Subscribe here: Apple Podcasts | Spotify | YouTube
Beyond the salacious details of the Nvidia–Jensen Huang scandal is a story about national security and the advanced chips powering the AI boom. But what exactly are these chips, and why do they matter? Ben Buchanan was one of President Biden’s AI advisers and helped write some of the U.S. policies on chip controls, in conversation with people like Huang. In this episode of Galaxy Brain, he breaks down what makes those chips so valuable and why they have such implications for the U.S.-Chinese relationship and the future of AI.
The following is a transcript of the episode:
Ben Buchanan: We had a lot of conversations with Jensen and with Nvidia. They were conversations about technology, not spies or sex or anything like that. So that’s a new dimension to it. But from a technology perspective, this is actually remarkably consistent with the concerns that we had in the Biden administration.
[Music]
Charlie Warzel: The further we get out from the Nvidia–Jensen Huang scandal, the more it seems clear that, despite all the salacious details, there’s a lot more to this story than meets the eye.
It’s an alleged Chinese-spy sex scandal, yes. It’s the fall of a tech giant. It’s this huge, Silicon Valley palace-intrigue story. But as time goes by, it seems that the epicenter of this story is in Washington, D.C. The Nvidia–Jensen Huang affair is, at heart, a geopolitical thriller. It is a series of cascading events that’s been set in motion by a cold war between China and the United States over advanced AI chips and the AI race.
And so to really understand why Huang had to resign, why he’s being called before Congress to testify, why the SEC has reportedly launched a probe into Nvidia—you have to understand how the U.S. got locked in this AI race with China, and how Nvidia’s chips work to power the AI boom. You have to understand the policies that have led to the export restrictions on the chips that Nvidia sells, and the convoluted loopholes that Nvidia and Alice Huang have allegedly used to get chips into China.
And my guest today: He not only understands every element of this story; he helps shape it. Ben Buchanan is the former White House special adviser on AI during the Biden administration. Buchanan helped design the chip-export controls on China that Huang and Nvidia lobbied against.
And along with Tantum Collins, he’s the author of this new book, The Bitter Struggle: Superintelligence, Superpowers, and the Fate of the World. He was in the room for so many of the events that led to where we are now. And today, he’s going to help us mere mortals actually understand what the heck is going on with China, chips, Jensen Huang, the AI race, and crucially, why all of it matters. He joins me now.
I’m Charlie Warzel, and this is Galaxy Brain.
[Music]
Warzel: Ben Buchanan, welcome to Galaxy Brain.
Ben Buchanan: Thanks for having me.
Warzel: So I’m hoping in this interview that we could do some very basic questions that will lead to this deeper understanding of the U.S.-China race, the labs; what’s going on. But first I just want to ask about Jensen Huang. Have you taken what we know now and reexamined any of either the interactions you’ve had with him or his company? Like, were there things in retrospect that now that look like signs of vulnerability? How are you processing this news moment, given your relationship to this entire story?
Buchanan: For background, I was the White House special adviser for AI and worked in the National Security Council, and was the architect a lot of the chip controls that seem to be at the crux of this story here.
In the course of doing that work for four years, we had a lot of conversations with Jensen and with Nvidia. They were conversations about technology, not spies or sex or anything like that. So that’s a new dimension to it. But from a technology perspective, this is actually remarkably consistent with the concerns that we had in the Biden administration. Certainly about the importance of chips to national security, the importance of AI to national security, and the necessity to cut off smuggling routes and remote access routes for China to get access to that technology. So, in some sense, it’s like a new chapter in a long-running novel. Maybe a particularly salacious and interesting chapter. But it fits the themes of the novel that I’ve lived, I guess now for six years.
Warzel: Some of this stuff, especially. And I want to get into the stuff about Megaspeed: the company that Huang’s alleged mistress, and alleged Chinese intelligence asset, worked for. That’s in your book. Megaspeed is in your book.
And so I’m just curious. Like, was all this on your radar, and you were expecting it all to come out in some fashion? Or were you pretty shocked by the fact that we’re all talking about this right now?
