--:-- Jordan Schneider: Welcome to the emergency podcast. We're bringing back Lennart Heim of RAND and Chris Miller. You guys know who Chris Miller is. To talk the new H20 drama. We went from banning it in April to now selling it with a 15% export fee. All right, Lennart, why don't you start our narratives? What is the H20? Why should you care about it? And what were the first few months of the Trump administration doing when it came to this chip?
--:-- Lennart Heim: The H20 is a chip which NVIDIA designed as a response to export controls in 2023. It's a typical game. You draw some lines and then new chips get created right below the lines. The H20 is such an example, but it did a neat trick. It maxed out the specifications which are not controlled: memory bandwidth. So it put the best high bandwidth memory the world currently has on this chip and created an export control compliant chip. I think it was introduced at the beginning of 2024, a couple of months after the update. It was sold over 2024 with lots of interest.
--:-- When the Trump administration started in January, we can say the Biden administration didn't get to it. I think it was definitely a way out of this problem. Many officials came out, many even favored it, but they just never got to banning it because of many stakeholders, many different opinions, and running out of time. Trump then banned this chip in April 2025, not via the normal means of passing regulation, but using a tool called "is-informed letters" because they're pretty fast. You can send a letter to the companies which produce these chips and tell them, "Hey, you can't sell these chips anymore because we expect there's an export control violation going on and we actually think this chip is too good." So that's where we started. The chip was banned. From my personal point of view, a big success. I think this chip should not be sold. We need to reduce our thresholds. This is just simply too good of a chip. Then in the last few weeks, we saw some flip-flopping back and forth, and every day reveals more information. While we talk, probably more things will come out over time.
--:-- [Audio clip of President Trump plays]
--:-- President Trump: [in clip] On China, if your administration agreed to send the most advanced or advanced NVIDIA and AMD chips... No, obsolete. No, obsolete. Well, then 15 percent of the profits. The H20s? No, no, this is an old chip that China already has. I deal with Jensen, who's a great guy, and NVIDIA. The chip that we're talking about, the H20, it's an old chip. China already has it in a different form, different name, but they have it. Or they have a combination of two that will make up for it. And even then some. Jensen also has a new chip, the Blackwell. Do you know what the Blackwell is? The Blackwell is super duper advanced. I wouldn't make a deal with that, although it's possible I'd make a deal, a somewhat enhanced in a negative way, Blackwell. In other words, take 30% to 50% off of it. But that's the latest and the greatest in the world. Nobody has it. They won't have it for five years. But the H20 is obsolete. It still has a market. So I said, "Listen, I want 20% if I'm going to approve this for you, for the country, for the U.S. I don't want it myself. Every time I say like, '747, I want for the Air Force.' So when I say I want 20, I want for the country. I only care about the country. I don't care about myself." And he said, "Would you make it 15?" So we negotiated a little deal. He's selling an essentially old chip that Huawei has a similar chip, a chip that does the same thing. And I said, "Good, if I'm going to give it to you because they have a stopper, what we call a stopper, not allowed to do it. It's really known as a restrictive covenant. If I'm going to do that, I want you to pay us as a country something because I'm giving you a release. I released them only from the H20." Now, on the Blackwell, I think he's coming to see me again about that. But that will be an unenhanced version of the big one. Like, we will sometimes sell fighter jets to a country and we'll give them 20% less than we have.
--:-- Jordan Schneider: It's crazy to see this discussion at this level of sophistication. Knowing the H20s, knowing the Blackwells by name, knowing what Huawei has. It's clearly a big topic now where this whole export control thing is in the spotlight. So I think it's a good moment to take a step back and look at the arguments in favor and against selling China AI chips. Against selling, you have the argument that selling AI chips helps upgrade the Chinese AI ecosystem that's going to compete with America's broadly. And that there are specific applications of the chips we would be selling to China that we would be very uncomfortable with, like military ones, intelligence ones, or broad human rights violations. Then the arguments in favor of selling include the idea that selling NVIDIA chips would retard domestic chip development, make it harder for SMIC and Huawei to find a marketplace. And the idea that selling chips into China would maintain Chinese dependency on the U.S. stack, keeping Chinese developers using CUDA and building infrastructure around U.S. technology. Maybe we should run through those systematically. Let's start with the biggest one: you shouldn't sell these chips to China because upgrading the Chinese AI ecosystem is a strategic threat to the U.S. I think we'll throw this one to Chris. This is a grand strategic question of how much of China's rise is okay and how much of it isn't. The military intelligence, human rights applications are almost secondary to just how scary you see a richer, more flourishing, more powerful China to be.
--:-- Chris Miller: I would even segment out the richer and more flourishing side and just talk about technological capabilities. They're obviously interlinked, but I don't think U.S. strategy has been to try to make China poorer or less flourishing in general. I think the question is just who's going to lead in AI. The trend over the last five years and really the last 50 years has been that if you want advanced AI, you need lots of advanced computing. There's a small number of companies who produce the chips in question. If you think that advanced technology has mattered in the past in geopolitical competition, which is pretty hard to argue with, it's probably going to matter in the future. Therefore, who wins in AI matters. Just like we would be less happy if we were all using Huawei phones and relying on Alibaba cloud, because there would be significant political ramifications downstream of that. So too, if we find ourselves in a future where either the US or third countries are relying on Chinese AI providers, whether for models, applications, or AI cloud, that implies less political influence for the U.S., a weaker U.S., and a stronger China. Those are the stakes. To some degree, both sides of this argument agree on that basic framing. The question is how best to get there. One argument is that you restrict compute access and thereby hobble the growth of Chinese AI firms. A second argument is, as Secretary Lutnick has said, get China addicted to the AI stack. The question to ask is how addicted are they willing to become? How addicted could you actually make them? Can you leverage that addiction in the future or not? These are where the empirical questions are really focused.
