MATCH法と中国のDUV輸入：液浸リソグラフィ規制は米中AI覇権の決め手か

It is a button that lets you win the AI competition between the U.S. and China. The U.S. and Japan can jointly have the effect of controlling all immersion systems to China, but China has recognized that we have a big cannonball-sized hole. It's not really a loophole. We call it a cannonball hole in our export controls, and they've taken advantage of that. But we have deposition and etching companies, and we've handcuffed our own with the expectation that Japan was going to follow suit, and they haven't, right? So there's a fairness. I don't know, Ockham. Fair's got nothing to do with it. Fair ain't got nothing to do with it. Truly, if you are seeking any form of strategic advantage, the foundation of that advantage is indigenous chip making. And the foundation of that is lithography. Let's rock and roll, baby. Let's pass the match back. DUV exports to China. Why should you care about lithography machines that get sold into the PRC? What is America or what is Congress in particular thinking about doing about this question?

And what does it mean for the future of humanity, light cone, great power relations, all of the above? Nick Brown joins us, a first time China Talk guest. He is a co-founder of the Center for Technology and Statecraft, a new think tank spun out of IFP. that's looking at long-term technology challenges and opportunities that the U.S. faces. Nick was a physics major, spent a while as a software engineer in and around government, also had an arc as an aspiring concert violinist. But we're here to talk not about that, but about what is in pole position for my paper of the year, his master class at exploring what would happen if the U.S. cut off DUV exports to China. Nick, congrats on the paper. Congrats on the think tank. Welcome to China Talk. Thanks. It's good to be here. So just to set the groundwork a little bit,

why should we care about how much compute China has in relation to the rest of the world? Yeah, I can take a stab at that. Ben Akib, you're welcome to. You know, ultimately, compute is one of these fundamental drivers of AI capability. You know, there's compute algorithms and data, and there's been a lot of work by a lot of people over the past, really, decade that has sort of established that of those factors, compute is one of the most, you could say, governable, is the one that is most possible to intervene and sort of shape where it goes. And then those factors combine to create AI capabilities, both through training where you develop the model and then where you actually deploy it out in the world. And then you realize those capabilities in some kind of context could be economic, governmental, security related. So insofar as you care about the exercise of AI capabilities, which I think is fair to say many do.

And if you care about the degree to which China is able to exercise those capabilities, particularly in comparison with the U.S., compute, which is the number of chips, the quality of the chips, the aggregate computational stock that's available to them, is one of the fundamental drivers that you might want to examine or push one way or another. akib from trying to talk co-hosting let's do a bit of the tail of the tape as of october 2026 how much compute does trying to have and how much compute does the rest of the world currently have access to yeah i mean i really appreciate the work that you guys sort of cited on this uh at least according to the median estimate i think first of all we have to understand that everyone's numbers are really hazy here uh but it seems like the u.s actually has a pretty commanding lead over China, right? As long as you're measuring an H-100 equivalence, you could say that China currently has maybe around a million,

whereas the U.S. has upwards of 30 million. So, you know, I agree. Compute's really important, but Nick, I thought we got them. We got EUV. So, why do we have to do more? I thought we're winning. Yeah, I assume that's sort of my tag-in to talk about DUV, a lithography. So I can maybe give a little bit of a background about... Wait, wait, sorry. Sorry, it's 30 to... We're not at 30 to 1 because you got to... There are asterisks with the remote control. Yes, yes, there are asterisks. Yeah, so 1 million Chinese design shifts in China, we so estimate. But at least when Nick Corvino and I ran some numbers, there's also almost another million in remote access computes that China can access in data centers in Thailand, Malaysia, usually in Southeast Asia. But again, it goes to show how hard this is to measure. But anyways, for like a worst case, we're still at a 10x advantage currently as of fall 2020.

So anyways, sorry, back to you, Nick. I would actually maybe just not to distract from my own line of thought, but just to comment a little bit about that. I do think that the distinction there is quite material. And that's between what you might call remote access chips, sort of one segment. And then exported chips is another segment. And, of course, there's been lots of the chip export wars continue to rage. You know, NVIDIA's H100 and then H800 and then H20 and then H200. And there's lots of debate that goes back and forth about that. I would nonetheless sort of segment that out from China's indigenously produced chips. And ultimately, the difference between all three of these categories comes down to leverage. For remotely accessed chips, there's quite a lot of leverage. And then even within that, you might distinguish between chips that are accessed in the U.S. versus elsewhere. The U.S., its companies, its entities that answer to it as a sovereign entity,

have a lot of leverage on how that compute is used. They could cut it off at any moment. For exported chips, there is leverage sort of at the point of export. Whenever you decide to export the chips, at that point you let go of that leverage. But in terms of the production of those chips, that's still within your grasp. Indigenously produced chips, there is very little leverage. Ultimately, at the end of the day, if China can make its own chips with its own supply chain and it can make as many as it wants, it just means that the compute choke point, as a governance choke point is just no longer really effective, at least for international shaping of power. And that is why it is so critical to look at China's indigenous production while we debate about how many chips to actually export to China. So I just agree that lithography and compute is really important, but I thought this is a solved problem. We already banned EUV, so I don't understand what else we got to do.

What's the deal with the EUV? Yeah, I'm glad you asked. maybe I can give a little bit of background, and then we can sort of get into the meat of the issue. So EUV and DUV are two different sort of technology generations of lithography. Those stand for extreme ultraviolet lithography, and then DUV is deep ultraviolet lithography. So we're not extreme, we're just deep. And EUV gets all of the press, and in a lot of ways it should. It is, by a lot of measures, the most advanced technology on the planet. Individual EUV machines sell for hundreds of millions of dollars. However, DUV machines, which are less advanced than EUV, but they still will go for $60, $80 million, are nonetheless extraordinarily powerful pieces of technology. The most advanced DUV machines are certainly capable of manufacturing advanced AI chips.

and we can talk about sort of the different degrees of what it means to be advanced. But, you know, China's current flagship AI chips are Huawei's Ascend chips that they manufacture indigenously. And at the technology nodes that those chips where they use the silicon from those nodes, many DUV systems can support those. And that means that those systems are good enough to facilitate China's indigenous AI chip manufacturing. And then, you know, happy to talk about the specific problem in our export controls. But broadly, you know, DUV as a class is, it's not quite as advanced as EUV, but it is still incredibly powerful. So let's play out the story for the Chinese domestic chipmaking industry if they're able to continue importing the machines that they're currently allowed to on the books. Why is DUV so important for the build out and where would that leave China relative to the rest of the world if you continue to play this movie out, you know, five, 10 years?

