Semiconductors, Tariffs, & Chip Restrictions Under Trump | Chris Miller

In Episode 388 https://hiddenforces.io/podcasts/semiconductors-tariffs-amp-chip-restrictions-under-trump-chris-miller of Hidden Forces, Demetri Kofinas speaks with semiconductor expert Chris Miller about the state and future of the semiconductor industry in the context of Trump's 2024 U. S. electoral victory, with all its implications for industrial policy, defense spending, tariffs, export controls, trade wars, and much more. This conversation is a deep-dive into the semiconductor supply chain. Kofinas and Miller discuss how money is currently being allocated under the CHIPS and Science Act, the progress being made in domestic U. S. semiconductor fabrication, the challenges facing TSMC's Arizona facility, and the troubles at Intel. They also take a hard look at the outcomes resulting from the U. S. semiconductor export and end-use controls implemented against China, how effective or ineffective they've been, and ways to improve them. They also assess China's own efforts at building out its domestic chip ecosystem and some of the loopholes that Chinese companies are systematically exploiting in order to become fully independent of and eventually overtake Western chipmakers and equipment manufacturers. They discuss the three largest limiting factors to scaling AI data centers, the national security threat posed by Chinese companies leapfrogging their U. S. competitors in the race toward AGI, and much more. You can subscribe to our premium content and access our premium feed, episode transcripts, and Intelligence Reports at HiddenForces.io/subscribe https://hiddenforces.io/subscribe. If you want to join in on the conversation and become a member of the Hidden Forces Genius community, which includes Q&A calls with guests, access to special research and analysis, in-person events, and dinners, you can also do that on our subscriber page at HiddenForces.io/subscribe https://hiddenforces.io/subscribe. If you enjoyed listening to today's episode of Hidden Forces, you can help support the show by doing the following: Subscribe on Apple Podcasts https://podcasts.apple.com/us/podcast/hidden-forces/id1205359334 | YouTube https://www.youtube.com/channel/UC8URhgYos5fjHqFSO4RSIEg | Spotify https://open.spotify.com/show/0TK6u251zGSlMgJKH6cOaP | Stitcher https://www.stitcher.com/podcast/demetri-kofinas/hidden-forces | SoundCloud https://soundcloud.com/hiddenforces | CastBox https://castbox.fm/channel/2629539 | RSS Feed http://hiddenforces.libsyn.com/rss Write us a review on Apple Podcasts https://podcasts.apple.com/us/podcast/hidden-forces/id1205359334 & Spotify https://open.spotify.com/show/0TK6u251zGSlMgJKH6cOaP Subscribe to our mailing list at https://hiddenforces.io/newsletter/ https://hiddenforces.io/ Producer & Host: Demetri Kofinas http://www.demetrikofinas.com/ Editor & Engineer: Stylianos Nicolaou Subscribe and Support the Podcast at https://hiddenforces.io https://hiddenforces.io/ Join the conversation on Facebook https://www.facebook.com/hiddenforcespod/, Instagram https://www.instagram.com/hiddenforcespod/, and Twitter https://twitter.com/hiddenforcespod at @hiddenforcespod Follow Demetri on Twitter at @Kofinas https://twitter.com/kofinas Episode Recorded on 11/06/2024

Speaker names marked ? are inferred from speaking time and the show’s known hosts — they are not voice-verified.

• 0:00Demetri Kofinas?

  What's up everybody? My name is Demetri Kofinas and you're listening to Hidden Forces, a podcast that inspires investors, entrepreneurs, and everyday citizens to challenge consensus narratives and learn how to think critically about the systems of power shaping our world. My guest in this episode is Chris Miller, an associate professor of international history at the Graduate School of Global Affairs at Tufts University and the author of Chip War, a book that chronicles the history of a decades-long battle to control the world's most critical resource, the microchip, and the commercial industry that supports it. This podcast was recorded the day after Trump's victory in the 2024 US elections, which means

• 0:44Demetri Kofinas?

  Chris and I had the opportunity to discuss the state and future of the semiconductor industry in the context of that win, with all its implications for industrial policy, defense spending, tariffs, export controls, trade wars, and much more. We dive deep into the semiconductor supply chain between the first and second hours of this conversation. We discuss how money is currently being allocated under the Chips and Science Act, the progress being made in domestic US semiconductor fabrication, the challenges facing TSMC's Arizona facility, and the troubles at Intel. We also take a hard look at the outcomes resulting from the US semiconductor export

• 1:24Demetri Kofinas?

  and end-use controls implemented against China, how effective or ineffective they've been, and ways to improve them. We also assess China's own efforts at building out its domestic chip ecosystem and some of the loopholes that Chinese companies are systematically exploiting in order to become fully independent of and eventually overtake Western chipmakers and equipment manufacturers. We discuss the three largest limiting factors to scaling AI data centers, the national security threat posed by Chinese companies leapfrogging their US competitors in the race toward AGI, and much more. If you want access to all of our conversation and you're not already subscribed to Hidden Forces, you can join our premium feed and listen to the second hour of today's episode on your mobile device using your favorite

• 2:10Demetri Kofinas?