--:-- Jordan Schneider: One more on the in-favor-of-selling argument: this idea that keeping China dependent on TSMC fab chips actually lowers the risk of a Taiwan war. I have some questions about this, but it's something that Ben Thompson has been pushing, which has percolated into the administration and Congress. My mom threw me that one the other day. What do you make of it? For me, it doesn't seem to be the main calculus behind it. I buy it on the margin, maybe a little bit. There are two levels to the question. First, at the political calculus to go to war or not, this would be an extremely weighty decision where the fates of nations would be at stake. Whether or not the chips are there, China gaining or losing relatively in AI hardware strikes me as the 12th thing you would be thinking about if you were a Chinese premier. Domestic political developments, developments in Beijing, developments in Taipei, how willing the U.S. seems to fight for Taiwan, how excited Japan is to let the U.S. fight from its territory—all of those strike me as much more germane decisions. There's a bit of technological myopia from tech analysts thinking the chips are the one thing that's the silicon shield stopping it. If you are a head of state making the biggest decision of your life, I don't think it's going to come down to, "Oh, well, Huawei tells me they can only make 750,000 chips in 2028, so then it's not going to work out." But one level down from that, there is this open question about to what extent the chips and the technology they enable end up reshaping the military balance of power. That is still very much an open question that smart people can disagree on. You can imagine a lot of different crazy futures where AI really matters. It could actually work in the other direction of lowering the risk of Taiwan war if America has a big lead in semiconductors, because a leader in Beijing would look at the military balance of power and the advantage the U.S. and Taiwanese forces get from being more AI-ified and see no chance of winning.
--:-- Chris Miller: The other key facet here is that if you look at sales of advanced chips from Taiwan's ecosystem to China, most of them are not AI chips. It's mostly smartphone chips and PC processors. AI chips are a portion, but a small portion. It also gets back to the question you've raised in a lot of shows, Jordan: how AGI-pilled is Xi Jinping? The answer is it doesn't seem that AGI-pilled. Best evidence for which is that SMIC and its 7-nanometer production is still producing a whole lot of smartphone chips, which you would not do if you thought we were in a race for AGI that will define the future. Both of those facets point against the silicon shield as it relates to AI chips being absolutely central here.
--:-- Lennart Heim: Just to clarify, they're not allowed to produce AI chips at TSMC. They can produce everything else there. Huawei not because they're on the entity list—they did some bad stuff—but almost every other Chinese company can just go to TSMC and produce chips there. So there's a significant flow of chips from Taiwan to China as we speak. Ideally, not AI chips. We had some hiccups in the past where it was also AI chips.
--:-- Jordan Schneider: A key question, Lennart, is how obsolete is the H20 relative to what Blackwells can do, but probably more importantly, what Huawei can do? Walk us through the numbers.
--:-- Lennart Heim: I don't think describing the H20 as an obsolete chip is fair. Chips have many specifications. Let me break it down into two simple ones we should care about: computational power—how many flops it has, how many operations per second it can crunch—and memory bandwidth—how fast you can read and write memory. One of the key inventions we've seen over the last few years, which AMD actually did first, is so-called high bandwidth memory, a pretty complex technology. We've got three companies in the world doing it right now: SK Hynix, Samsung, and Micron, building this HBM. The H20 is really bad on the flops. 7x worse than H100, even more worse, 14x worse than the upcoming chips in the B100s or more. It's definitely not a competitive chip there. But on the memory bandwidth side, which is key for deploying chips, it's pretty good. It's even better than H100 because the H100 uses five units of HBM, whereas this one has six units of HBM. It gets a mind-boggling four terabytes per second of high bandwidth memory. No Chinese chip has such good high bandwidth memory. More importantly, even if they have the 910C, which has some HBM at 3.2 terabytes per second, they're not allowed to buy it anymore. It's been banned since December 2024. Right now, China would be struggling to get their hands on this HBM. They're trying to produce it domestically, but that will take a bit. Even if they produce it domestically, it would initially be worse. So I don't think the H20 is an obsolete chip. It's a pretty competitive chip. It's definitely fair to say it's a worse chip than many others. But on this other dimension of deployment, it's really, really good.
--:-- Jordan Schneider: Some people say the goal is to stop China from training high-end models, so you focus on the flops. If your goal is to constrain inference, you focus more on memory bandwidth. Walk us through the way these different chips are used.
--:-- Lennart Heim: That's a fair debate we should be having. We should think about what we want export controls to achieve. Right now, it's fair to say the H20 is not an amazing chip for training AI systems. There are some things numbers don't always capture. You still build on top of the NVIDIA software stack. If you're a company that used NVIDIA before, it's a pain to switch. There are problems with Huawei chips you don't see in the specifications: they overheat, you need more of them, the software stack isn't great yet, and you can't even get enough. All of these things mean the H20 is not a great training chip. On the training goal, I think it's still being achieved. Where the debate begins is what we think about deployment. If your goal is only to stop them from training big systems, it's really hard to be precise on all of this. AI is ever-changing. The biggest thing we've seen over the last year is the rise of test-time compute—AI models thinking. How do they think? They produce tokens. That's what the H20 is really amazing at. One could say the usability and importance of the H20 only went up because we got models which do more thinking, generating more tokens, and also generating tokens to train the next generation of AI systems. These are the arguments people would make: this is a pretty good chip for producing these new things that are more important in the AI development lifecycle.
--:-- Jordan Schneider: The other argument the president made is that Huawei already makes these chips, which is true to an extent. Walk us through the numbers there as you see them. I know there are questions both about the quality of Huawei's chips as well as the numbers that can be produced. Secretary Lutnick said they can produce 200,000 a year. How does that compare with what we're going to see with H20s?
