CEO Jensen Huang slowly won over Trump, eventually persuading him to let Nvidia sell the H200 — its second-most powerful chip — to the Chinese market, with the US government taking a 25 percent cut. With America’s most advanced chips flooding into a market known for brazen IP theft, Sullivan is deeply alarmed that AI companies are literally delivering the United States’ technological advantage to a country that openly wants to surpass America. But even though the AI arms race is often compared to the nuclear arms race during the Cold War, there’s one crucial difference. The government funded nuclear research for the primary purpose of national defense, but tech companies are funding AI research for the primary purpose of making untold amounts of money. “Maybe that was a failure of imagination on my part,” admitted Sullivan.

Below, Sullivan and I have wide-ranging conversation about the current state of Trump’s AI policy, the lessons that Nvidia should take away from Tesla’s failures in the Chinese market, what he would have done if DeepSeek had been unveiled while he was still National Security Advisor, and — given that it was barely a day since the United States’ shocking abduction of Venezuelan President Nicolás Maduro — if the Trump administration would cite increased AI data center power consumption as a pretext for seizing Venezuela’s oil industry. (His assessment: “I would not put it past them.”)

“[I]n the short term, they will be able to book more orders. In the long term, they are contributing to a competitor that is going to squeeze them over time.”

This interview has been edited for length and clarity.

Tina Nguyen: You’ve spent years warning about the risks of China overtaking America in the AI race. But then Trump comes in, rolls back the entirety of Biden’s AI executive order, and makes America’s AI policy friendlier to the industry. Republicans in Congress and the White House would argue that lifting regulations on trade and development would achieve the same goal of beating China in the AI race. What are your thoughts on that?

Jake Sullivan: Well, one of the most important moves that we made during the Biden administration was to impose export controls on the sale of advanced chips to China for use for training of frontier AI models. And the fact that President Trump is now permitting the sale of some of these chips, in particular the H200, is just a gift to China, whose main constraint at the moment is not power, not money, not talent. It’s chips. And we are now basically filling that hole for them. I see no way in which that could possibly be anything other than an advantage to China, in the race between the US and China at the frontier of AI. 


And it goes beyond just the frontier, because the compute constraint for China also goes to their ability to serve their models globally — to become the AI backbone upon which the world runs. Prior to this decision, China was having trouble building data centers outside of China because they couldn’t produce enough of their own chips to do so. Now, they can have Tencent and Alibaba and other companies essentially building data centers using American chips to serve Chinese models to the rest of the world.

I do not see how that works on anyone’s terms to benefit the United States — and especially on the terms set out by David Sacks and Donald Trump, who want to see the world run on an American technology stack.

A couple days ago, the US announced that it would allow TSMC to import US-made equipment into their Chinese factories. I believe the argument they’re making is, “Oh, we’re basically using American technology to manufacture chips in China.”

In general, my view is that export controls, both on semiconductor manufacturing equipment and on the most advanced AI chips, is necessary to protect America’s national security and advance America’s edge, vis-a-vis China in AI. And I don’t believe that it will come at a long-term commercial cost to the United States, because the demand for this equipment and the demand for these chips is so great that there will be a growing market in the rest of the world for American firms, whether it’s Nvidia or one of the American chip equipment makers. So I don’t buy the argument that we have to allow them to sell some of the most advanced technology into China to make sure that they can earn sufficient returns to continue to grow their companies.

Just to give you one example: Nvidia, in 2022, objected to the export controls that we put into place and said, If you put these export controls on the A100 and the H100 chips, we are going to lose a lot of money. At that point, Nvidia’s market cap, I believe, was south of $400 billion. Fast forward three years: Those controls have been in place the whole time and Nvidia’s market cap is now in the multiple trillions of dollars.

It’s the most valuable company in the world at this point.

Exactly. And why? Because the demand outside of China is so great, and will continue to be great long into the future. So I have a real concern about the willingness of this administration to trade very short-term commercial gains for what I think will be a long-term cost. History over the course of the past few decades has been littered with cases of American companies wanting to sell into China to make immediate profits, only to learn later that once China mastered the technology and then poured money into the industry, [the companies] ended up getting screwed over time.

I mean, that’s basically what happened to electric vehicles, right?

Exactly. The idea had been that Tesla and Elon Musk thought, Get me a foothold into the Chinese market. Play nice with the Chinese government. This will give me access to a billion-plus people, and at the end of the day, it will help Tesla.