  Podcast app, just like you're listening to this episode right now by going to hiddenforces.io/subscribe. If you want to join in on the conversation and become a member of the Hidden Forces Genius Community, which includes Q and A calls with guests, access to special research and analysis, in-person events and dinners, you can also do that on our subscriber page. And if you still have questions, feel free to send an email to [email protected] and I or someone from our team will get right back to you. And with that, please enjoy this timely and deeply educational conversation with my guest, Chris Miller.

• 2:53Demetri Kofinas?

  Chris Miller, welcome back to Hidden Forces.

• 2:58Speaker 01

  It's great to be back.

• 2:59Demetri Kofinas?

  It's great having you back, man. How you doing?

• 3:01Speaker 01

  Doing good. It's been, what we were just saying, it's been a while since we recorded our first episode, I think two years ago, probably this month. So, nice to be back on the show.

• 3:09Demetri Kofinas?

  Right. I actually think it was, believe it or not, I'm pretty good with numbers and for some reason, I think it was September 6th when we recorded the episode. I'm pretty confident it was in September. And you came on our Genius Community for a Q and A at some point a year later or so to give us an update. It might have been around some news, or it might have just been a general update on how the export controls and end-use controls on semiconductors are going, or were going at the time. We're going to have a chance to talk about that today, as well as the CHIPS Act, how money is being spent and allocated there, how it's going overall, TSMC's Arizona semiconductor fabrication facility. Intel has also been in the news. It's dealing with a lot of problems. It's an integrated device manufacturer, an important

• 3:52Demetri Kofinas?

  one, an important part of the US industrial strategy to reshore the semiconductor supply chain. So that's a big part of this conversation. We're going to talk about China's domestic chip ecosystem, how the export controls are going, the AI build out, but- Coincidentally, in fact, I hadn't actually planned this. We're recording this the day after the election and we have a definitive result. Donald Trump has won a decisive victory. He's also won a majority of the popular vote. And also, I believe the Republicans will have control of both houses. So it's going to be a real opportunity for the vision of a Trump team and a Republican White House to really manifest through industrial policy and the economy. Trump has talked about tariffs. I'm curious, when you look at, and given the conversations you've had with people

• 4:36Demetri Kofinas?

  that are affiliated with the Trump administration, when looking also at the outside, studying what the proposals have been made, how do you think this impacts the policies around reshoring semiconductor usage, the chip restrictions, the trade wars, the China, all of that? What is the impact of a Trump administration in 2025?

• 4:56Speaker 01

  You know, I think if you step back and look at the last three administrations, from late Obama through the first Trump term and through Biden, there's actually a surprising amount of continuity on all of these issues in terms of subsidizing domestic manufacturing. The CHIPS Act was first passed under Trump, implemented under Biden. If you're looking at restrictions on China, those were really ramped up under Trump and then continued and provided even more structure under Biden. Tariffs ramped up under Trump, continued and actually expanded under Biden. And so I think just looking at that trajectory, it's safe to assume that all three of those sets of policies continue. Now, what the level that Trump pursues them at, I think is the question.

• 5:40Speaker 01

  He's obviously articulated tariffs as a priority. Other semiconductor issues haven't really been part of his focus in the campaign.

• 5:47Demetri Kofinas?

  All right, we're going to have a chance to dig into that. Just out of curiosity, since, again, we haven't spoken too often, and although we did have one chance, but it wasn't for the podcast, what's your life been like since publishing chip wars, given just everything that's happened in the industry?

• 6:02Speaker 01

  Well, it's been a fun time to focus on semiconductors. Obviously, the US-China tensions are a big part of that. AI is another big part of that, which I think I expected some degree of that when I published Chip War, but just the surge of investment and therefore the interest in AI and data centers and Nvidia and ChatGPT and its semiconductor implications has been a surprise level. And I think the third thing has been electronics and computing in general, the supply chain shifts that are underway due to both those dynamics are also really interesting. And so I spent a lot more time than I would have guessed in the last two years, not just in Taiwan and Korea and Japan, but also in India, for example, in the Middle East, both of which want to become bigger players in this industry.

• 6:45Demetri Kofinas?

  Yeah, because you published the book in 2022 and ChatGPT 3 was released when in December of 2022?

• 6:52Speaker 01

  November.

• 6:52Demetri Kofinas?