The fundamental underlying sort of point is that lithography from an export control perspective is what you would call the strongest point of leverage for the U.S. and its allies. And that ultimately comes down to the fact that it is by far the hardest tool to indigenize. It is the hardest tool for China to make itself. And that's reflected in sort of a lot of different, you know, you can see that in a lot of different aspects of industry. The cost of lithography tools is far higher than the cost of other tools. The R&D expenditure by ASML is far higher than the expenditure by, maybe not far higher, but higher than the expenditure of other toolmakers that, you know, are across many different product lines. And if you sort of dig into the numbers and the state of play, we feel pretty confident that China will not be able to make its own DUV systems that are good enough to make AI chips at a large enough scale to really expand its own chip making until into the 2030s.

And so what that means, what that all adds up to is a really large point of leverage for the U.S. So that's sort of like scene setting point number one. Scene setting point number two, I alluded to this earlier, but to get sort of more specific about it, there's many different generations of DUV systems. They're like the very most advanced ASML systems. And then there's the very earliest ones that were made back in the 2000s. But the version of the ASML systems, and that's the NXT 1980 for the nerds out there, that we allow into China is good enough for, again, the 7-nanometer logic and HBM2E, maybe even a bit of HBM3 that are used in Huawei's chips. Right now, we allow that to go into China, and there's various ways within China that those tools can end up contributing to AI chip production that we can get into if it's of interest. but China has very astutely recognized that we have a big cannonball-sized hole.

It's not really a loophole. We call it a cannonball hole in our export controls, and they've taken advantage of that and are importing a staggering number of systems. We've estimated their entire installed base, and we've estimated their rate of imports at about 90 systems in 2024 and 2025, which is over $10 billion, close to two-thirds of ASML's entire immersion, DUV immersion exports, and really a vast amount. Now, to get to sort of your question, Jordan, about scenarios and how things play out, I think as Akib was saying that right now the U.S. is ahead of China. We are dramatically ahead of China, in fact, in overall chip making and compute capacity. And that is because China is constrained by a number of different things. You know, not just lithography, other kinds of SME, etch and deposition that have been restricted by U.S. controls.

And that's kept them to a very small number of chips. but there is good reason to expect them to over time indigenize various aspects of the SME supply chain. Litho is almost certainly the hardest aspect for them to indigenize. It's going to be the hardest tool for them to develop and the last tool that they develop. So what we have sort of done is sort of estimated their lithography bounded capacity and so that's assuming when they do indigenize these other segments like etch and deposition, what is the overall capacity for chip making that's enabled by their current stockpile of lithography systems? And there are a couple of high-level findings that fall out of that. The first is that their current stockpile, which is enormous because our export controls have not been tight enough, is nonetheless physically incapable of matching U.S. AI chip production through 2035.

That means that even if they were to take all of their systems and divert all of them to their most advanced fabs and spend 100% of those systems making AI chips, you know, not make a single smartphone, even then, just at the basic level of physics and lithography throughputs, even then in that most extreme scenario, they will be well below, you know, something like a tenth of U.S. production that we estimate. And that's, again, a very extreme scenario. However, if they continue to import lithography at the rate that they've done in 2024 and 2025, that picture changes. I'm not saying that they necessarily are guaranteed to catch up or surpass, but the curve sort of bends upward, and you can see we've got a lot of nice plots in our report. But there are scenarios in which the U.S. production sort of tapers off, and their production accelerates where not even their physical limit,

but if they sort of make a concerted effort to prioritize AI production, that you could be looking at near parity, maybe even parity in 10 years. So at the highest level, it is the difference between essentially a guaranteed advantage and at minimum a much smaller advantage and real uncertainty as to, you know, the structure of the overall computational equilibrium. I wanted to get a sense of the probability of how you think about these scenarios, right? Like there's a couple of variables here. One, that they continue to import as much as they do. to the share of how much they dedicate to AI chip production. How should I, like, is this something that I should consider a 1% probability, or is this something that should keep policymakers, you know, awake at night? And then I guess leads to the second question of, you know, this is a gradual thing of China keeps importing these tools,

and then they have the capacity to beat the U.S. When is the, I guess, perfect time to strike, or when is the time to actually pass policy to shut the door but actually know it's the right time? Yeah, maybe two responses. First, I would maybe push back against the implicit premise there of a binary, beat the U.S., not beat the U.S. That's sort of a frame that ultimately I'm not sure it really is pointing at something that's that meaningful. the right frame is ultimately relative chip making advantage. And so that means that it's not like if China makes one extra chip more than the U.S. that everything changes. That's not the way things work, even computationally, but certainly not when you're considering strategy and national power. But the relative advantage, the relative, you know, the ratios at which chips can be made, that does matter.

First of all, I would again emphasize that these are lithography capacity bounds. We think that they are informative because China will indigenize the rest of the supply chain eventually, and it will almost certainly be before litho. But that is one thing to flag, that these are litho capacity bounds. So let's stay on that for a second. Like, if you think that they won't be able to indigenize everything, or they'll do a worse job, or, you know, the yields will only be X good. Like, don't we get, like, a whole discount factor before we even get to our list of things that might make you feel a little bit better about selling these machines? It depends on what time scale you want to feel better. You know, lithography systems can be used for, you know, many years after they are sold. So it is true that, you know, in the next year, in the next two years, We think that they are primarily limited by deposition and etch and some of these other SME where we have, in fact, done a better job of controlling them.