--:-- Lennart Heim: The key dimensions are quality and quantity. Many always talk about the quality argument. I personally think the quantity argument is way more important. You mentioned a number from Lutnick and Kessler also testified: 200,000 Huawei chips being produced in 2035. How does this compare to the US? I think we're churning out around 10 million chips this year. Significantly more. If we're selling—and I think there have been projections about NVIDIA selling a million H20s—we sell them five times more than what they can produce. This is where the debate starts. The quantity argument is really key here. If you only sold them a couple thousand or 200,000, that's a vastly different debate than selling them a million or potentially even more. The fact that China wants to buy them speaks to their problems with producing domestic chips. On the quantity side, China is simply not there yet. They're getting better and producing more chips as we speak, but there are many difficulties. Even if they produce more chips, do they have enough high bandwidth memory? How good is their smuggling operation to get this memory? How good is their packaging yield? All of these things add up so that you eventually just can't really produce competitive chips. If you compare the Ascend 910C to NVIDIA's best chip right now, the B200, it's way worse on the high bandwidth memory part and on the computation performance. It's also worse than the H20, which we're selling, at least on the memory part. The point is, if you're selling the H20, there's a chip called the H20E. What does it do? It doesn't use HBM3, it uses HBM3e. I previously said it has 4 terabytes per second. If you use HBM3e, you can probably go up to 5 terabytes per second or even more. What indications do we have that this chip is not getting sold? The flops are still being capped, but the memory just continues going higher and higher. As long as we don't have updated regulations, we don't know where the line is going to be drawn in terms of quality of memory bandwidth, but most importantly in terms of quantity. If I could ask for one thing, please reduce the quantity. I think that's the key thing we should pay attention to here.
--:-- Chris Miller: One of NVIDIA's lines that Jensen has been saying a lot is that they used to have 95% market share before the restrictions, and now it's down to 50%. First off, they've never actually given numbers for that. Second, my guess is that they were the only people making accelerators that people wanted. So even if it did go down to 50%, it's not like it was the same pie. The pie went down such that that 5% it used to be now turns into 50% of the whole pie. That number does not tell you that Huawei necessarily has the capacity to fill it up. As Lennart said, Jensen cares about this because lots of Chinese companies are willing to spend—his projection is what, like $15 billion a year in sales? Huawei, Baidu, and Tencent are not dumb. They are going to spend billions and billions of dollars of CapEx. This CapEx number seems small if you're talking about Google and Meta, but it's actually pretty large relative to the total CapEx from the Chinese hyperscalers. They're doing this because they think it is useful and important to their AI ambitions going forward, not to do Jensen a favor.
--:-- Jordan Schneider: Could we talk about who in China will be the large-scale buyers of these chips? You mentioned Tencent, Alibaba, AI firms like DeepSeek, ByteDance—a huge player in China's AI ecosystem. Lennart, do you have a sense of numbers if any of those are public, or at least talk about who the buyers are inside China?
--:-- Lennart Heim: I don't think we have public reporting of who exactly. There's definitely been some reporting that big hyperscalers—the big cloud companies like Tencent, ByteDance, and others—are definitely interested in this. I'm not sure about ByteDance because they're building tons of clusters in Malaysia, which can buy whatever chips they want and just continue building. The normal hyperscalers will continue buying these kinds of chips, but they all hedge. They all also get Ascend chips. They're not stupid. With the policy flip-flopping, they don't know when they're going to get cut off. They're all just hedging with Huawei's chips while they're getting better. We would just subsidize that transition while we do this. That's the thing I'm worried about. Huawei will get better. They will produce better chips. The chips will be significantly worse and significantly less quality than the US, but they will get better. That's the thing we all need to acknowledge. There was a policy at some point which told Huawei they would need to produce their own chips. That's a pathway going down here. There's no going back. The question is what do we do in the meanwhile? How big will the gap potentially be? I'm a firm believer that this will be quite a massive gap, which will have a big impact on the AI competition.
--:-- Chris Miller: That is one of the key lines of debate, but also empirical questions that are hard to research or get hard data on: the decisions of the private tech firms in China—the Alibabas, the Tencents, and others. To the extent that you're right that there's a meaningful quality difference between NVIDIA and Huawei GPUs, they have a strong incentive to build as much as possible on NVIDIA. You can see an argument that says they're going to buy Ascends but put them in the closet or not really take them seriously because they want to build their products. But you're saying that's probably not the case because even those firms, which don't have a strong incentive on their own to help out Huawei, do in the context of potential future export controls and loss of access to NVIDIA chips. The argument that controls align the incentives of Tencent and Alibaba with Huawei and the Chinese state—you think those incentives are already at this point fully aligned?
--:-- Lennart Heim: Yeah, I think so. More importantly, let's walk through the arguments for it. There are arguments in favor of selling H20, and I think that's a sane debate to be had. On the other side, it's sometimes lacking technical details. The market share argument is a fair argument—you want to maintain for NVIDIA a bigger market share and reduce demand for Huawei. I just don't think that's the case. It's an existential priority for China to produce the semiconductor industry. Importantly, it's not like the semiconductor industry only gets better because of AI chips. The majority of chips the world produces are not AI chips. Who's producing at the most advanced node at SMIC, but also at TSMC? Apple. Usually we produce mobile phones there first. They're pushing it forward anyway for the newest Huawei smartphone that will probably produce something like a six-nanometer node, which will then be leveraged to produce better AI chips. Even if you reduce the market demand right now, the semiconductor industry will get better, and these will lead to better AI chips eventually if they transition to this. Then there's the tech stack argument. Sure, we keep them hooked on CUDA. It's a pain to go from CUDA to MindSpore to the Huawei ecosystem. We can model this as a one-time transition cost. Many American companies have done this. Google switched to TPUs at some point. Amazon is using Trainium chips on AWS. They pay a significant cost to switch and run these different hardware stacks. But eventually they do it, and they will also eventually do it with Huawei. It's not like if you use CUDA, your systems are more aligned. If you would sell them AI systems that don't spit out CCP propaganda, I'm in favor of that. That's spreading American values, liberal values. That seems fine. But if you're just selling them chips, there are no values, no constraints that come with selling chips. You can just do whatever you want on it. That's where I'm just like, it's missing this tech stack component. We kind of got it right in the UAE: sell them the cloud, let Microsoft build there. Here we just sell the underlying component that can build whatever they want on top of it. That's just missing in the debate.