  November of 2022. So that came right afterwards. What are you seeing in terms of interest from people in the industry, from affiliate industries who are trying to, who are reaching out to you in order to get a better grasp on the direction of where things are going?

• 7:08Speaker 01

  You know, I think in the chip industry, there's still a ton of focus on US-China dynamics. And the executives who currently run chip firms, they all came of age in a period in which engagement with China, trade with China, investment in China, that was the way you build businesses. And they've all really struggled as a group to acclimate to this new environment where competition with China and restrictions on China are now the status quo. And so I think the chip industry has had a lot of learning to do on that front. I think in other segments of the economy, the learning has been how to make sense of the investment in AI and data centers. And so I'm struck by not just the number of private equity firms that I end up speaking to are interested in data center investment, but the number of energy investors who are

• 7:51Speaker 01

  now trying to deliver power to data centers that hadn't thought about data centers at all, hadn't thought about chips at all, are trying to put Nvidia on a map. They now realize that power demand is increasing because AI is demanding more power. And that's been another really interesting change that I hadn't expected.

• 8:07Demetri Kofinas?

  Yeah, it's interesting because as I've tried to understand the nature of this dynamic of the rush to, well, not just the general business imperative to make advancements in the chip industry, but also the geopolitical imperative, there are Some obvious bottlenecks like computational capacity capability. Also, the number of data centers you can spin up to sort of offset, in the case of China, the hit you take to compute by just aggregating more chips together. But there's also energy. Energy is another bottleneck. So I want to have a chance to talk about all that. Let me ask you one more thing about Trump and the Trump team before we switch to a conversation about the CHIP Act, which is the first real discussion point I want to go through. Who within Trump's orbit will be leading the charge on America's industrial policy?

• 8:52Demetri Kofinas?

  Do we know who those people are besides Robert Lighthizer?

• 8:55Speaker 01

  I think Lighthizer is going to be the really interesting figure to watch. I've heard speculation that he might end up at the Treasury Department or at the Commerce Department, both of which would be very powerful roles. And I think he's been the primary person who has been advising and in the first term implementing Trump's tariffs. And so that will play a really major role in shaping industrial policy. I think beyond that, we are still waiting to hear some of the names that will be actually implementing these types of policies. And I think that's something worth tracking pretty closely over the next couple of months as we not just hear who are going to be the secretaries in that given cabinet departments, but who are the undersecretaries because they're often actually playing a huge role in implementation.

• 9:36Demetri Kofinas?

  So I also want to recommend the episode that you did do. I failed to do that. If people haven't heard that episode that we recorded a few years ago, which was based off of the work you did in your book, CHIP Wars, I recommend they actually pause this recording and go back and listen to that first because that's really foundational. But I'm curious today, when we look at the amount of money invested in a given calendar year across all aspects of the semiconductor manufacturing industry, how much is being invested on the part of Chinese state actors and companies? And how much is being invested across the entire Western chip ecosystem that also includes countries like Japan, Taiwan, and South Korea?

• 10:16Speaker 01

  You know, it depends how you slice the numbers. If you look solely at chip manufacturing, what you find is that investment in China has increased a lot, but that there's still more investment happening outside of China, in Taiwan, in Korea, in the US and in Japan, first and foremost. And if you look at different types of manufacturing, you can get different nuances. So for leading edge or close to leading edge manufacturing, the types of chips that go in phones and PCs and data centers, almost all that's still happening outside of China because China can't buy the types of tools you need, like the EUV lithography machines that you need for cutting-edge chip making. If you look at lower-end chips, the more foundational chips that are going into dishwashers

• 10:57Speaker 01

  and going into cars, that's where China is putting most of its investment dollars. And that's where you're going to see a really dramatic increase in Chinese production capacity over the next couple of years. And so if you look at the best private sector forecasts of the number of facilities under construction in China and the rest of the world over the next couple of years, you find China's almost building as much as the rest of the world combined, not exactly, but almost, which is a really extraordinary change because in the past, China was a pretty small player when it came to producing even lower-end chips. And so that's where the market shifts are by far the biggest.

• 11:31Demetri Kofinas?

  And what is the strategy there? Is the strategy there to dominate the supply chain and take market share from Western companies that are servicing clients in less economically developed and unaligned parts of the world, like countries in Africa and Latin America?

• 11:46Speaker 01

  You know, I think it depends on who you're looking at in China. I think if you're running a chip company in China, your goal is to build a business that's viable. And so that points you in different directions than if you're the government. The government wants independence from Western ships. And if you put yourself in China's shoes, you're still importing huge volumes of even lower-end chips from the West, from Taiwan, from the US, from Europe, from Japan right now. And so I think for policymakers in China, they're still worried in a worst-case scenario, a conflict over Taiwan, they're going to lose access to the chips that their dishwashers require, as well as the chips that their AI systems require. And so that Alone is enough to build out this big manufacturing base for for low end chips.