But when you're looking at the, let's say, the mid to long term, like three to 10 years, three to 15 years, I think it is just ultimately not tenable to assume that China is going to fail to make progress in indigenizing etch and deposition. I should say that they are making progress. CSET actually has a really good set of reports in 2025, and then again, they updated it this year. And, you know, they've got – they're making inroads in edge deposition, CMP, a lot of these other segments. One segment where they have 0% market share is lithography, and they continue to not, you know, obtain meaningful market share, although I'm happy to talk about their own indigenization trajectory because it is important. But with that said, in the near term, yes, I think that it is fair to sort of treat these as like this is a useful set of information, but it is less of a projection.

But again, when you're projecting out 10 years, I think you have to take it seriously. Certainly in the US we expect into the 2030s that production will be at least partially potentially limited by EUV There been some good I think podcasts on that topic Anyway on the scenarios on the modeling I think there sort of two factors There is one maybe three factors One is just how much do they import? Do they import sort of at the same pace that they've been doing, less pace, even slightly more of a pace? Like if we take as a premise that AI will become more and more important, which is not an unreasonable premise, I think, in fall 2026, is actually not crazy to assume that they might modestly ramp up their pace, even though in 2024, 2025, it was quite a high pace that was at least partially driven by stockpiling. But when you're getting into the 2030s, expecting a potentially even slightly higher pace is not unreasonable. And so we've got various scenarios along that axis. And then within China, there's

the question of what fraction of these machines are dedicated to AI. And that's sort of dictated both by what fraction of them are sitting in the fabs that are capable of making AI chips. That's places like SMIC and CXMT. And then within those fabs, what fraction are making AI chips versus other advanced chips like you might use in a cell phone? And, you know, one thing that we broke out in our modeling is what we call current trends, which is sort of we assume the current allocation that we sort of estimated in our installed-based estimate, we assume that continues, that roughly the same fraction of their machines as today will go to advanced fabs and then roughly the same fraction are allocated to AI. What does that look like? And then we also consider what's called an AI push. And that's if, you know, if China decides AI, this is a national priority, like truly at the top of every list.

Not that it's in a lot of ways it is already a national priority, but as anyone who's worked in government can tell you, there are different levels of urgency that are that are possible depending on how much leaders are bought in on an issue. And the AI push just assumes that they reach the allocation to AI that we forecast outside of the U.S. ecosystem. And that's the scenario in which if they keep importing systems at or slightly above the rate they've done in recent years and they reach that allocation, you start to see potential overlap or near parity in the U.S. and China's chipmating. So there is some legislation that purports to deal with this very question, Nick. You know, we've we've China Talk listeners will be aware of the Trump administration's reluctance. To go at themselves.

So we've had Congress come into the breach over the past year, spinning up various pieces of legislation, which, like, you know, could be done by the Commerce Department tomorrow. but have not. And, you know, enough senators and Congress people are annoyed about that to have passed bills with pretty strong majorities. So we're going to focus on the Match Act today. Nick, what is the Match Act and how does it relate to UV immersion lithography and, you know, bending the cones of the future in ways that advantage the West? Yeah, so the Match Act is a bipartisan, now bicameral bill that deals with exports of semiconductor manufacturing equipment, including, but not limited to, I will say, deep ultraviolet immersion lithography. And, you know, right now it's in the Senate version of the NDAA.

And so, you know, it'll be discussed in the NDAA conference over the next couple of months, which is always a very hectic and Byzantine process. What does the Match Act do? I will say I would maybe break it out. There are sort of the things it seeks to accomplish and then how it tries to accomplish them. And those are sort of two separate topics. In terms of the things it sort of tries to accomplish, one is, you know, they both fall into the category of enhancing controls specifically on SME, semiconductor manufacturing equipment. Match is all about SME. And there's sort of two different buckets of SME and then ways in which it tries to enhance those controls. One, you might call sort of, you know, the really big bad SME, a bucket that's of the really good stuff. And notably within that bucket is a deep UV immersion lithography.

But there are a couple other categories of tools that are thought to be extremely capable tools that really advance chip making. And what MAC is seeking to do is ensure that the U.S. and its allies place countrywide controls on those tools. And that sort of maybe calls for a little bit of explanation. In export control land, there are different ways to control tools or chips or whatever have you. One is to control to the entire country, and that basically says prevents them from entering China at all, or it could be Russia or whatever other country of concern. The other is to allow it into the country, but to restrict it from going to specific entities. And so you might say this can go to China, but it can't go to SMIC, or it can't go to CXMT, or it can't go to Huawei, or what have you. So for these tools, MATCH is basically seeking to elevate the control status from, you know,

restricted only to specific entities to restricted countrywide to prevent them from entering the country at all. Then there's sort of another bucket of tools, which is actually an extremely large bucket of tools that basically covers all SME. And that's where there is somewhat of a miscalibration between the U.S. controls and controls that are enacted by notable allies, most notably Japan and the Netherlands, where the U.S. controls two specific entities, two SMIC and CXMT, China's most advanced fabs. The U.S. basically controls all items that are subject to its jurisdiction, notably all SME. So there's not a single SME that can go to those fabs if it's coming from the U.S. And that's not true for Japan and the Netherlands. They control some SME to those fabs, but a much, much smaller set. So MATCH is seeking to sort of align those controls.

I should say that the nationwide controls, the big bad bucket of SME, is also in a very meaningful sense. What the act is doing is it is aligning the controls on these allied tools with steps that the U.S. has already been willing to take on tools that are produced by its own SME makers. So that is also in a very meaningful sense an alignment, and hence the name, the MATCH Act. briefly how it does that it sort of it directs the commerce department to engage in negotiations with allies for I think it's about 150 days because all else equal everyone would prefer if these tools are controlled nationally using the authorities of the Dutch governments and of the Japanese government and and all other governments when it is that government's toolmaker that makes it. For example, for ASML, of course, that's the Netherlands. Everyone would prefer that solution. It is better for a number of reasons. However,

many times it is hard to get allies to agree. They have their own interests. They have their own domestic politics, economics to think about. So what MATCH directs BIS, the Commerce Department, to do is, if negotiations do not succeed, to unilaterally impose controls. And there's a number of mechanisms, legal mechanisms, that BIS has at its disposal, and in fact that it is used in the past to control both chips and tools, including some of these very tools, actually, at various points during the long saga of the Biden administration. And so MATCH is, you know, it is directing BIS to take that step to ensure alignment if negotiations can't produce it on their own. So the MATCH Act does two things, right? It would restrict the export of these tools to China or further export, but it also restricts the servicing of these tools, right?