--:-- [Ad break for Horizon Fellowship]
--:-- Jordan Schneider: You've made it 30 minutes into a podcast about H20s. Do you want to do this full time? The Horizon Fellowship provides a direct path into the world of emerging tech policy, a fully funded placement at a federal agency, congressional office, or think tank in DC for up to two years. For its 2026 cohort, Horizon is actively seeking candidates with expertise on China's technology ecosystem, its policy landscape, and broader strategic goals. Prior technical or policy experience is not required. The application deadline is August 28th. Check the show notes for a link to learn more and apply.
--:-- Jordan Schneider: I think this is a key aspect to the export control debate, which is fascinating. A lot of people don't get that if you restrict sales of tools, then you hurt the toolmakers, but you help the users of those tools. In the chip industry, if you sell fewer lithography tools, it's bad news for ASML, but it's actually probably good in the long run for TSMC and other companies that face less Chinese competition. Similarly, if you sell GPUs to China, it's good news for GPU sellers, but bad news for US AI firms who face stronger competition. One of the strategic questions is which level do you try to cut off? The U.S. has, until recently, decided to cut off at multiple levels and is now shifting. This week, it seems like it's shifted towards a policy of sell the GPUs, but keep the controls on the chip making tools.
--:-- Lennart Heim: That makes sense. If we were reverse selling them extreme ultraviolet lithography machines from ASML, it would be way more of a rampage than selling them AI chips. I would complain more if we start selling them Blackwells over H20s.
--:-- Jordan Schneider: So I think that's a fair debate we should be having here. People can fall on different positions. And again, we can disagree on some arguments here. You have these different types of controls which stack with each other, and the AI chips are the first ones to fall. I think that makes sense.
--:-- Lennart Heim: One of the arguments is that if you make China addicted to AI chips, you gain long-term leverage. The mental model people think of here is if you get them using EUV lithography tools, they don't have their own ecosystem, and it takes a decade to try to replicate your own tools. So maybe this is going for Leonard. Does the same dynamic hold here, or if not, why? What are the differences?
--:-- Chris Miller: Yeah, there are many different facets of being addicted to something. In the ideal case, it just means all Chinese firms are really reluctant to adopt Chinese chips, right? Therefore, they have less revenue. SMIC is wondering why nobody wants to buy their chips, and instead all the Chinese just buy US chips. I already talked about how SMIC and Semiconductor gets developed anyways independent of AI. But it's a fair thing to say the less people use Huawei's AI software ecosystem, the worse it is. I think that's a fair argument to be made. I just think they know they want to produce it anyways. They know we need our own AI chips at some point. They're not full steam on this. I think they could go strong if they wanted to. Maybe they're full steam on it, but I just don't know for many reasons. So them using the US tech right now maybe delays it to some way, even subsidizes it. Let's just think about Volkswagen, my German heritage. They love to export to China. How's that going right now? Did this stop BYD? Not really. I expect the share of Volkswagens being sold to China in the future will be low. The argument to be made here is they made a ton of money in the meantime. And I think that's a fair argument to be made.
--:-- The reason why I just feel nervous about AI chips is you supplement, you increase the total compute deployment training capacity in the time. If again, let's just say AGI is a singular point, AI is just not going to matter in the next five years, then all we discuss here doesn't matter that much. Because the good thing about AI chips is they get exponentially better. We're not going to talk about H20s in five, six years from now because we have exponentially better chips already here. So I think that's an argument. We can just say, don't worry, we just sell them. We make some money. They get a little bit better AI, but AI is not going to be decisive in the next four to five years. But then later, ideally, we stop it. And we have better chips that are exponentially better. Again, it goes back to where do we draw the threshold? And when and how does AI matter, which is a diffuse question. I have a pretty uncertain view here. I'm just like, AI could be a really big deal in the next three to four years. It seems likely it's going to be a big deal, bigger or less big, depending on how it goes from transformative economic growth to determining the future of the military up to just fizzling out. I think we should address this uncertainty here. I just work on national security risk and I'm trying to minimize downside risks. I don't see the benefits in the long run of why we should sell them. So fair argument. I think there's some good arguments here. But overall, I think it doesn't cut it, at least for me.
--:-- Jordan Schneider: When you look at some companies, it's a really big deal having Chinese market access. For NVIDIA, excuse me, for Intel, 35% of their revenue is selling CPUs into China. This was a big deal for the tool manufacturers. In some years it was 30, 40% over the past few years. NVIDIA is a $4 trillion company. They will be just fine and still be able to deliver you that exponential curve of rapidly improving AI chips even without the extra $10 billion of sales.
--:-- I guess there's the maximalist version of this question: if you are 100% sure that AI does not matter and it's not a strategic technology, then yeah, sell it, go crazy, do whatever you want with it. But it's a tricky line of thought where we're writing an AI action plan where we want to make AI dominance. We think this is going to usher in a new golden age, but we're willing to take some of this downside risk that we're making it easier on a China which we identified as a major strategic threat in order to... I don't know, make some money?
--:-- Chris Miller: Well, if we do think this is a big chip, there is a broader context of the relationship that you can try to trade things in. And say, we wanted them to scuttle some submarines or stop messing with the Philippines. There are lots of other asks you can make from a balance of power, regional dynamics perspective that you could have put on this. And it's kind of wild that it didn't even seem to be in the context of the debate, of the discussion between the US-China trade deal, but was just a decision that Trump made independently because Jensen got to him and we wanted to have good vibes in the relationship. And the 15% tax we're putting on it, is it just going to pay off the national debt? What if it went to buy drones for Taiwan or to shore up funding for BIS so they could do a better job of tracking down all the chips that are getting leaked out into China? There are some lines of this where if you really are going to follow the premise that China is a strategic threat and we got to shape just how much they're going to gain from a relative technological competition perspective, there are other moves you can do to use this card more in your favor than just letting the other side pocket it.