Buchanan: I never knew of the sex thing. As best I can recall, I did not know of Megaspeed in the government. I definitely don’t think I ever heard the name Alice Huang. Megaspeed is in the book because The New York Times did a good story on it 13 or 14 months ago. And it looked to us, when we were writing the book, as a great avatar or example of exactly the kind of smuggling and remote access we were worried about, and we were trying to prevent.
While the particulars of this case are new to me, or at least were not something we discussed in the government, the general idea of circumvention of the export controls because this technology is so important was something that we spent a lot of time on and in fact were actively trying to combat.
Warzel: So I wanna ask some basic questions here, because I imagine I’m not alone in what is essentially like a pretty shallow understanding of the story in Nvidia. I think, you know, reading the news, a lot of people understand they make advanced AI chips, power the boom we’re experiencing here. They have an extraordinary amount of market share. But walk me through just to set the table here. Like, why is that? What makes Nvidia’s chips so great, so necessary, such a fundamental part of this whole boom?
Buchanan: I would look at Nvidia as a key part of maybe the world’s most advanced supply chain and manufacturing process. So Nvidia designs the chips that are incredibly intricate—hundreds of billions of transistors at nanometer scale—that enable this AI revolution that we’re part of, for better or for worse. And those chips are made in Taiwan almost exclusively at TSMC using equipment from ASML in the Netherlands.
These are three companies we profile in a lot of detail in the book, because they’re three of the most important companies in the world. Everyone knows of Nvidia; some people know of TSMC; almost no one knows of ASML. But it’s that kind of supply chain, and then all their subcontractors from there, that is at the core of AI.
In AI, we have this notion called the bitter lesson, which is the inspiration for the book’s title, The Bitter Struggle. And the bitter lesson is that AI scientists want to make AI systems almost in this divine-creation process, in which they’re handcrafting the algorithms to perfectly mimic or encapsulate human knowledge. But actually, what works much better, we’ve observed, is giving the systems a flexible algorithm from which to learn. A lot of computing power. And getting out of the way.
And that’s why it’s computing power—and in particular these days, Nvidia’s chips—that are at the absolute forefront of this AI story, and therefore this geopolitical story. Because what companies have realized is they can scale up the amount of computing power. They can scale up the amount of Nvidia chips that are used to make AI systems, and they get much more powerful systems as a result. In a very predictable fashion. And that, more than any other single factor, I think, has been the tailwind that’s propelled AI progress over the last 10 or 15 years.
Warzel: And it’s what’s made them these companies just basically insatiable, right? In terms of how much they want to purchase. How much they want to use.
Buchanan: That’s right.
Yeah, it’s worth dwelling, for one second, just on the numbers here. Because they’re absolutely mind blowing. And they underscore one of the other important elements in the book and in our policy—which is that AI is the first revolutionary technology of the last hundred or so years that comes from the private sector. Think about nukes and space, and so much else, coming from the government. The government’s cutting the checks at a minimum, and has some capacity to steer where the technology goes. The modern paradigm of AI, this computing power–driven paradigm that we were just talking about, comes from the private sector.
And it really is just eye-popping: the amount of investment the private sector is making in pursuit of the bitter lesson, or what’s sometimes called the scaling law. In 2026 alone, the estimates are that the hyperscalers, the big compute companies, will spend $700 billion on infrastructure and growing to a trillion or so dollars by 2027. Just taking the 2026 number, that’s akin to a Manhattan Project basically every two or three weeks.
That’s more than America spent building the entire interstate-highway system, adjusting for inflation over 35 years. So it’s just so hard to overestimate the amount of computing power at play here in making these AI systems.
Warzel: And it sets the stakes of this, I think, extremely well. Before we get into the nuts and bolts of this Nvidia story, I wanna step back; talk to a little bit about our technological relationship to China. You’ve said in the book, in the past, that AI will be the fulcrum on which the U.S.-China rivalry turns. So help me understand why that is. And also why it’s so bad that China gets access to these chips.
Buchanan: We entered the White House in 2021 with the belief that AI would become very relevant, if not at the absolute center, of national-security affairs. Not just with China, but around the world. There’s a number of ways in which that importance could manifest over time, which we thought would take five or so years.