So there's ASML engineers, there's Tokyo Electron engineers currently stationed in China that helps fix these tools whenever they break in a Chinese fab. How do you think about that? Do you think it's necessary to also restrict servicing? How much impact would that have? Or do you think it's just necessary to stop future exports and let the servicing keep rolling, especially since that's so much of these companies' revenue? Yeah, I'm happy to talk about that. I think it is right and it is wise to consider those independently of each other. They do have different implications, both in terms of the core impacts on China's chip making and also on the tool makers. So first, you know, I'm just speaking for myself now. In my view, the fundamental thing that that match does and the thing that I think is most impactful in the long term is the items exports, is cutting off the exports into China.

Simply doing that, even if servicing continues at some of these fabs, massively shifts, you know, the state of play. With that said, you know, leveling up servicing controls, because to be clear, there are already controls on servicing that apply at different fabs. The U.S. actually has very extensive servicing controls. The Dutch and the Japanese have substantially less extensive controls. For servicing, where I would make a distinction, you know, if I were waving a magic wand, is the advanced FAB versus legacy FAB question. So just for the viewers out there, advanced FABs are, you know, those that have advanced logic or memory nodes. That's like your 7-nanometer logic, 16-14 nanometer logic, and below is the technical definition in expert-controlled parlance. And then there's an equivalent one from memory that ends up capturing about 18 nanometer half pitch.

Those fabs are where AI chips are made. You know, legacy nodes in general are not sufficient to make AI chips. And broadly, you know, it has been U.S. policy to not try to destroy China's legacy capacity, at least so far. Now, there's a whole set of arguments for what to do about legacy chip making. I mean, there are some people that would like to curtail it, but that at least in the current set of export controls, they focus on advanced capacity. So for servicing, you know, certainly one thing that I think Congress could consider is differentiating between servicing performed at advanced versus legacy fabs. That also has, you know, it substantially affects the revenue calculus for the tool makers, because if 70 percent of your tools are in legacy fabs or 50 percent or what have you, if you're allowed to service at those, then, of course, that's revenue that that you get back.

So that is one where I think, you know, and no doubt it will be probably written 500 different ways during the conference process. But that's one where I think a lot of thought is merited. And, okay, so if we pull this servicing from advanced labs in particular, I think one of the biggest questions I have had when trying to do the research for the Match Act and the biggest source of unclarity is what actually breaks down, right? So it's my understanding that, you know, there's ASML and Tokyo Electron engineers there, but they use a lot of Chinese contractors or a lot of those engineers themselves are Chinese, and they're not going to leave if you restrict the servicing. Maybe some of the parts that you're replenishing are also made in China. Does anything actually break? Is this something that, you know, hampers Chinese production tomorrow? Or what, you know, what breaks first? What are the timelines for it? How should we sort of forecast this in our heads?

That is a good question. And I will say that that is one where it is quite an opaque topic. in an overall area of research filled with opaque topics. That's a particularly deep, dark rabbit hole. With that said, there are real strategic questions there. If all of the servicing ends up just being shifted over to Chinese contractors, then arguably you may not have done much. now. I think in reality, there likely will be impact. There will be impact. I think that it will probably not emerge immediately, but sort of all of the subcomponents, you know, they have like a timeline on which they start to degrade. And, you know, there are certain, there are different levels of degradation. There are some that can maybe easily be fixed by sort of an on-site engineer. There are some where maybe they need, you know, a couple of spare parts. There's

others where they really need to call back to headquarters and, you know, get some specialized expertise. And for some of the most advanced subsystems, particularly, you know, the light source is one that often comes up, some of the stages, some of the sensor systems. That is one where I think it's almost certain that you will see some level of degradation. I wouldn't go so far as to be confident that production will stop. In fact, I would probably bet against that. But seeing some degradation in yield and throughput, probably over the timescale of, say, starting maybe 6 to 12 months and continuing over time, I think that's not unreasonable to expect. And then on implementation of the MATCH Act 2, we're not known for doing things very quickly this is a very deliberative process

as far as I know the current version of the Match Act would basically have 150 days before it goes into full effect isn't that like that's 150 days for trying to stockpile right is this you know how much would they be able to stockpile actual photography machines and also just those tools and parts that you were talking about such that when they're imported you know they can keep servicing it themselves for a much longer period of time Yep. Stockpiling is always a risk, particularly when we announce our controls, you know, many months in advance, which if anyone was, you know, if you were here for the Biden administration controls, the, you know, the famed HBM stockpiling of 2024, there's a rich history of that phenomenon. So realistically, they will potentially stockpile. You know, that is something that you just sort of it comes with the territory. And as we've seen in 2024 and 2025, they might stockpile a lot.

Again, I really can't emphasize enough the power that one of these tools provides. You know, we estimate in this report that a single one of these 1980 systems can probably produce about over 100,000 Ascend chips per year, even assuming China's yields, which are much worse. That is an extremely large number of chips. And, you know, they will probably import tens of tens of those systems, at least maybe over 100 before before the act would actually go into effect, particularly because you can actually extend it if negotiations continue. And there's nothing that government likes to do more than continue negotiations. It's really in governmental DNA. But nonetheless, in some ways, the way to think about I think what the Match Act is doing is it is zooming out a little bit from the month to month discussion of how many chips will Huawei make. This is an act that gets at the fundamental choke point, the fundamental point of leverage for a decade.

and so in that sense the numbers are bigger and the impact is bigger and so even if they they import 50 or 80 or 100 if exports are cut off after then you will see impacts at least strategically on on the ceiling of what they can do and that stretches out 10 years maybe more So I would prefer that they not stockpile. I would prefer that, you know, the government move at the somewhat faster timescale than it does. But I will say it would not ultimately, it does not fundamentally degrade what Match Act, I think, is trying to do. I agree. These machines are super powerful. They print so many chips, right? I also have family in Pakistan, and they use Huawei phones. So I'm kind of like, you know, just steel manning the argument against you here. What about the collateral damage to the, you know, non-AI chips part of China's advanced semiconductor industry?