--:-- Lennart Heim: Jordan, that's an excellent point. If you would just have more insight here because what we've seen so far is the H20 got sold again, then post-talk somebody said it was part of the trade talks, others denied it, then the Chinese came out denying it was part of the trade talks. What I don't like about it is export controls were always a national security thing. When the diffusion framework came out and there were certain countries like Poland, Switzerland in this tier two, many were complaining like you're dividing a European trade union. But it's a national security thing. It's not a trade deal we're doing here. This was at least where export controls originally came from. And now we're mixing them with these trade things and now we get 50% of the revenue share and amazing. Let's pay off the debt. Let's do other great things. I don't think national security is for sale here. If we would get other national security concessions in return, that'd be amazing. It would just be nice to hear more and communicate about this. There are definitely things where people like me and others are willing to walk back. Hell yeah, let's sell the H20 because we got a beautiful deal out of it. I just don't think 15% of the sales makes the cut here. Just money doesn't help you. They get the computing power, you make money. You might even make more money because NVIDIA is making money and maybe Microsoft or your favorite hyperscaler in between, while you still have more control and more leverage.
--:-- I still generally think many just miss the point. You don't need chips in your basement to run them. You just access them remotely. So they could literally dial into our beautiful new UAE 5-gigawatt cluster or dial into the US and existing cloud providers. And then in the future, in case they go rogue or you really want to make sure it doesn't go to certain military-linked entities, you just have way more leverage. If we do the concessions, we talked about the different things you want to walk back before you sell chips, just tell them you can use the cloud, which is perfectly legal as we're speaking right now. So if they really want the computing power, use our cloud. It's totally legal. You can go for it. We still make money.
--:-- Jordan Schneider: There's a political economy dynamic here which, Leonard, I think you're referencing, which is getting back to if you sell the tools, you enable the chip maker. If you enable chip makers, that type of competitive dynamic. What we've seen is GPU sellers, NVIDIA most prominently, be very vocal on this issue. We haven't seen hyperscalers be vocal at all, even though I think one should conclude this implies more competition for them. And then we've seen mixed responses from AI model companies. I think Anthropic has been pretty vocal opposed. I haven't seen OpenAI. And so it strikes me that companies that have a lot at stake have been taking very different strategies, some being vocal, some not. I don't know what exactly explains that. More observation.
--:-- Lennart Heim: I got an explanation. Go for it. You know which GPUs they're all using? NVIDIAs. And if you speak out against them, Jensen's going to get you. So if you look at Anthropic, who's slowly migrating to more Google GPUs and Amazon Trainium, you can see the deals. They can speak out against it where everybody else is reliant on Jensen. I can at least confirm from many conversations with many people in these companies, the only competitor here is AMD. The market share for NVIDIA is only going to go downhill from here. The total market will go up. AI is a big deal. But AMD is getting better. Google GPU is getting better. Microsoft chip is getting better. Amazon chip is getting better. We have more and more startups getting better. We just have way more AI chip competition. I think NVIDIA is also slightly nervous on all of these issues. I would love to live in a world where NVIDIA had a smaller market share and see what the hyperscalers and AI companies would say. I think many of them would actually come out. OpenAI at least came out in favor of export controls historically when it talked about energy dominance and more. I think right now they're all just quiet because somebody else might then knock on the door.
--:-- Jordan Schneider: It's fascinating in the context of I've gotten lots of questions of what does industry think? And of course, what you're saying is, well, which part of industry are you looking at? Which segment, which specific companies? Fascinating political economy there. Jordan, back to you. Why don't you do the HBM political economy? This has apparently been reported that the Chinese government is asking for high bandwidth memory as part of concession number two. What does that tell you, Leonard?
--:-- Lennart Heim: If I'd be running China, I would ask for high bandwidth memory over asking for H20s personally. Because I've got my sovereign drive anyway, so I want to build better and better AI chips. If I look at my current AI chip industry, I would like to want EUV, but maybe this is too much to ask for because we did this early on. Trump did it back in the days. But what is the thing we've only recently done is banning high bandwidth memory units. So we got our chip and next to the chip we put the memory, and these memory units being produced by Samsung and SK Hynix and Micron. They're not allowed to go to China anymore. We've seen reporting that the Chinese put forward, could we trade HBM? Is that something we can do here? I hope the US government will draw a clear red line here. We talked about how you would walk back things. There are arguments in favor of selling chips. We talked about them. What we do here is not selling them our chips. What we do here is enabling them to build better chips. The best way how the 910C or the 910D, whatever the next chip is they produce, will get better is by having higher bandwidth memory. Right now, China does not have the capacity to use even HBM3. There's reporting about first trial production of HBM3. In contrast, NVIDIA and others are starting to equip HBM4 and using HBM3 right now. Again, don't get me wrong, China will get better. They will eventually produce high bandwidth memory. There's lots more to be done. We could stop them from producing better memory. But in the meantime, while they're scaling up this production and trying to get better, we should probably not sell them our high bandwidth memory to make their chips more competitive because we might regret this in many years where we're then competing across emerging markets and Huawei has a better chip which can better compete with our chip. The interesting dynamic of the memory space is that two of the three producers are not US but Korean. That's also why we see probably tons of smuggling because it's pretty close by and there are certain tricks to get more HBM. Don't get me wrong, I think China is smuggling HBM right now. We just throw sand in the gears. I'm in favor of throwing sand in the gears. Ideally, we get better enforcement. So yeah, they get less HBM eventually.