Logically, there’s some basis to believe that. But the life of policy is not logic, but experience. And experience has taught us repeatedly that what happens is: We go into the Chinese market. The Chinese find ways to essentially use the presence of the US there — whether through technology transfer, outright theft, or just brutal competition — and end up surpassing, and then ultimately evicting, the American tech or the American product from their market. We saw that happen with electric vehicles. I think we’re going to end up seeing it with respect to these advanced AI compute technologies, because China is determined — and has said this from the very top for many years, long before these export controls were ever put into place — that they intend to end their dependence on American chip technology, and to try to become a major competitor to American chip technology globally.

So what you’re saying is that the companies themselves will also see a long-term detrimental effect from selling to China?

My view is that in the short term, they will be able to book more orders. In the long term, they are contributing to a competitor that is going to squeeze them over time. I can’t make exact predictions about the future, and I’m not sitting in these boardrooms and I’m not seeing exactly what’s happening. But I would just say we have seen many, many examples of companies thinking that this is going to be a long-term success story, only to find ultimately that their entry into the Chinese market hurt their capacity over time.

Now, one thing I will note in particular. I think Nvidia is looking at chip demand globally and has a concern that it now has competition outside of China: Google’s TPUs, Amazon’s Trainium, potential other entrants into the GPU market.
And so it’s thinking, What do we do about this? Part of their answer to that threat, that some of the other hyperscalers in particular will not be sourcing Nvidia chips in the future, is to think: We need the Chinese market for this reason. I see that logic. I ultimately have questions about whether that will work for Nvidia. But what I know for certain is that it will not work for the United States of America, from our national security and national interest perspective, to allow Nvidia to sell these advanced chips to China.

I was reading an interview that you did with Time magazine last year, and you mentioned that during your time in the White House, you ran a war room simulation about the China/US AI race. To the extent that you’re able to talk about it, did you ever run a scenario where the private sector’s interests won out over government?

I confess that we did not factor in the possibility that we would actually be rolling back our export controls. We were operating scenarios presuming continuity of the export control policy. Now maybe that was a failure of imagination on my part, but having seen the first Trump administration — where they actually pursued a policy consistent with what the Biden administration very much built up on and expanded in our time in office — I was not expecting to see this administration come in and make these moves. And frankly, it’s come as something of a surprise to me. I think it’s come as a surprise not just to critics of the Trump administration, but many [of his] supporters in Congress, Republicans in Congress, who are stunned to have seen the Trump administration making moves along these lines.

Now that you’re a private citizen, could you possibly game out the results of this scenario?

There’s a few things happening now that deeply concern me. We’ve just talked about the chip controls issue, both with respect to the chips and to the manufacturing equipment that makes the chips.

The second thing that concerns me is the private sector. Many elements of the private sector are saying to the Trump administration, Get entirely out of the game of safety and security alignment issues.
We want a regulation-free, completely laissez-faire, let-’er-rip approach. And that has been reflected in the public statements and the public policy of the Trump administration. My concern is that we may end up in a circumstance in which China is dominating the global conversation over standards, with the United States virtually absent from the table. I think that is another area in which China could actually seek gains in its ability to become a dominant global AI player, where the United States is simply absenting itself, in part, at the behest of a private sector saying, Hey, the government shouldn’t be involved in this whole question of standards or regulation or safety or the like.

The third area that concerns me is when it comes to the question of talent and America’s ability to attract the best talent from around the world. This administration is taking a series of steps that is putting up a “not welcome” sign. It is going to make it harder for us to get the very best engineers and very best minds when it comes to AI going forward. I think that that is going to put us at a disadvantage as well. 


And then the final area is: America’s long-term innovation edge has always been built on basic research, and this administration is slashing the budget for basic research. I think that that is problematic because if you rely entirely on the private sector to do all of your innovation, they can do amazing things, and our AI companies have done amazing things, and I have huge respect for the breakthroughs that they have made. But you also need research at universities, with National Science Foundation grants, that are looking not for a commercial advantage, but just for very basic scientific breakthroughs. It is that kind of research that ultimately leads to the greatest revolutions over time. Slashing that money and relying entirely on, basically, the big technology companies to do the work for innovation — I think [this] is problematic from the point of view of the US. It is disrupting a formula that has worked for us time and time again over multiple decades.

One of the things I focus on is Trump’s transactionalism with Big Tech, and what he wants from AI companies in return for doing them this massive regulatory favor. 
You’re seeing OpenAI’s leaders donate to MAGA super PACs that boost him and his candidates. Now Trump is asking them to prevent “woke AI,” attacking Colorado’s approach to prevent algorithmic discrimination. What are your thoughts about the AI industry itself becoming partisan?