How much would it restrict, you know, things like smartphone chips? And, you know, is that a consequence that we want? As far as the collateral damage goes, it is true that advanced nodes are used for AI chips and they are used for smartphone chips, too. And so if you constrain advanced chip making, that can constrain smartphone chips. With that said, for the case of lithography in particular, they have in fact already imported such a large amount. We have in fact allowed them to import such a huge stockpile that realistically they've got enough for all the smartphone demand for the next 10 years for sure. In fact the only thing that probably capable of saturating and exceeding that litho capacity is something like enormous AI chipmaking So you know I not going to say that there no impact whatsoever

but I do think that the situation that we are dealing with or the impact, you know, comes if they are already redirecting vast quantities of their chip making towards AI and maybe away from consumer devices, which is, of course, its own issue and something we may face here, which is its own complex issue. Should America care if Huawei can make, you know, chips on the edge? Yeah, you know, like, it is really true that, like, if you talk to different policy folks, You know, China's national ambitions and its consumer companies, they are intertwined in various ways. Their consumer companies are also legitimately consumer companies. You know, certainly people are buying these smartphones to, you know, to call their parents, call their grandparents. So I think there is really a reasonable range of perspectives to be had here. My own view is that ultimately the technology that has the most strategic relevance for the U.S. is AI.

So that's where I would focus. If I have a limited number of hours in the day, that's where I'd focus. But if you talk to certainly in the national security community, people that are looking at just how China projects power abroad and, you know, enshrined its own technological stack abroad, that does play into it. I guess with that sort of lukewarm answer, I guess that's what I'd say. I focus on AI. That's what I'll say. And just two seconds, just to clarify. When you talked about China, like, you know, they've been importing a bunch. They're stockpiling those machines. They're not literally stockpiling them and, like, keeping them in a warehouse somewhere. They're all in active use. Okay, yeah. Yes, that is right. I should be maybe a little bit more precise. You know, in export control parlance, oftentimes stockpiling literally means they're sitting in a warehouse and they're just waiting for the control to kick in and then they'll disperse them. In this case, these tools are, we believe, in fabs.

Now, that doesn't mean that that capability is not there and available to be redirected potentially towards AI uses. So to talk about China's stockpile is still very much a relevant concept. But yes, so far as we're aware, not a huge number sitting in warehouses gathering dust. Definitely not gathering dust because these systems are the cleanest things on planet Earth. So you think it's a big deal. Sort of unclear to what extent the Chinese government thinks it's a big deal. I mean, the strongest argument I've heard against this is that passing this act brings back summer of 2025, where you have the rare earths card and then the factories get shut down and then the Trump administration figures out some way to cave. And that will be awkward, right, because this will have been law.

But, you know, TikTok was banned and we figured that one out somehow. So, you know, when you're sitting here today, like, what do you think is the I'll give my answer afterwards. What do you think is the right way to kind of like conceptualize what the what Beijing's response would be to this legislation getting passed? Yeah, it's a good question. I will say that although, of course, the TikTok law saga did happen, I do think that the potential for retaliation is meaningfully lower with an act of Congress. The thing about executive actions like export controls is that they can be done and undone with just the president or one of his appointees just has to sign a paper or probably not send an email in this day and age. With Congress, I do think the game theory does change a bit.

you know, at the end of the day, it will still be law, even if even if there are ways to get the administration to delay or, you know, to tie it up. You know, there's, of course, there's always the chance that a new administration comes in and decides they have to enforce the law. And and and I do think that it is substantially less likely to get tied up in the trade balance. Another thing I will say is that at least for lithography, and this is where, you know, some of the servicing controls are relevant. Like I said, this act is really, it is a mid to long term kind of act, a lot of the key impacts. It is something where like when you zoom out and look at the overall picture, you can really see a substantial impact. I think it is less likely, certainly the imports bans. It's not like this is cratering China's chip production overnight.

With servicing, there's more to consider, but certainly the items bans. It is not the kind of thing where China will feel instantaneously. And, you know, governments, there's nothing like something that's instantaneous to cause action that's going to bite you next week. I don't think that China's going to feel a bite from the import bans in the next week or the next month. or in many cases, even the next year, in something that becomes stronger and stronger over time. And so because of that, I also think there is less potential for retaliation. I'm not going to say there's none. I don't know the mind of Xi Jinping. There's a whole industry of people that are sort of dedicated to constantly reading the tea leaves there. And I'm not going to say there's none. But I do think it is a different situation in a very meaningful way than some of the sort of crossfire that the administration ended up finding itself caught in in 2025. My take is like, yes, anything's possible.

But the times that you've seen the big escalations over the course of 2025, it's exactly the point that you made, Nick. These were sharp, acute things that had very clear, focusing impacts, right? So 200%, 100%, 150%, 200% tariffs, that will focus the mind of an economy which is focused on exports. The 50% rule, also taking 10,000 plus Chinese companies off the global market for buying various types of technology, a thing that focuses the mind. Like, you know, 150 day window and a promise to hopefully get, you know, the allies on board at some point with an administration which may or may not be super excited about enforcing this. I mean, it's just like it's going to be this long, slow burn. And, yeah, if you're making a play, which is a 15-year one, where this kind of comes in drips and drabs and you're not getting cut off from the current tools that you have in your ecosystem and you have the political economy, right, of the firms saying, hey, don't worry.