--:-- Jordan Schneider: Should we switch to the political economy of this in China? July 15th, we get the news that the Trump administration is letting NVIDIA start to sell H20s. A week later, MSS publishes a notice to the public saying to beware of digital spying via foreign-produced chips. Ten days later, the CAC, the Cyberspace Administration of China, summons NVIDIA representatives over risks of being able to control a Chinese remotely and accuses them of having a kill switch in them. Then we have a private leading cybersecurity research firm in China, Tianxin, they publish a report which goes viral talking about all the ways that there could be backdoors. And 10 days after that, just three days ago, August 9th, we have state television doing a whole big report talking about how there might be backdoors already in these H20s. They cite former ChinaTalk guest Tim Fist's CNAS report on this topic, which he came on to talk about on ChinaTalk a few years ago. So you're welcome. Chris, what's your read on this kind of interesting brushback pitch we've gotten from the central organs about H20s in China?
--:-- Chris Miller: I think there are three potential explanations, not mutually exclusive. One is that the Chinese security services are paranoid and the discussion in Washington of the Chip Security Act, which would mandate geolocation verification, which has been happening simultaneously to the H20 debate, has intensified those concerns. That's explanation one. Explanation two is that it's part of an effort to discourage the private Chinese tech firms from using H20s and that there are people around Huawei or in the government who are afraid that H20s will actually take into the market share. So this is a way to say buy more Huawei chips as well. And then I think the third explanation is that this is actually pressure on US firms like NVIDIA to say, we need you to do more or else we're not going to let you back in the market. We've seen this in other segments of the tech sector where China will ramp up pressure on a private US firm with the aim of having that firm then try to use its resources to shift the debate in Washington. You could maybe envision the HBM debate being part of what China is looking for, the broader trade negotiations that are underway. But it certainly wouldn't be a very attractive endpoint for NVIDIA if they got approval from the US side and didn't get approval from the Chinese side to sell. So perhaps China thinks it has some leverage there. How exactly to attribute these three causes? I'm not exactly sure what the shares I would put on each of them, but all three seem potentially relevant.
--:-- Lennart Heim: Did we talk about they also put out guidance a while ago on energy efficiency? This was shortly in April or May when the H20 was resolved before it got banned initially. They put out guidance that the H20 is famously pretty energy inefficient if you look at flops because of the export control bandwidths. Basically, I don't know exactly what this guidance means, but it discourages companies from using it. I guess nobody's been following it because now they're buying it in the single millions, this chip. But I think it feeds into the same narrative here. You try to push certain companies or you create artificial demand for Huawei chips and slowly tell them, hey guys, at some point we want to do our own AI chips. As Chris was saying, I think all of these stories are simultaneously true. There's no big downside for them to do these kinds of things, only benefits.
--:-- Jordan Schneider: Actually, there's a same media source, I don't know if it's one of you referencing, Jordan, but one of its criticisms of the H20 was that it was environmentally unfriendly.
--:-- Lennart Heim: Yeah, they cite this exact NDRC line that Leonard talked about. The goal is five teraflops or half a teraflop per watt and the H20 can only give you 0.37. It's pretty bad. Pretty environmentally unfriendly for training but pretty damn environmentally friendly for deployment of AI chips. Way better than any Huawei chip. I can tell you that.
--:-- Chris Miller: I think here's a moment where some mirroring might be in order. We've just had an hour-long conversation about how messy and convoluted and how many conflicting priorities we have in American policy towards artificial intelligence. The same thing is happening in all these different ministries in China. This is big news. This is a change in the landscape and people want to have their say and make their stamp on it. You don't necessarily need to attribute some 4D chess of "oh, now's the time to squeeze them." But I'm sure the people in the MSS read Tim's thing and are like, "This would be really stupid if we bought all these chips only for them to turn into bricks or spy on us or have bombs in them that are going to blow up like beepers in Lebanon or something." I'm sure folks in CAC feel the same way. And then the same debate that we've been having for the past hour of "is it net positive or net negative for domestic self-sufficiency to have a competitor to Huawei potentially take a big chunk of the market domestically" is something that's being played out in China.
--:-- I agree with you that at a broad level, now is the right time to ask for more stuff from NVIDIA. Now that they've gotten the green light and there is this $10 to $15 billion of demand and these chips sitting on a lot somewhere in Taiwan that they're really excited to ship out, to say, "Hey, you better step it up or cut the price or do an extra screen to make sure there aren't any kill switches on it or whatever." The way this is playing out on Twitter is, "Oh, China's saying they don't want them. That means we should definitely sell them." I don't think it is necessarily dispositive just reading that Chinese state media or state organs are saying something means it is true. It's not that hard to play the "let's not even give this credit for 4D chess. This is just 2D chess of saying, 'Oh, no, we're worried about the chips. We don't even want them.'" That changes the political economy of the debate in Washington, where it makes selling these chips potentially easier to go down. So that's also something to watch out for as we see the Chinese government saying, "Ah, no, we didn't really want these all that much. This isn't actually a big concession. We're kind of worried about the second order effects of this." But the fact is, the demand is not going anywhere. It's not like Alibaba is not going to buy these chips because of these sort of warnings. Alibaba would be pretty sad if they suddenly only need to rely on other inferior chips where they can't produce enough of them.
--:-- Lennart Heim: Ideally, if I were on the Chinese government, I would put out so many regulations that I can sell all of the Huawei chips I can produce and then fill the rest with some nice NVIDIA chips here. But I think also what's interesting is there's just some misunderstanding of what the Chip Security Act is supposed to do with location verification. The idea is not to check if a chip is in China and then we have a problem. The idea is we put this tracker on a chip in Malaysia, Singapore, wherever you think they'll be smuggled, and then check they don't end up in China. This was never supposed to go on chips which go to China because ideally we don't have any chips going to China, at least not the advanced ones. This whole debate of these hardware-neutral mechanisms and location verification was big in the UAE and Saudi Arabia and Singapore and Malaysia for all of these smuggling hotspots which people were worried about. Some people have been pushing, and I think if you think about if we now stop selling chips, I argued we should sell them cloud. That's one line. People got to say, we can sell them chips, but put something on the chip. But just knowing a chip is now in China and we know it's in this location and this city over there, how does this help us? Everybody can dial in remotely, even if it sits at Tencent. Who says that the PLA is not using it? You can just dial in remotely. So I don't know what's going on there. If there's some misinterpretation of documents, just some vibes, it's a confusing status.
