Dmitri Alperovitch

Thanks for having me.

No, I don't believe there's any evidence of that. In fact, I believe they're actually working. And they work even better if we were better at enforcement and were more extensive with these controls. So, first of all, China did not, you know, usually this conversation is about China, right? And preventing China from getting dominance in this critical area of technology. And China did not start their production of semiconductors or massive investments in this field in 2018 when the first export controls started to go up or in 2022 when more of them came online. This began really in the early 2000s with the establishment of SMIC the Chinese main semiconductor manufacturing company. And obviously even earlier in the 1980s with the establishment of Huawei which later went into the chips manufacturing and design phase. But particularly with Xi Jinping coming to power in 2012, he initiated massive investments in this field to the tune of hundreds of billions of dollars predating any even conversations about export controls to China because at the time we were still in this friendly partnership competition phase. You know, President Obama in 2015 announced standing next to Xi Jinping in the Rose Garden that we welcome China's rise. A lot has changed in the last 11 years since that statement was made. But even in 2015 when Xi Jinping put together his Made in China 2025 plan, he announced that independence in design and production of semiconductor was a critical national security priority for China. So predating any export controls, right? China did not need any incentives to focus on this area and they had started on this long before.

Now let's talk about those export controls. So we start actually with equipment. During the first Trump administration, we realized that we have an incredible choke point in the west on equipment that is used to manufacture chips and particularly the more advanced chips because they're basically made in three countries. They're made in the US. There are three companies in the US primarily focused on this field. Applied Materials, Lam and KLA. There are a couple companies in Japan, Tokyo Electron being one of them, Nikon another. And one company that many people have heard of in Netherlands called ASML that manufactures the only company that manufactures extreme ultraviolet lithography. The other companies are focused on other parts of the manufacturing phase of semiconductors but you need them all to actually manufacture chips and particularly advanced chips. And the first controls started going in place on export particularly of the EUV technology to China in 2018. And since then Chinese have been really stalled in their ability to get to the advanced nodes. They managed to get to seven nanometers using really known process because TSMC of course the largest semiconductor manufacturing company in the world based in Taiwan showed around that time around 2019 that you could use the older generation of equipment called DUV deep ultraviolet lithography to manufacture 7 nanometer nodes and about 5 years later the Chinese got there as well using the DUV technology that we're still allowing to be shipped to China. They've done some experiments at 5 nanometers, but nothing that has been as far as we can tell productionized or at any significant yield levels. And the same is true, by the way, at 7 nanometers. The yields are very very bad with those process nodes from what we're able to tell. But they can go to three, they can go to two, right? The latest cutting edge nodes. And with the EUV technology, they're unlikely to make more progress on that. So those export controls have actually worked in largely being able to stymie China from producing advanced chips. They would work even better if we had stopped them from purchasing the DUV machines from ASML. Because then they wouldn't even be able to manufacture 7 nanometers. But you know we were myopic and did not go far enough.

And then with regards to restrictions on the chips themselves and those started going into place under the Biden administration in 2022, particularly the more advanced compute chips, the GPUs that you can use for training and running inference on AI models. They have unequivocally worked because we have statements literally from leading Chinese AI companies saying they're compute constrained, that the lack of sufficient chips in China is the biggest impediment that they face. Now again, we were half-hearted in those efforts because we said, 'Well, we won't let you get the very advanced chips, the most advanced chips, cutting edge chips, but we'll let you get one generation behind the so-called A100s and H20s and others from Nvidia.' And those are actually good enough in many cases. They're not as great as the Blackwells and the next generation of Rubin that's coming online from Nvidia, but they're pretty good. And the leading Chinese AI models are actually running on those chips. In fact, you know, when a year ago, almost exactly a year ago, everyone was freaking out about DeepSeek and the progress the Chinese have made, very few people actually bothered to read the paper that DeepSeek put out saying that they basically trained their models on Nvidia A100s and H100s that had been allowed to be sold to China. If we hadn't sold those chips, they wouldn't have made as much progress. So those export controls have worked. They would work much much better if we tighten them up if we don't allow any GPU exports to China. If we don't allow any equipment export to China. And we would be much much further ahead.

No, look, you know, to make advancements in AI, you basically need four things, right? You need talent, which China has plenty of, right? And they have lots of people, lots of smart people. So, we can't stop that. You need energy, which frankly they have more of because they have fewer restrictions on being able to put up nuclear power plants or coal power plants and the like. You need data, which they also have plenty of. You know, anyone can crawl the internet. So, that's not a limitation. And the fourth one is you need compute. And this is the only thing that we do have a choke point on China because today the United States is the one that designs almost all advanced GPUs, right? Whether it's Nvidia, whether it's AMD, whether it's Google, whether it's Amazon, and increasingly Microsoft and Meta getting into the space as well, the design is all happening by American companies and there's one company that manufactures all those chips and that is TSMC. Now there are experiments with Samsung in Korea and Intel increasingly wants to get into this game as well but effectively right now it's TSMC and maybe soon a few more Western companies. So this is a choke point we have, the only choke point to curtail China's advancements and make sure that we remain in the lead.