like you know this is just a good phase on phase off like we'll get there you know we can do this we can do this whole duv thing um by manufacture the tools i mean then you know there are a lot of kind of dampeners there are lots of like different mutes that um uh come into play and plus there's just going to be so much other noise happening right like this will pass right as she and trump our meeting and it's like if you don't have that moment uh you know we're gonna have a change in control in congress in all likelihood um i can see this like i would put a pretty high probability on this not being a thing that triggers world worth um you know trade war 3.0 because honestly it's like pretty embarrassing right if you are the chinese government and you're saying that like this one thing which maybe can be controlled in three years um where you've had all these you know

you've had she making speeches for years and years and years saying we need to indigenize we need to indigenize like i i think the story that the chinese government ended up telling themselves when it came to importing the uh nvidia chips is likely to repeat itself in this context of being like all right you guys don't want to sell them at some point in the indefinite future fine like we'll figure it out you know we got this whole crack team of engineers we're stealing asml blueprints like we'll get there like we'll see you on the other side um as opposed to uh you know really throwing the gauntlet down and you know forcing trump to cave on something that you know is kind of hard to when it's long. I think that's right. Maybe one comment to sort of go off of what you were saying, Jordan. I do think that hanging over this entire issue,

the DUV issue, is a fundamental question as to when China will indigenize its own DUV. Because in some sense, I was just saying that, you know, we think they'll indigenize edge deposition and metrology and CMP and all these other segments before they indigenize lithography. But it really is true that at the point that they can make tons of litho systems, a lot of the strategic dynamics change, and there's just a much less ability to direct things. There's a couple points that I might make about indigenization. One is that it is hard to forecast because it is a, you know, it is the serial scientific technological industrial process where you figure out one thing and then you have to figure out another thing and then combine them. And it is always hard to make that forecast. With that said, although, you know, I think China will certainly, you know, probably feel that they can indiginize fast and certainly their domestic companies will say that.

We do feel quite confident, and I think I phrased it earlier as, that they will not indigenize litho at a scale sufficient to substantially expand its chipmaking until into the 2030s. And just to briefly comment on that, there's a couple of different segments there. One is, at what point do they sort of master the technology enough to get a prototype? And then, stacked beyond that, is at what point do they convert that prototype and qualify it for commercial production? And then beyond that, there is how quickly do they scale up their commercial production to the point where they are making high tens of systems, 100 systems per year. As I said, right now, you know, we find that they've got 340 ASML litho systems already at their advanced fabs, which is a shame. Actually, sorry, not at their advanced fabs, in the country, split between advanced and legacy fabs. And it is a shame that we have sort of let that many flow into the country. And I think in retrospect, a strategic loss.

But what that means is that to really move the needle, China's indigenous production is not five systems, 10 systems, 20 systems. Ultimately, don't move the needle that much. And, you know, we've in our modeling, we include indigenous production, including, you know, as early as 2030. But I will say that I've also run the numbers, even assuming what you might call a hyper aggressive indigenization, where China announces a prototype for seven nanometer, like an equivalent of NXT 1980 next year, which would be far faster than ASML did in the past. And then the year after that, they achieve commercial production. And then they scale it up 100% year over year, which would be faster than at any point in the ASML, you know, U.S.-allied litho ecosystem. And even then, in that most aggressive scenario, China's existing stockpile and their indigenous production still does not come close to U.S.-allied production.

Really, the fundamental conclusion remains, even under the most aggressive indigenization scenarios that aren't even in the report because of how aggressive they are. If you cut off the imports, you are closing the door on anything approaching parity between the U.S. and China. It is a button that lets you win the AI competition, at least as measured by chip making. If you keep letting imports flow in, And that guarantee is just not there. It's a very different strategic situation. So I think there's three basic questions you have to answer to determine whether or not you support the Match Act, right? One is the one that you said is how quickly do they indigenize? I think the other is, we were talking about this earlier, like what is your bar for feeling comfortable? I agree with what you said earlier that it's not like, oh, if China's one ship, you know, ahead of the U.S., then we're losing. and if we're one chip ahead, then we're winning. But, you know, if we're outproducing 30 to 1, I feel pretty comfortable.

But if I am outproducing 2 to 1, do I still feel comfortable? Like, where do you draw that far? I'm also just curious as your thoughts, because I feel like I have to answer these questions all the time as a researcher. And then lastly, I guess it's AGI timelines. I think, you know, we were talking about this earlier, about it doesn't hit right away, right? But if you think the singularity is happening tomorrow, then maybe you want something that hits right away. So I guess, like, how do your, like, these sort of timelines factor into it, too? Yeah. So first, in terms of, you know, U.S.-China advantage, really it is, you know, especially in the long term. If you are, you know, looking at, you know, tens of millions, hundreds of millions of chips. There's a lot of ways in which this can intersect with strategic issues. You know, there is sort of the direct national security question of, you know, for uses in a national security context.

You know, which could be cyber, could be direct military. There is also the question of sort of the balance of power around the world. and the norms and laws that govern advanced computing. And especially insofar as the U.S. and potentially China in the future may get to the point where they are exporting chips or hosting data centers abroad, having the productive capacity to determine something about how those chips are used, That is a what you might call a diffuse, but extremely significant, in my view, factor. In a political economy sense, I do also think that there is a there is a domestic question. And this maybe gets into your third point, which is. You know, if we accept the premise that I will induce large changes in society, which is, again, I think not so controversial these days.

I do think there is a real, you know, the tech industry, of course, we'll talk about China a lot and, you know, and wanting to beat China. But there is a question as to to beat China to what and what is the outcome that we seek to achieve as a society and, you know, as a governing entity, if you're talking about the government. and having the time to engage with that issue, both in the military and national security context, but also because it is a dual-use technology domestically, without having a rival, an adversary, another great power sort of breathing down your neck. I think there's real benefits to that. How do you determine, like, winning, or how do you determine your threshold for feeling comfortable? Like, there's no, it doesn't feel like, it's all vibes. There's no specific number, it feels, right? And it also depends on so many assumptions about how they allocate their compute.