--:-- Jordan, you previously made a point about Intel, which I think is an interesting one. We were saying Intel made a lot of money in China, right? Intel is still allowed to sell their CPUs, and Intel's CPU share in China is only going down. I think we will see the same with NVIDIA and AI chips. Even if you're allowed to sell, your share will potentially be going down. Why is this the case? There is similar guidance, for example, for all government computers to go to homegrown domestic produced chips. We can't trust Intel anymore on this. We will see the same on AI chips. So yes, Intel made a lot of money in China, but the share is going down over time and they're pushing on the self-reliance to produce their own AI chips. They also named security concerns here. That's why the government is coming first. I don't know the exact numbers of Intel's sales right now in China and how much money they're making there, but I'm pretty confident it's been going down and the government is not buying any more Intel chips because they just put out this guidance. So we've seen this playbook playing out before. The only difference is now we have this confusion with which chips are allowed to be sold, which ones are not to be sold, and how good are they actually. But the story is nothing new.
--:-- Jordan Schneider: Could we talk about what we know in terms of the big buyers of AI chips in China and their relationship with the state? You've got the private tech firms, Alibaba and Tencent. You've got the AI labs, DeepSeek most prominently. I don't know what Leonard or Jordan would have a view here, but one of the key questions seems to be what is the relationship with the state today and how is it changing? To what extent should we see them as an arm of the state is certainly not accurate. Totally independent is certainly not accurate. There's a spectrum. So to what extent are these political priorities shaping their procurement decisions?
--:-- Chris Miller: There was some reporting which was clearly sourced by the intelligence community over the past few years that after the MSS hack of the SF-86, that's the form you submit to the US government when you want a security clearance, which talks about your confession of sorts to the Catholic church where you talk about all your divorces and all your debt and everything that a foreign intelligence community might want to know about you. That data was actually that the MSS tapped Alibaba and ByteDance engineers to put into a more useful format. We've just seen over the past few weeks reporting from Business Insider about a public tender from some corner of the PLA that wanted H20s to do whatever they wanted to do with it. I used to be more sanguine on this type of thing, but I think this is the dual-use technology to beat out other dual-use technologies. So it just seems to me to be a little preposterous that insofar as this is a strategic resource that the Chinese government would not be able to leverage data centers that are living in China that the US did not have any sort of kill switches or on-chip governance on in order to do whatever they want to do with it, whether that's build a surveillance system or help with weapons manufacturing. To be clear, the Pentagon has now signed a $200 million contract with OpenAI. This is just the beginning. So clearly this stuff is useful that we're willing to pay a lot of money to get it into the Pentagon in one form or another. So if what Leonard says is convincing is that you're selling a lot of H20s materially raises the amount of usable functional compute that can be put into anything in China, it would be really surprising for me if you didn't have the Chinese government want to take out these new tools for it. If you didn't have the Chinese military police complex want to take these new tools out for a spin.
--:-- Jordan Schneider: I think there's kind of two points you need to add. One is, if AI tools exist, will the military use them? I think obviously the answer is yes. But if you're a data center procurement official or executive at Alibaba Cloud, to what extent is your decision making there shaped by what you read in state media versus what your boss tells you to build an effective cloud, in which case maybe H20s are your best option versus the sense. How do we think about that? Because those are the people who are going to decide how many Ascends to buy, unless they're getting a dictate from the top.
--:-- Chris Miller: I guess the sort of counter example I'm thinking to myself is, there was a time when parts of the US military were using Chinese drones. Not because there was a policy to use Chinese drones, but because they didn't have any US drones. So is there a scenario in which your procurement executive at Alibaba is just going to try to ignore the sounds because they were told to build a good data center?
--:-- Lennart Heim: Yeah, I think at some level, yes. These are companies that report quarterly earnings that pay their employees based on how well the company performs. People get stock options. So I think by and large, the incentives of the people who are buying these chips is to drive the most revenue for the money you're spending on your CapEx. But it only goes so far. I do think there is this broader strategic realization, which you don't need Beijing to tell you, that this door can be closed at any time. You really ignore Beijing. I think we've opened it to be closed already.
--:-- Chris Miller: I think maybe now is an interesting moment to talk about the sorts of things that could change the dynamic we're on now on chips and on the broader US-China relationship. We have Congress as a variable. There have been a number of senators and congresspeople who've been like, "Wait, what are we doing selling these chips to China?"
--:-- Jordan Schneider: I thought we banned that and said it was our golden ticket to the 21st century. And then I think just because Trump is doing this at such a personal level, like we've seen him turn on Putin. We've seen him go from all Putin to "we're going to ask some questions about this guy." We'll see what the hell happens in Alaska. But I do think that there is like Jensen saying the wrong thing, taking too much of a victory lap, or she doing something really obnoxious. There are a lot of interpersonal dynamics that could change what the Trump administration ends up doing, which is probably the more relevant variable than whether or not Lennart can convince you that Huawei can only make X amount of chips.
--:-- Chris Miller: It's an interesting moment in time because we have all the trade negotiations. Everything is volatile and certain things are just on the table and they'd be willing to discuss them. We see the Chinese bringing forward, at least according to reporting, the idea of HBM. It will be interesting to see what the government is going to say and where it's going to draw a red line. We had statements before the trade negotiations in London that H20 is above the red line, so they wouldn't negotiate it. Again, we can all try to put together the story of what happened here, but we won't know for sure. There will be more discussions about these kinds of things. The Chinese can bring it up.