Let’s take the one that’s the most concrete right now, and that’s cyber operations. It’s no secret that the United States and China engage every single day in cyber operations; cyber espionage against one another. That this is one of the key contours of competition between the two nations—that if you have an advantage in cyber operations, you have a substantial national-security advantage. I don’t think anyone can test that at this point.
What was speculative in 2021—but that we believed would come to pass before too long—was that AI would change cyber operations. In 2026, it has. In 2026 alone, AI systems have discovered thousands, if not tens of thousands, of critical vulnerabilities in vital software all over the internet.
The AI companies that have built this software, working in concert with the U.S. government, have turned these vulnerabilities over to vendors and to other network defenders, so that they can be patched and remediated. Essentially making the internet as a whole more secure. My belief, and I think this is pretty well supported, is that if China had access to such a technology and was finding these kinds of vulnerabilities in software, it would not be turning them over to be fixed and remediated. It would be using them for intelligence purposes. And that would put key parts of American critical infrastructure at risk.
Just recently, this past week, as we’re recording this here, an AI company using an advanced AI system disclosed that they had found a vulnerability in Java, one of the absolute core languages that makes the modern economy go. A couple of weeks ago, the same company announced that they found a vulnerability in PHP, which is the back end for 70 percent of the websites on the internet. So we are talking about mega-vulnerabilities that AI systems are finding that have profound national-security implications in the wrong hand. And our view was certainly that it was better for the United States to get to that threshold first.
Warzel: So you’ve also said that this race to unlock all these capabilities, including the cyber capabilities, may be “winner take all,” I think is how you described it. Help me understand why that’s a zero-sum game.
Buchanan: I think we should be very precise here. So the dynamics are close to winner take all—which is to say, if you get to these critical thresholds first, then you can aggregate for yourself a lot of the power that comes from reaching those thresholds. We just gave the cyber example. I think there’s no doubt that the nation that got there first has a lot of power. What you choose to do with that power does not need to be zero sum.
And in fact, I would argue throughout American history, a lot of the moments where American policymaking was at its best was when it had a fair amount of power and in some ways ceded that power to institutions, or built a stabilizing framework. So I don’t know that it’s zero sum in the geopolitical sense.
Certainly, when I was in the government, we spent a lot of time trying to negotiate with China. And trying to say to China: Look, we’re taking competitive actions here, of course. But we think there’s interest, and that both of us have, that are shared. Let’s work on making this a positive-sum relationship. The Chinese largely blew us off. We can talk about it if you’d like. But this was certainly a key part of our overall AI strategy.
Warzel: How would you describe the “they blew us off” part of that?
Buchanan: So in October 2023, President Biden signed an executive order that included a line that when you translate from the legalese, essentially, said the United States will engage on AI safety—not just with its allies and its partners, but with its competitors. And that of course meant China. And President Biden personally took that line to President Xi [Jinping] three weeks later when the two leaders met in Woodside, California. And essentially he said: I’m serious about this. We have to work together on this and have a positive-sum relationship.
The two leaders agreed to set up a dialogue of the AI experts in Geneva in May of ’24. It happened. I was there. It was maybe 10 or so U.S.-government officials, 10 or so Chinese-government officials. We took it incredibly seriously. We sent essentially everyone who was an AI leader in the U.S. government, went on this trip. The Chinese did not.
And the Chinese subordinated it to their political apparatus, and spent the entire meeting essentially saying, like: We’re not sure why you’re talking about risks. We don’t see risks in AI. All this discussion of risks, even shared risks, is American imperialism.
We pushed anyway. And we said, No, we really think there’s shared risks here. We really think there’s things we can do that are positive sum.
Eventually, after six months of pushing on that point, we got the most common-sense thing, the thing that you should think would take six minutes. And that was an agreement between the two leaders, President Biden and President Xi, when they met for the last time, in November 2024—to not put AI in nuclear command and control. And I think we can all agree we don’t want AI firing nuclear weapons. So this should be the most simple thing in the world, but it took 13 months of pushing to get China to even get there.