Yeah, truly that is one where it is, there is no magic number. I will say that there are different ways to think about, you know, two frames for thinking about compute capability is advantaged. versus thresholds. And by advantage, I mean relative advantage. Like, you know, if China has one chip, how many chips does the U.S. have? Versus, you know, then there's a question of just how many chips do they have and what absolute capability do those provide? And I think realistically, you know, you have to engage with both frames. It's hard for me to sort of point out a, you know, a specific number where you say, Okay, at that one, if it's $49,999,999, we're good. We're going to call it a day. That's a tough one. On AGI timelines, the pace of progress, it is true that the bans on imports, like I said,

the impacts are likely to be realized over a three to ten year time frame. I will say that servicing impacts probably will be realized at least to an extent sooner. You know, over the next one year, the next three years, I think it's fair to expect those to emerge sooner. AI progress, you know, has obviously been very fast. You know, there's lots of people that expect even faster progress in the future. I definitely think that insofar as that happens the strategic situation would not be impacted substantially by at least the import span But realistically in those situations there are lots of other crazy things are happening And there are no doubt many other options that are on the table. You know, I will say I myself, although I do continue to expect

loss of AI progress in terms of impacts on the world, I've got a, you know, fairly wide distribution of what I think is possible. I think that in a lot of ways, impacts on the world are maybe tougher to predict and tougher to estimate at specific points in time than, in some ways, than the emergence of specific capabilities even. And so I think, you know, as somebody who, you know, I don't want to just think about the next year or the next two years. I think that Certainly as a governance community, certainly the U.S. government should still, even in this day and age, consider thinking sort of long-term. And I think from that perspective, as I said, I think the impacts of Match are pretty undeniable. For the real heads, still an hour into our conversation here. Akib, what's your initial read on the Match Act applied to the kind of Japanese depot side of this whole equation?

Yeah, I think the Match Act maybe makes a lot. I mean, so in some ways it makes more sense for lithography and in some ways it makes a lot more sense for deposition and etching, right? We don't have, America doesn't have any lithography companies. so in some ways it feels really unfair at least to the Netherlands to impose this on them but we have deposition and etching companies and we've handcuffed our own with the expectation that Japan was going to follow suit and they haven't so there's a fairness I don't know, Ackham, fair's got nothing to do with it like, let's talk to the American tax payer and the other thing I agree with Nick, right, that deposition and etching, China seems much more able to indigenize so it seems like the impacts are unclear like it's not going to be as much of a I don't know choking of the Chinese trip industry as revoking lithography would be. I feel a similar level of opaqueness when we think about what happens to servicing

and parts. Do you have any do you have any reads on this Nick? Maybe one one thing I will say how do these lithography controls compare to controls that already exist on deposition and etch. And that is quite relevant, I think, to the political economy of match. You know, match is the match act, right? Certainly the branding and in many ways the impact is aligning allied controls with U.S. controls. Now, one example of this is, you know, for these entity controls. As I mentioned, the U.S. controls everything and the allies just control a small set. So it levels that up. But even for the countrywide controls, for the litho, I think in a very meaningful sense, it is an alignment action. Now, it is true that, you know, for litho, it's made by ASML in the Netherlands with a small sliver by Nikon in Japan, but it's really all ASML. But we hit on this a little bit. The U.S. has demonstrated many times, including in October 2022, the initial set of controls.

Along with the chip controls, the U.S. hit, I think, 11 deposition tools. At the time, the U.S. was the only toolmaker that would make. I think they were mostly AMAT. And the U.S. did that unilaterally. It was not asking the Allies to do that at all. And that was sort of a demonstration of the commitment to fully control countrywide tools that benefit advanced production. and you know if you look at the whole scope of the advanced controls even outside of lithography for deposition, etch, metrology there are many many US tools with the same profile as the NXT 1980 which is this ASML litho system that are controlled countrywide and the Dutch have been more resistant to doing the same I think as reported for lithography and so what these countrywide controls is they are bringing up the controls on

let's say the most advanced, the most high leverage choke point to the level that the US has already done many many times across deposition, etch, these other segments at the cost to its own tool makers so in that sense it is still fundamentally a matching action an alignment action Do you, I was like, you know, maybe the Netherlands, maybe our allies, you know, follow suit of their own accord or things might get ugly if we FDPR it, right? My initial intuition is that maybe Japan as an ally, this would be a much easier pill to swallow just because there's a lot of competing company interests within Japan. like the Match Act is probably good for Keoxia or it's probably good for you know Rapidus if they're doing anything these maybe have like some countervailing balances whereas it seems just really bad for the Netherlands and their economy

or in a way that they maybe don't want to do it yeah I will say if you get into the political economy of Match one thing that is I think very important to note about Match which is not necessarily true for the bills that you know chip exports is that it is in general fantastic for U.S. and allied chip makers because, you know, fundamentally these litho systems will be used by Huawei, by CXMT, which is China's memory manufacturer to enable advanced chip making, which will then potentially flood out of China and, you know, and compete in the global marketplace and eat into the margins of U.S. and allied chip makers. So from a political economy perspective, anywhere in the U.S. salad ecosystem, including Japan, but also, you know, Samsung and SK Hynix in Korea, and the U.S. chip makers, you know, NVIDIA, AMD, Intel, they stand to gain. Now, in terms of Japan itself, it is, you know,

I won't claim sort of no Japan's internal position on match. I'm sure it is on their radar. but ASML is the company that is selling the vast majority of the immersion tools and Japan, you know, Maicon has a much, very small segment of the immersion economy. Now, Match does potentially, depending on where it winds up in, you know, conference, might end up controlling other tools where Japan has more exposure, but at least for Litho, for DUV, Japan is far less exposed And it does have, it's got its own memory manufacturer. So this is not certain that Japan would be more positive on it than the Netherlands. And I will say, we noted this in our report, that actually if Japan, even if the Netherlands cannot be convinced, In fact, if Japan could be convinced on litho, on DUV, the U.S. and Japan can jointly have the effect of controlling all immersion systems to China without using the FDPR or any extraterritorial authority.

The reason is that the U.S. and Japan both have a national company that makes light sources, which are this vital, vital subsystem. And you have to use either the U.S. one, which is Symer, or the Japanese one, which is Gigafoton. And so if you write a national control that says this light source cannot be exported, if it is going into a DUV, that's going to China. Sorry, a DUV immersion. Since I should be clear, match does not hit all DUV. It only hits immersion DUV. But because of that, the U.S. and Japan sort of have between themselves the power without resorting to extraterritorial authorities to control all of DUV immersion to China. But ASML, they own Symer, right? So it would just be saying, like, your own company can't sell to you in some sense? Well, I would maybe distinguish selling from exports. Yeah. You know, yeah.