--:-- Lennart Heim: But I'm also more interested in the semiconductor manufacturing equipment companies. If NVIDIA got this beautiful deal, I know what they're doing. They're all trying to give the president elsewhere a ring. It seems like a handful of people are making these decisions, and I just hope they're well informed of which things are more important. If I see any news about EUV machines being sold to China, I'm probably going to get a heart attack. I just really don't want this to happen. From a personal transaction perspective, there isn't someone in the semi-cap equipment ecosystem that Trump's going to give the time of day like he felt he had to with Jensen, because Jensen is America's most important CEO. I don't think any of those folks have the sort of panache and skill to make it work. Even Ben Thompson, who I gave a hard time to earlier in this podcast, understands very clearly that there's a lot of downside risk in selling more tools to China than we already have.
--:-- Jordan Schneider: I would go even so far as to say it wouldn't be good for Jensen if Huawei is good at producing AI chips. So wouldn't it be in their interest to say, "Hey, yeah, let's make sure we sell them all chips. Let's make sure to hit them on every single dimension we can to make sure Huawei is just less competitive"? I would love to see that. That would be at least a good part of the story here.
--:-- Chris Miller: I think going back to Congress, Congress will be interesting to watch on this issue. The trend in Congress has been to vocally push for tougher controls, both in the first administration and under Biden. Not universally, but that's been the predominant push. So now it will be interesting to watch Senator Cotton, for example, and what he does or does not say publicly on this issue.
--:-- Jordan Schneider: Chris, do you want to tease out the Russia comparison a little bit? Congress was really not happy. They ended up putting some sanctions on the table. What have the dynamics been there over the past six months?
--:-- Chris Miller: Well, the last six months in Russia have seen Congress officially not play much role at all. They put sanctions legislation on the table and then pulled it back after Trump requested it. But I would say there have been a number of Republican senators who have been influential in shaping Trump's thinking. Lindsey Graham, for example, seems to have played a role in shaping Trump's thinking on Putin over the last six months and the way Putin is stringing him along. We're going to Alaska later this week, so maybe all that will prove irrelevant if Trump changes his mind. But it does seem like you could argue that even though Congress has done nothing on Russia, it has helped change thinking in the White House. I wonder if this will be true here, but this seems like a place where Trump is going to make more of his own decisions, especially as far as it intersects with the China trade negotiations, which it seems like it may well.
--:-- Jordan Schneider: Yeah, and it's a less salient thing than a land war. There's no domestic constituency. Just weirdos with tech national security podcasts. Other stuff we should get to? Lennart, do you have anything to say about the fact that before this week it was reported that NVIDIA was coming out with a downgraded version of some new downgraded chip post-H20? I guess that's now irrelevant.
--:-- Lennart Heim: Oh, yeah, the B40 or B30. That's now irrelevant because of the H20. It's unclear. If people flip-flop the decision on the H20, notably there is still a license requirement. NVIDIA got the license granted. If they wanted to go all the way back, they could have removed the license requirement. From October 2023 to April 2025, there was no license requirement. Then they introduced the license requirement, which is still intact. The only thing that happened as of speaking last Friday is they granted the licenses according to reporting. If they still want to sell a chip which is not subject to export controls, they would produce a new chip called B30 or B40. It needs to lower the computational power threshold, the same as the H20, and also have lower memory bandwidth. According to reporting, I think the FT leaked what is in the S-Inform data, it needs to be less than 1.4 terabytes per second memory bandwidth. The H20 is at 4 terabytes per second. The B40 would then probably not use HBM anymore. It would probably use an inferior memory technology, but significantly cheaper because why use HBM if you can't have that much memory bandwidth anyway? That's called GDDR technology, which we usually use for our graphic GPUs.
--:-- Jordan Schneider: If people talk about this being only the fourth best chip, I don't think the H20 is the fourth best chip. I think the B30 or B40 is a more fair description of a fourth best chip.
--:-- Lennart Heim: I would still not call it an obsolete chip, but it's definitely a worse chip. It's a chip where the US government at least decided, "Here's where we draw the new lines. This chip is fine to be exported before the license." So it could still be coming. I have not heard they're stopping the production there yet. I guess NVIDIA is making a calculus right now on how much demand there is, but it's clear the H20 is better. The question is, will all the licenses be granted going forward? Trump said at the press conference a couple of days ago that he'll consider a downgraded Blackwell.
--:-- Chris Miller: Are there ways we should think about what that might look like if, in fact, it materializes? With huge questions over whether or not that's actually real.
--:-- Lennart Heim: One thing that stood out: he said 30 or 15 to 50% less performance. What many people are missing about AI chips and computing is that chips get exponentially better. If your chip is 15% less, that's nothing. That's still the same generation. If you really want to sell worse chips, you need to go back a few generations, and then the chip needs to be like seven times worse, not only 50% or 15%. There's an argument to be made that you want to sell worse chips, but it's not a little bit of a downgrade. We really need to take the exponential into account. If we trim down a Blackwell chip, for example a B200, by 15% to 50%, it's still twice or three times as good as a Huawei chip. And we can produce millions of them, but Huawei struggles, according to reporting, producing 200,000 this year. That's a key thing to get right here. People need to keep in mind the exponentials. Chips get exponentially better. A 15 to 50% trim is nothing in the grand scheme of things. I would make my voice heard to say this is probably not a good idea of what you'll be doing here. The government drew lines before, and the lines are way lower. I think that's where they should be.
--:-- Jordan Schneider: All right. Well, it's been a long road since October of 2022. Thanks for sticking with us, everyone. This information is for educational purposes only and is not a recommendation to buy, hold, or sell any investment or financial product. This podcast has been produced by a third party and may include paid promotional advertisements, other company references, and individuals unaffiliated with A16Z. Such advertisements, companies, and individuals are not endorsed by AH Capital Management LLC, A16Z, or any of its affiliates. Information is from sources deemed reliable on the date of publication, but A16Z does not guarantee its accuracy.