I’m hopeful China’s posture is different now. Certainly if we had had a second term, we would have—we were making plans to continue outreach to China on this point. But there wasn’t a lot of doubt in my mind that China was not, at the time, ready to have a serious conversation about shared interests and shared risks, and how we could all work together.
Warzel: Do you have any sense of where you think China sits now on this? Given that, you know, the capabilities have skyrocketed? This is clearly a more of a top-of-mind geopolitical issue probably than it was, you know, in the during the Biden administration.
Buchanan: I can point to some pieces of evidence that suggests China’s taking it more seriously. There’s some suggestion they’re thinking about cyber risks and talking to their companies about cyber risks. There’s some suggestion in some of the Chinese Communist Party publications that they’re warning about the dangers of AI and so forth.
So I think there’s some evidence. And certainly, those are the threads I would be pulling if I were in the government. But also, China is essentially a nation run by one person.
And I think it’s very hard, you know, even when you have access to U.S. intelligence to know exactly what Xi Jinping is thinking. I don’t have access to U.S. intelligence right now. So I wouldn’t wanna pretend to know what’s in his head when it comes to AI. And that, really, at the end of the day is the ultimate question when it comes to China.
[Break]
Warzel: So the Biden White House has a, you know, very specific policy about China having access to Nvidia’s chips. There’s a real focus on this amongst all the AI-policy folks, including yourself. Take me to [Jensen] Huang and Nvidia. What is Huang’s policy toward China when it comes to selling to them or around this period?
Buchanan: He said the same thing in private that he said in public. Which is, we put this policy in place in 2022. We said to him, essentially: The supply of chips in the world is fixed, because it’s so hard to make them. There’s not gonna be that many more coming off the line. And we want those chips to not go to China. Because we’re worried they’re gonna modernize their military; they’re gonna repress their population with these surveillance tools and the like. Instead, we want those ships to go to the rest of the world, but especially companies in democratic nations.
And Nvidia’s posture was that this would really hurt them; that this was bad for them economically. And that they needed the China market in order to compete. And sometimes, to give the sort of steel man of Jensen’s position, Jensen would often say things like—or Jensen and, I should say, his supporters, because there’s a whole group of folks who had this view—would say things like: Well, we need to keep China dependent on American chips. And we’ve got to sell them the chips, so they don’t make their own.
We disagreed with the two prongs of that argument. The first prong we disagreed with was that this would be very bad for Nvidia. And essentially, our view was that the bitter lesson that we talked about before was so dominant, that the tailwinds of demand for American compute—I think you said the companies in the United States are insatiable when it comes to compute. That’s absolutely right. Jensen said that as well. And they would buy every single chip Nvidia made, and they would buy it for a significant profit. And that’s exactly what happened.
Nvidia’s market capitalization in 2022, when we did the controls, was something like $350 billion. Just before the Jensen scandal a couple of weeks ago, it was $5 trillion. They’ve had a rough week here. But depending on how you do the math, they’re probably up 10 or 15 X on where they were when the compute control started, because of the tailwind that we foresaw about the scaling law and the bitter lesson and the demand for compute. So Nvidia has done great.
And in fact, on the Nvidia earnings call just this week, when asked about a recent Trump action to close some of these loopholes and reassert export controls, they downplayed the impact of it, because they’re now acknowledging the demand is so big around the world.
So we don’t think we were hurting Nvidia. We think Nvidia was going to do very well under our policy, and indeed they did.
The second question is: Are we hurting the United States? Because, do the chip controls cause China to indigenize production in a way they wouldn’t otherwise have? And therefore we lose some leverage over China because they’re no longer dependent on American chips? I also think this was wrong, and we thought it was wrong at the time. The reason why is that China knew the importance of advanced chips. As far back as 2014, when America was off fighting ISIS, China recognized, up to the Xi Jinping level and the State Council of China, recognized how important chipmaking was. China at the time was spending more money importing chips than it was spending importing oil.
And they couldn’t make any advanced chips at home. So Xi Jinping began an effort when you count the public and private capital upwards of $200 billion to build a chipmaking industry in China. So they had the pedal to the metal on indigenization before we ever took any action.