I think it’s dangerous for any industry to wear the jersey of one political party or the other. When I was national security adviser, I worked hard to have good relationships with the leaders of all the major frontier AI companies. I think I succeeded in doing that, even though it was obviously widely reported that there were tensions between the administration and many of those companies. I think it’s important for senior government officials to develop good relationships, to understand where these companies are coming from. These companies are, in my view, a national asset of the United States. It is good that we have the most cutting-edge technology companies in the world, but it is also right and proper for the US government to ensure that there are proper regulations, norms, guardrails in place, both to protect the safety of Americans and to protect our national security.

If an entire industry decides we’re going all in with one administration or one political party, or cutting deals, both political and financial, with an administration, it worries me that that is going to end up eroding what is one of America’s greatest strengths, which is that we don’t have this kind of entanglement of the private sector with our government, in ways that other countries have seen and have suffered from. So I would hope that the technology industry would think about the long term, and about the need to have good relationships with political actors across the spectrum, and to stay out of trying to make big bets on what exactly is going to happen in American politics, since none of us can can predict that, by any stretch of the imagination. 


So this is a bit of a hypothetical, but if you had still been national security adviser when DeepSeek was unveiled, what would have your response been?

I think that the Chinese government actually played a role in hyping the DeepSeek release in order to send a message through their propaganda, relentlessly — and sadly, somewhat effectively — to voices in the West, basically saying: Resistance is futile. Your export controls cannot work. Give up on them. The whole chip strategy has not succeeded. That is what they tried to do. And that is nonsense. 


You know how we know that is nonsense? If I had still been national security adviser, I would have been pointing this out at every chance I got: DeepSeek’s CEO himself has repeatedly said that their main constraint is chips. Their main constraint is compute. What has made it difficult for them since that release at the beginning of last year, to continue to march forward at the pace they want to march forward, is they have not had access to high-end American GPUs. President Trump is now making it easier for them.

I would have been stressing that there is more hype here about what this means for America’s compute advantage than there is reality. And in fact, the DeepSeek release, in a way, reinforced the need for effective, enforced, and in some instances, expanded American export controls when it comes to high-end GPUs.

Do you think the Trump administration has done anything right in the AI race against China?

Well, first, I think they have put a lot of emphasis on the importance of the AI revolution that we’re in the midst of. I think they have properly recognized that this is an incredibly important moment. They have put a lot of high-level policy attention behind it. They are not neglecting it. I think the choices they’re making aren’t great, but I’m glad to see them elevating and prioritizing the issue.

Second, they have largely abandoned the safety agenda. I think it is wrong and history will judge it as having been wrong. But one area where the AI Action Plan did, in fact, continue the Biden administration’s approach, was on safety when it comes to the risk of bio-uplift: the potential for convergence between AI and bioweapons in ways that could harm the security of the American people and people everywhere around the world. So I think that that has been a good thing for them too. And there have been these interesting initiatives like the Pax Silica, an effort to work with some friends and partners of the United States on chip technology, on other aspects of AI science and tech. I think that work is valuable work and should continue, and I want to see a further doubling down on engaging with allies, friends, partners in Europe, Asia, and other places as we go forward. 


But [overall], I think the administration has taken steps to reduce America’s innovation edge, increase China’s access to advanced compute, and erode America’s leadership role in terms of the standard setting necessary for being the global leader on AI. So I don’t have a very good story to tell, I’m afraid, although some of these steps, I think, deserve praise.

Given how rapidly the situation is changing, I only have one Venezuela question, but I do want to loop it into the context of AI data centers and energy consumption. The International Energy Agency projects that by 2030, the US’ total energy consumption is going to rise by 10 percent, probably more, due to the demand for AI data centers. Forty percent of it is likely going to come from fossil fuels. What are things to keep in mind as Venezuela gets taken over by America and possibly has its oil seized?

First, the idea that we would engage in military action, with the primary purpose appearing to be increasing preferential access for American oil companies to a foreign country’s oil — that, I find disturbing, and frankly, a little bit bizarre.

The chief issue that I see in the United States right now, when it comes to the power necessary for the AI revolution going forward, is about creating generation and transmission capability and improving our grid. It’s not about pumping more oil out of the ground. And frankly, China’s gotten the memo that the Trump administration has completely refused to read. Over time, the [dropping] price point of clean energy — including solar and other forms of clean energy technology — [is] going to mean that, as this goes forward, betting big on heavy crude from Venezuela as part of an answer to an AI energy challenge? I must say, I don’t find that particularly convincing. I haven’t heard them make that argument, but I don’t think it holds a lot of water. 


It seems like something they would totally just throw out there.

I would not put it past them.

If you want a nice, relaxing change of pace, my suggestion is that you go read our live coverage of CES this week. We’ve got a team reporting from the expo floor in Vegas, editor-in-chief Nilay Patel is taping a live episode of Decoder, and I’ve already found my new favorite consumer electronic of 2026:

Happy New Year, everybody! See you next week.