And why is it important for us to be in the lead? Well, two things. One, there's undeniable evidence that AI is going to transform virtually every aspect of our lives, but the first one or one of the first ones that it's going to transform is combat, is war, right? The military is one of the earliest adopters of AI. You've just had an announcement from the Department of War saying that they're adopting Grok. They're already using Gemini and will be using all really leading edge models. It's not just for autonomy. It's transforming logistics. It's transforming how we train the force. It's transforming how we organize. Lots of applications. And the question is, do you want the Chinese military, the PLA, to benefit from those technologies given that they are the primary adversary, as stated in the national security strategies of the United States, not just the last one, but really all of them since the 2017 strategy from the first Trump administration. I think the answer to that is no. And then outside of the military, do you want the Chinese economy and their manufacturing sector and their technology sector and all other areas to benefit from this technology more so than we benefit from in the United States and to better compete against us and to continue losing our market share to them. I don't think that's a good thing either. Right? So, I think preservation of our lead in AI is extraordinarily important for both national security and economic security. And we do have this choke point. It won't last forever because the Chinese will eventually figure this out. But as many years as we can get out of it to establish that lead, to establish the dominance of US models, to establish the standardization on those US models, the better off we are.

Well, I think you know in 2018 when we started putting restrictions in place on EUV ASML machines, we could have said we'll put restrictions on all western equipment from Netherlands, from Japan and the United States. None of that should be going to China because why would we enable China to produce chips at any node sizes? How is that helpful to us? Because we know what they do. Aside from shipping it to their military, they also engage in dumping and try to use that tried old playbook that has worked since the 1980s to subsidize industries that engage in dumping to destroy western competitors and then establish dominance in market share. They have that now in critical minerals and particularly in rare earths. They have it in solar. They have it increasingly in batteries and EVs and all sorts of energy applications. Why would we enable that in chips? Hugely myopic.

And then on GPUs in 2022 and later in 2023 when the export controls came into place, you know, the Biden administration had this policy articulated by then national security adviser Jake Sullivan of small yard high fence, right? This idea that we're going to define a very very small area of technology that we're going to restrict from being exported to China. Have a high fence around it, but it's going to be very small and everything else is going to be allowed. I think that was myopic. I think we should have said let's enlarge the yard, make the fence even higher. But why would we sell any compute to China particularly GPU compute that can be used for inference or running or training of AI models, any of them, right? Even if it's a generation behind what the US companies use because you know that will drive more efficiency from them, they'll figure this out and even if it's slightly worse, you know, they can open source these models and have the world use them even if they're not as good as the frontier models from the US which is exactly what's happening today, right? China is maybe 12 to 18 months behind because we sold them those chips. They would be years behind if we hadn't.

Yeah. The Trump administration right now is trying to do this initiative they're calling PC Silica. Jacob Helberg from the State Department just announced it a few weeks ago and basically that's the idea that you are going to build a coalition of western and middle eastern countries for AI supply chains because we do have one problem with AI today in the US and that is energy, right? We don't have enough energy, we don't have enough data centers, energy costs are not inexpensive here. So it does make sense to partner with countries for example in the Middle East where they have plenty of energy. Maybe others as well. But making sure that there are strict controls on those data centers and the chips going into those data centers to make sure the Chinese companies can't get access to them, right? Because even if we're successful in blocking export of chips to China, if the Chinese companies can just rent space in global data centers to train and run their AI models, then it kind of defeats the purpose. So you want to make sure that they are not able to do that and use that loophole. But if you can do that then it does make a whole lot of sense to bring partners closest to us and train models together and use that energy capacity and frankly other advantages like innovation of talent that you can use in Europe and elsewhere to help accelerate western dominance in the sector.

Well, look, since 1945, we have enjoyed incredible advantage that the United States military was the most powerful force on the planet, right? That was securing not just the national security of the United States and our territorial integrity, but also the trade routes that we use around the world as we import and export goods across the globe. If China wins the AI, I think it would give them a huge leg up first and foremost militarily. And do we want to see the PLA becoming the world's preeminent military force? Because we see how they operate, right? Look at South China Sea. Look at East China Sea where they bully countries like the Philippines, Japan, Taiwan, right? To try to enforce their will, right? They're bullies. And you know, some people will say, well, the United States bullies, too. You know, if you're coming at it from an American perspective, you know, you want to have the strongest military, you want to be able to set the rules of the road for the rest of the world. And what we don't want to do is have China do that and bully our partners and allies around the world and dominate the sea routes that we use currently to ensure the economic security of the United States. Right? So that's first and foremost.