The Trump administration, in their first term, made it a little bit harder. They cut China off the top end of equipment called EV machines—from the company I mentioned before, ASML. And then when we took the controls, we put the controls in place in 2022 on chips. We also put controls in place on chipmaking equipment, another generation of chipmaking equipment, which were imperfect, but what held China back.
So the bottom line is even today, even despite the controls and despite China’s more than a decade-long effort to indigenize, chipmaking is so hard that China has failed. And don’t take my word for it: [Commerce Secretary Howard] Lutnick, one of the men at the center of this recent export-control story, testified in 2025 China can make just 200,000 chips.
And together, 200,000 for the whole country is not nearly enough. It’s not even enough for a single data center. So those were Nvidia’s arguments, and what our responses were to them.
Warzel: So in 2022, as you said, the U.S. implements these pretty sweeping export controls on semiconductors or on the chips. Can you briefly explain to me that policy in broad source for people who are not familiar with, you know, the export controls?
Buchanan: Basically, our view was the risk of diversion was so high—that the chips could be sold to a civilian buyer, but would end up in PLA, the Chinese-military data centers, or end up in the domestic surveillance apparatus—that we had to control the sale of chips. Not just to the particular military buyers in China, but to the whole country.
So the policy in 2022, and then some tightenings thereafter, was to say to companies like Nvidia: You may not sell your product; these advanced chips. You can sell lower-end chips, but you cannot sell the these particular advanced chips into China, full stop. And that was a big step.
Warzel: This is a dumb question, but what’s the penalty there?
Buchanan: It’s a crime. The Department of Justice and the Department of Commerce have together pursued investigations to arrest and charge people who have violated U.S. export controls or smuggled chips into China, or things like that.
Warzel: So the export controls, as I know it in the story, is ultimately one of the decisions that brings Huang and Nvidia—who are not, you know, huge political players leading up to this—more into the D.C. fold, right? It seems that a lot of like their advocacy, especially Huang’s, really ramps up in response to this.
So I’m curious: Tell me how Nvidia started rolling up its sleeves and lobbying against those bans starting around 2022, when you guys put them in.
Buchanan: We consulted with Nvidia as we were doing it, even before it happened. Because we were trying to understand the technology, and we were certainly not trying to surprise them out of the blue. So it’s not like they woke up one morning and said, Well, you know, the policy is in effect right now. So there were conversations all along.
I think those conversations obviously intensified once the policy was in effect, and as we were making adjustments to it and so forth. But these were the reasonably standard conversations that the government has with companies.
And you know, when I was the White House special adviser for AI, I talked to all the AI companies pretty regularly. I think it’s part of the job of the government, especially when the private sector is inventing the technology. It’s part of the job of the government to understand what they’re doing, and what its implications are.
Warzel: But I’m curious. Like the tenor, the tone of it—as someone who had the opportunity to be in the room. I’ve seen some reports. There’s a report that I read recently from Semafor about Nvidia’s more recent lobbying efforts. Quoting an anonymous congressional staffer saying their MO has been largely gaslighting: creating an intense trust deficit between government, the Hill, and Nvidia.
Like, were these conversations acrimonious? What’s the tenor of these conversations?
Buchanan: I think they were strident, maybe even aggressive, but it didn’t bother me. I don’t—you know, I just felt like Jensen was doing his job, and talking his book. And those of us in the White House, and they were a group of us in the White House who had very strong conviction on this, felt like we were here to serve U.S. national security. And serve the public, and also serve American innovation.
But we certainly felt like we had the technical depth to have detailed conversations with Nvidia, and we did. And you know, the policy ended where, I think, in general the policy should have ended. I would have probably a little bit more aggressive, if you had asked me, in terms of some of the controls.
But in general, I think the policy got to where it needed to be. Probably got there slower than it needed to get there. But in general, I’m happy with certainly our core strategic and technological thesis, which I think has been borne out at this point.
Warzel: So now we’re at the current scandal. The affair with Alice Huang—no relation, and former executive at Megaspeed. Allegedly the Chinese intelligence asset. With the caveat that Megaspeed has denied all allegations of wrongdoing, can you explain a little bit about how the this alleged chip smuggling works in practice?