You know, they are welcome to sell to ASML for any end use other than, in this theory, this notional scenario for any end use other than, you know, shipping to China. You know, export authority is a different thing than selling, you know, restricting sale of systems. I agree that in some sense, you know, from ASML's perspective, it's like, man, why does our subsidiary have to be in the U.S.? But to be fair, Simon was in the U.S. well before it was an ASML subsidiary. So we had him first. Fair ain't got nothing to do with it, Netherlands When people ask me about what matters for compute policy I'm usually of the opinion that like We're 90% of the way there Like the thing that I think really has the most juice Is remote access, right? And there's just so many other things of like You know, distillation, RL environments All these other things How do you sort of rank it on your totem pole of like

DUV is really important in terms of Strategic lead over China or effective compute Yeah. You know, I do think that it does, it depends on the timescale at which you are looking at things. I do think on anything approaching a long-term timescale, and even a midterm timescale, I would put DUV as my golden baby. You know, again, and it is not just because, you know, I expected to have such an impact on chip making. It is because I do think that the leverage is a vital point that ultimately remote access, although China may be accessing many chips and no doubt is remotely, that is a different situation and there is much more levers available to shape that and how that plays out than it is when China has their own chips. Truly, if you are seeking any form of strategic advantage or even equilibrium,

And the foundation of that advantage is indigenous chip making. And the foundation of that is lithography. So for me, because, you know, I want to be thinking at least in the midterm. You know, even if it's not 20 years out, at least, you know, five or 10 years. Litho is number one. I just want to come back to how good this paper is. You know, what you did here, you know, extremely technical about the actual machines themselves. We have this whole legal angle of it of, like, the different regulatory regimes and how they play out in different countries and the sort of regulatory questions that fall out from how you do it this way versus that way. We have this sort of business side of, like, you know, what are the sort of commercial incentives for all these players? We have this, like, industrial competitive analysis of, like, how good are the Chinese ones versus the U.S. ones. This is an enormous, these are an enormous amount of sort of like domain expertise and skill sets that you've been able to sort of internalize and develop over time.

And like generally when you have someone who does something that's like a policy paper, you hope that you get like one or one and a half of the angles to the question covered in a way that teaches me something new. And like, you know, as someone who has been staring at this, albeit from the outside, but like incredibly closely for, my God, like six years now. Getting to learn so many more things in a higher level of fidelity about this topic, I just thought was a real treat. So, I don't know, Nick, what do you want to tell the kids out there who dream of one day writing a report as wonderful as our DUV immersion lithography, the foundation of USAI compute advantage? How can the kids follow your path? Well, thank you, Jordan. That is very nice of you to say. Anyway, maybe this might be a good opportunity to plug CTS, arguably, which is the Center for Technology in Statecraft, which is this new think tank that I started with a couple of colleagues, and we're six now and may continue to grow.

So, you know, I do think that in policy, as in any form of analysis, there is often a very large gap between getting 80% of the way there and getting all the way there in terms of completeness of analysis and fidelity of analysis. Because at the end of the day, if people are going to make decisions based on analysis, they really want to see everything. And our aspiration at CDS is to sort of really, for everything we tackle, really give it the full once over. You know, we are not sort of intending to be putting out tons of op-eds every week necessarily. It is sort of a place for attempting to develop decisive analyses. because I do think that, you know, our government and our society need a lot more clarity on a lot of issues related to not just semiconductors, but AI and emerging tech, and we have a lot of interest and aspirations.

So I have a technical background, so that helps, and otherwise just grinding away, I guess. But you didn't have a technical background in immersion lithography tools, right? That is true. That is true. Yeah, it does help, I think. Maybe not, I don't know. But oftentimes, you know, if you're technical in one area, you can sort of grok some numbers and then you can use those and sort of catch on to other numbers and sort of then you're off to the races. I will say that a lot of this did take a lot of grinding away because, you know, one thing about software, which I was a software engineer in a previous life, it's all visible. It's all inspectable. You can sort of inspect a software program, whereas ultimately this is a question about manufacturing processes, which are these sort of hidden things. They're happening in fabs, and they're legible to the company that's doing them, but maybe not even to a competitor. Certainly, one of the things we have in here is a full yield model, and that took some doing because yield is extremely closely guarded.

If you could, like, hit anyone with truth serum and steal them for an hour, like, who would it be next? well this is no form of a threat certainly but some ASML leadership that could be helpful I do think that certainly as far as lithography goes they are the masters there is no comparison and ultimately that's just to bring it back to the theme that is why there is so much leverage available because almost nobody can match ASML. Alright, so Akeb, you were wavering at one point. Has the prodigal son returned to the match act? What's your conclusion here? I need more time to think about it. I'm not sure yet. The clock's ticking, man! It's October 14th! Gun to my head!

Let's rock and roll, baby. Let's pass the match act. Okay. All right. I think we'll end it there. Nick, thank you so much for your report. Thank you so much for being part of China Talk. Thanks for having me on. The Match Act. All up in your fab. Hoy, it's the Dutch. Aligned in spirit. True. It's our biggest export. What else can we do? Biden and Gina told us it was fine. We only sell a little to them friends of mine. Sorry, boys. New sheriff in town. This Congress is gonna shut it all down. More free rides, no more. We'll see you don't. Circle back, you answer to me. Make just one more scanner out that door. Try it out. Hold the plug on your whole damn door. I'm a man every skin and every groove.

If I can't, then neither can do. Every service call, every crew Match it or I'll make you Every scanner, every crew If I can, then neither can you Every service call, every crew Match it or I'll make you 5% tornado, babe, I'm so proud Now 0% to mini-mix, say it out loud Nikon, say cheese, now say no Tokyo Electron, put the etcher down slow B.I.M. John, small guys in a fast machine Still the scariest office I've ever seen Did you like the screw bubbles at our reception? Too sweet, sorry, no export exception 36 to 8 in markup we played Now put us in a big beautiful NDA My, my, my, my, my Can I test the groove? If I can't then neither can you

I'm amazed at the call, let's recruit Match me or I'll make you Can we keep having productive conversations? Just ask us nicely, take it slow Three years of that, no text, no citation Repeat after me, N-O I'm amazed at the call, let's recruit If I can't, then neither can you I'm amazed at the call, let's recruit I'm amazed at the call, let's recruit Oh, policy. denial. Oh, oh,