Secondly, because by the way, you know, the Pentagon, the White House calls China the pacing threat as we speak. The United States military is preparing for a war with China. So, do we want China to have the world's greatest military that could actually defeat us in a conflict, right? We would face pretty dire consequences should that happen. And secondly, do we want China to become the dominant economic power in the world? Today, the United States is still by far the largest economy. We're 25% larger in GDP terms, nominal GDP terms than China. At their current growth rates which have fallen down to single digits, it's mathematically impossible for them to catch up to us in the foreseeable future. But AI could change all that, right? The AI and the productivity improvements that they could get from that could dramatically accelerate their growth rates. Given that they're engaging in non-market competition practices, engaging in dumping, engaging in theft of intellectual property, bribery of countries and companies around the world, it's very very hard to compete when they also have really really good technology. Right? So do we want to lose market share around the world? Do we want our companies to be less competitive? Right? We've seen what happens in industries, right? Telecom is a great example. We used to be the leader in telecom technologies in the 1990s and then Huawei came online. They stole some intellectual property as has been proven in some court cases. They've started using the support and subsidies from the Chinese Communist Party including the bribery that they've engaged in around the world. And now much of the world is running on Huawei telecom infrastructure and there's virtually no US competitors that are left in that space, right? All those companies Nortel in Canada and others in the US have gone out of business or have shut down those divisions. And you've had that happen now in solar, you've had that happen until recently in critical minerals and many other areas. Do we want to see that across our entire manufacturing base, our entire technology base, right? Do we all want to use just Chinese produced goods and technologies and be essentially beholden to them? And we see what happens when they corner a market. They've done that just in the last year on rare earth, right? Where they have over 90% of the world's refining and processing capacity on rare earth. They started using that as a bully tactic essentially against us, against Japan, against other allies to try to enforce their will. Imagine if they have even more power in all these other sectors because of the advantages they're going to gain from AI. That's the world that I don't want to live in. I don't want other Americans to live in. And that's why it's so important to curtail China's advancements in AI.

Look, we've kind of been zigzagging on this policy. So when the president first came into office, he announced export controls on the H20 chips that the Biden administration had not blocked, these Nvidia GPU chips, kind of an older generation. And then he reversed himself and then just last month he announced that he will allow sale of pre-advanced H200 chips, even better chips to China. I think those policies have been a mistake. I think that we should not be selling any GPUs to China. We should be ramping up enforcement to make sure that they do not smuggle those GPUs from other countries, that they don't use GPUs in data centers around the world where we ship them, that we enforce strict KYC know your customer rules in those data centers if they're going to receive our chips.

And then on equipment, I would cut off all equipment, any manufacturing equipment going from the United States, Netherlands or Japan, which we can do by the way pretty easily because we have a powerful weapon in our arsenal called the foreign direct product rule wielded by the commerce department which basically says that any technology made anywhere in the world by any company in the world that has at least a certain amount of US IP, US content, we can enforce global restrictions on its export anywhere in the world. And basically ASML and the Japanese companies all rely on US technologies, on US chips and other things. And we can basically tell them you shall not sell to China. And by the way, you shall not maintain the existing machines that you've already sold to China. You will not resupply them with parts, do any sort of maintenance on those machines. And we can cut off at the knees the Chinese semiconductor manufacturing industry which would benefit the western sector as we're trying to invest in the sector, right? We've authorized $52 billion in the last four years on the CHIPS Act, right? To invest in manufacturing here domestically. Those companies are not competitive with Chinese production because of the subsidies and the dumping that the Chinese are engaging in, right? So if we want that industrial policy to succeed, you have to make sure you're not also enabling your enemy. Like, why would we do that? Why would we essentially sell the equivalent of modern weapons, which are chips and equipment to make chips, to our primary adversary as defined in our consecutive and bipartisan national security strategies? It is just absolutely myopic and ridiculous that we're engaging in this behavior.

Right. During the Soviet Union, can you imagine having a public debate about whether we should sell supercomputers, essentially GPUs or the modern day supercomputers to the Soviet Union? You'd be laughed out of the room, right? Even though you could say, 'Well, these companies need more money because, you know, they could get access to the Warsaw Pact countries and invest more in R&D. And by the way, if we cut them off, they'll invest in their own production and R&D, and they're going to compete with us.' I mean, same exact arguments could have been made, but no one in their right mind would have said that, right? And by the way, why stop there, right? If the same strategy applies to chips, why not say, 'Well, we don't want the Chinese military-industrial base to be so advanced, so why don't we sell them aircraft carriers? Why don't we sell them F-35s? Why don't we sell them, you know, our advanced destroyers and other weapon systems so that they can get addicted to the American military tech stack,' right? I mean, that is ridiculous. Of course, but the same ridiculous argument has been made on chips that is not being challenged as much as it should be.