Buchanan: So I want to be clear here that I don’t have firsthand knowledge of Megaspeed. I’m only going off the news reports. I’ve left the U.S. government, obviously, in January 2025. But it does seem like Megaspeed is an exemplar of two categories of risk that worried us quite a bit when it comes to China’s access to chips.
The first is that the chips themselves would be smuggled into China. So they would be purchased in a place like Malaysia or Singapore. There might be some facade of a buyer, potentially a company like Megaspeed in one of those places, that is taking the chips into their data center; is saying, Yes, we’re going to run them here. Everything is good. Don’t worry.
And then they’re taking the chips out of the data center, and they’re shipping them into China. And they’re ending up in a Chinese-military data center or the data center of a company that works for the Chinese military, or something like that. And that obviously is a very significant concern for the United States, because that’s direct smuggling of the chips themselves. And we don’t know—once they go into China, we don’t know what the chips are going to be used for.
The second category of concern is what’s called remote access. The chips still go to a kind of shady company in Malaysia or Singapore, or a place like that. They go into a data center, but they stay there. The difference is China—Chinese military, Chinese companies—just access that data center remotely over the internet.
So the chips stay there, but China can train their next-frontier AI system on this data center, remote. So it’s a different way of circumvent export controls. And we made a decision in one of the export-control updates to try to block that. To say: As a condition of taking these chips, you have to pledge that you will not use them to enable Chinese AI training.
President Trump, when he came into office, declined to enforce that rule. So there’s at least a legal gray space—of maybe it was permitted, or the administration wasn’t enforcing it.
But those are the two big categories of concern: the actual physical smuggling of the chip, and then the remote access. It’s worth saying Nvidia spent a great deal of time in 2025 and 2026, including Jensen personally, saying chip smuggling is not an issue. And I think it’s obviously incorrect. Not just because of the Megaspeed case, but also because of other cases in which the Department of Justice has brought really well-documented indictments and criminal charges against people for smuggling chips in the way that I described.
Warzel: In terms of that remote access to those data centers that are outside of China, for training purposes, it’s known that China has distilled—or sort of reverse-engineered, you know—U.S. models in order to kind of keep up in the race here. Is that where that is happening in your mind? Like, inside those places? Or is that happening more locally? Or is it a mix of the two?
Buchanan: Assume it’s a mix. So let’s just be very precise about what’s happening. Distillation is a process in which you can essentially—you know, if folks here are Formula 1 fans, you can draft behind the car in front of you. And you can get the outputs of a leading AI system and come up with the facsimile of that system, imperfectly. You’ll lag capabilities, but it’s cheaper.
That is a general concern. And one of the ways in which China has been trying to compensate for the lack of computing power. Essentially, is copying the Americans’ homework. That process of distillation still requires some computing power. And I would imagine China is using whatever computing power it has available to do the process of distillation, whether that’s chips overseas or chips in China.
Companies have gotten much better at stopping distillation. So Anthropic, a company that I advise, but I don’t speak for, reported that they came up with methods in the summer of 2026 to stop distillation. I think it was like up to like 99 percent. And I think OpenAI is making some moves in that direction as well.
So my suspicion, hard to know for sure, is that we will look at this kind of adversarial distillation as an artifact of a period from 2025 to 2026 where China really struggled with access to computing power, in part due to the export controls, and was using distillation as a way to try to keep up with the frontier. But I don’t think that will be sustainable for them in the long run. And then they’ll be back to the same box of not having enough computing power.
[Break]
Warzel: Given what we’ve learned about Megaspeed, and what you’ve said about the Department of Justice also bringing cases about people trying to, you know, navigate around these export controls—what is your sense of just how much Nvidia helped China get access to these shifts? Like, how bad is this going to get for Jensen, for Nvidia, in terms of where this can go?
Buchanan: It’s a great question. Again, I don’t have firsthand knowledge of the facts here, and I don’t want to pretend. It does seem like the amount of chips going through these channels was really significant, potentially even more than the number of chips China could make as a whole country itself. I think The Wall Street Journal reported something like $12 billion of chips. I’ve seen other reportings saying it was $4 billion to $5 billion of chips.
Megaspeed is just one company. There were other companies that were potentially doing this. What Nvidia knew, what Jensen knew—that is all for congressional and maybe criminal investigators to find out. I don’t pretend to know.
But certainly, as a national-security matter, I do think that the flow of chips to China in this way—smuggling and through remote data centers—is one of the biggest national-security weaknesses the United States has. And what you’ve seen really is a pretty clear bipartisan consensus to change the policy. And there’s bipartisan bills in the Congress—introduced by, you know, MAGA senators like Tom Cotton and Jim Banks, and people like that—that would cut China off chips and chipmaking equipment as a matter of law. So the president couldn’t change it, even if he wanted to.
And then, even the Trump administration, this past Monday, changed their policy and cut off remote access to data centers and required licenses for shipments to Southeast Asia to try to cut off the smuggling. They’re not quite done; there’s still a couple loopholes in their rule. But that was a very significant action for an administration that really, you know, from the middle of 2025 to today, was only taking export controls away and not enforcing them, rather than putting them into practice.
Warzel: So there are calls from Silicon Valley right—from the shareholders, from the, you know, VC types, the other tech founders of some esteem. They want Jensen Huang to come back. He’s a visionary; he’s integral to the AI economy. Help me understand the rationale here, because China is enemy No. 1 in terms of the race to superintelligence. But also, it just seems like they’re abandoning the China fears all of a sudden, or excusing this dalliance. This potential dalliance.
Buchanan: I think a lot of the folks who are making those calls—which, to be clear, the Nvidia board pretty clearly rejected on their earnings call this week. A lot of that constellation of people has been pretty dismissive of the China concern. So they’re often the ones supporting selling chips to China. So at least they’re consistent on that front, in that they’re just not worried about this. I think there’s probably a key difference between my view and their view. Which is: I think AI really is going to matter for national security. And at least their revealed preference is they don’t think that, because they’re willing to let China in the game. But as a national-security matter, I certainly am very concerned and think these activities should be investigated.
Warzel: What is so hard for me to understand with that is that the national-security imperative is part of the imperative of the companies, that are saying: We can’t afford to truly pace the frontier—or “stop” or “pause” or whatever—because it’s imperative that we beat China to this. I mean, I do feel like they’re speaking out of both corners of their mouth here.
Buchanan: I do think, yeah. So if what you’re saying is sometimes the same people who are saying We can’t regulate AI are also saying Let’s sell chips to China. I mean: We can’t regulate AI, because we have to beat China are also saying Let’s sell chips to China.
I agree. That is totally incoherent. It doesn’t make any sense to me. President Trump himself—in, you know, militaristic terms that I would not use, given our discussion before about how we hope to have a positive-sum outcome here—President Trump is very fond of saying: Whoever wins AI, wins.
You would think if that is your policy, you would not want to sell chips to China. I don’t get it. I think it’s incoherent for the reason you describe. To be clear about what I do believe, my view is: American companies in fact do have to slow down. And I support pacing the frontier.
We haven’t talked about it, but you know, part two of the book is all about how we regulated the AI companies, and what more we need to do on the regulation side. That was sort of huge part of our policy. And my view is that, in fact, the way in which we can create the space for regulation—as far back as 2021, we believed—was to slow China down.
And China’s absolute dependence on American computing power was a way for us to slow China down, to build a lead for the United States. And then to, if you’d like, spend that lead on safety, to when the technology got to a critical moment. Which is what we’re in right now. To regulate the companies pretty aggressively, if I had my way. To make sure this technology is safe before it’s rolled out. Without this argument of: Well, we’re in an arms race with China; we must forge ahead. Which I thought was just very destructive.
Warzel: So where does this all leave us right now, as we’re sitting in the story here? We’re talking Wednesday, November 18th. Obviously, things are changing by the minute.
Buchanan: Always good to timestamp these conversations, ’cause it’s been changing every 24 hours.
Warzel: Genuinely, it is. How much hot water is Jensen in? How dangerous and serious is this for Nvidia, as Huang goes to testify before Congress in December? Like, what are the stakes for both Huang and Nvidia going forward?
Buchanan: I think they’re very high. Congress, in a bipartisan way, said it’s gonna do an investigation here. I don’t know what’s gonna turn up, but certainly they’ve got a lot of threads to pull. Whether or not there’s a criminal investigation, I have no idea. But I do suspect there will be a lot of government investigations into what happened here, and there should be.
I guess in the medium term, Nvidia’s chips are still gonna be really good, and there’s still gonna be big demand for them. So as long as Nvidia can keep making great products, I think they will continue to sell those products. I guess it’s an open question of: Can they make great products without Jensen? We’ll see. They’ve got 31,000 research engineers, or something like that. So there’s a lot of smart people there, and they’ve got a lot of great technology. So we’ll see what happens to them technologically. But yeah; in the context of the investigations, certainly I would describe this as a serious moment.
Warzel: What should senators be asking? Or what should anyone in Congress be asking Huang when he’s up there, you know, in front of Congress?
Buchanan: It’s the most important, age-old question in any kind of D.C. investigation, which is: What did you know, and when did you know it? And they should be asking Jensen that, and they should be asking the commerce secretary, Howard Lutnick, that. Who knew what about where these chips were going. Who was getting them. How active was Nvidia’s role in enabling smuggling? How active was Jensen’s role?
Again, I don’t have firsthand knowledge of the answers. But those were all the questions I’d be asking.
Warzel: And I think lastly here, what has this done for the AI ecosystem right now? Like, is it in a worse place than it was? Or let’s just say, the American AI ecosystem?
Is it in a worse place than it was two weeks ago, before all of this stuff really trickled out into the press? Or it was about the same?
Buchanan: I think it probably creates some turbulence in the waters, but my suspicion is we’ll look back at this in six or 12 months, and the big story of AI will be largely unchanged by this. And the big story of AI, you know—as we spent a lot of detail on the book—is: You have really powerful chips being used by really capable algorithms, at larger and larger scales. Yielding extraordinarily capable systems, with profound national-security implications and profound implications for everyday life and democratic values. That story.
To be clear, I’m not a cheerleader for that story. You know, you don’t write a book called The Bitter Struggle if you’re a cheerleader. But I think that story is largely undiminished by these events here.
Warzel: What about China? Where are they better off? Are they worse off than they were?
Buchanan: They’re worse off. They’re absolutely worse off. Because there was a bipartisan consensus that was forming on denying China access to computing power. But they had big allies that were saying, You know, let’s sell China the chips. And those allies were having some success. Chief amongst some, Jensen. David Sacks, Trump’s—my successor, as you know—AI adviser to the president, was advocating for this, at least based on the reporting. President Trump had advocated for in some context.
And now the bipartisan consensus has, at least so far, started to harden against them. That is bad for China’s access to compute, and ultimately that is bad for China.
Warzel: Ben Buchanan: Thank you for coming and helping to demystify this; talking about it in these very plain terms. I wanna have you back on, to talk about the regulation part of this, later. But this has been very helpful.
Buchanan: It is great to talk about this subject, and it’s great to have interest in this subject of the chips and chip controls. Because this is a wonky subject for sure—that now has gotten this gloss of a sex scandal and a spy story. But I don’t think it’s an exaggeration to say this is one of the most important subjects in the world for U.S. national policy.
Warzel: And we’ll keep following it. Thank you very much, Ben, for coming on.
Buchanan: Thanks for having me.
[Music]
Warzel: If you liked what you saw here, new episodes of Galaxy Brain drop every Friday. You can subscribe on The Atlantic’s YouTube channel, or on Apple or Spotify, or wherever it is that you get your podcasts. And if you want to support this work, and the work of my fellow colleagues, you can subscribe to the publication at TheAtlantic.com/Listener. That’s TheAtlantic.com/Listener. Thanks so much, and I’ll see you on the internet.
This episode of Galaxy Brain was produced by Hadley Robinson, with story editing by Dave Shaw; the video was edited by Alisa Selman; and the audio was engineered by Rob Smierciak. Claudine Ebeid is the executive producer of Atlantic podcasts, and Andrea Valdez is our managing editor.Learn more about your ad choices. Visit podcastchoices.com/adchoices
Leave a Reply