• 12:31Speaker 01

  So then if you're receiving big subsidies from the government or from the provincial leadership or from your municipal government, well, and you're a factory owner, well, why not keep building capacity because the government's footing the bill? And so I think you've got different rationales for this build out with different actors. But the net effect is that there's a ton more capacity coming online. And it's coming online fundamentally for non economic reasons. It's not that demand is mandating more production of volume in China. It's that the subsidies are making it possible.

• 13:01Demetri Kofinas?

  All right, so from that perspective, what is the calculus that's happening at Western firms? When they're seeing all this money that's being spent in conjunction with the Chinese state by Chinese companies in the legacy space, legacy chip space? Does this have a chilling effect on investment on the part of Western firms in those areas? And then so that they end up focusing on leading edge to the detriment of legacy?

• 13:26Speaker 01

  Yeah, I mean, I think it already has had a chilling effect to some degree. Part of the nuance here is that a lot of Western chip firms aren't impacted because they don't produce these types of chips. So if you're in Nvidia or AMD or Intel or Qualcomm, you're producing primarily leading edge chips. And so the Chinese lagging edge build out doesn't really affect you. And it could actually benefit you if the systems that your chips are going into are getting cheaper Chinese components, at least in the short run, that could actually to some degree be a positive. It's only the Western firms that produce legacy chips. So think of Texas Instruments or On Semi, also European firms and Japanese firms that are facing really head on competition from a lot of these heavily subsidized

• 14:10Speaker 01

  new Chinese chip makers. And there we've seen, for example, major investors in semiconductors publicly call for reduced capital investment in Western facilities. One of the reasons being that China is going to be flooding the market in certain types of commoditized chips. And, you know, that's why in the US and Europe and Japan, there's active discussion about should there be tariffs imposed on the import of Chinese chips into Western manufacturing? Supply chains.

• 14:37Demetri Kofinas?

  Actually, that was gonna be my next question, not specifically whether tariffs would be imposed, but what has been the response by government? So is tariffs the preferred method? Is there also desire to use in part, again, some kind of state subsidies? Or maybe that's not the right word, but some way to offset the costs for domestic manufacturers to make up the shortfalls in legacy production?

• 14:58Speaker 01

  Yeah, you know, I think the CHIPS Act in the US and similar EU CHIPS Act, Japanese subsidies had the effect in part of offsetting Chinese subsidies. But the scale of Chinese subsidies is much greater than the scale of Western subsidies going into the legacy segment of the market. And that's why you've seen Western governments consider two other policy proposals. So one is tariffs. And the Biden administration actually earlier this year doubled the tariff rate from 25% to 50% on chips that are imported directly from China to the US. And that's actually not super impactful because the US imports hardly any Chinese chips directly. They're sent to Vietnam and then packaged in Malaysia and put in a dishwasher in a third country and then imported.

• 15:44Speaker 01

  And so chips that are part of a bigger system and imported don't today face any tariffs because the system faces a tariff rate based on what type of system it is. And so one of the topics of discussion is should you impose a component level tariff? So if your dishwasher is made in Malaysia, but it has a Chinese chip inside, should it pay a tariff based on the Chinese chip content when it crosses into the United States? And there's a lot of support for those types of component based tariffs in both Congress and in certain parts of the federal government bureaucracy. And that would be one way to disincentivize Western manufacturers and Chinese manufacturers from using Chinese ships. So that's option one. Option two is just to say, we don't trust Chinese ships.

• 16:25Speaker 01

  We're going to ban them from critical systems, from defense and aerospace, certainly, from autonomous driving, where we've seen the Biden administration take steps there, from data centers. And so I think we're going to see more on that front, too, of just broad bans of Chinese ships on security grounds.

• 16:41Demetri Kofinas?

  All right, man. So I have so many questions, but just one just popped into my head. So there was recently found that some of TSMC's chips had found themselves inside of Huawei products despite the export controls and restrictions.

• 16:53Demetri Kofinas?

  And so my question is like, how effective is it to do this kind of niche targeting? Or does that inevitably create loopholes that will be exploited? And do we end up, are we eventually going to get to a place where we're going to see blanket bans?

• 17:06Speaker 01

  Yeah, you know, I think we're already moving in the direction of more blanket bans because niche targeting is really hard. I think in a lot of ways, the 2022 export controls were the first step in that direction. Before 2022, the US rules more or less said you can't sell chips to Chinese military companies. And we found that there were lots of examples of that happening. And so the US shifted the rules and said, you can't sell AI chips to China, period. And that is kind of a direction of travel towards blanket bans as it is. Now, I think the next potential step in that direction is rather than say, TSMC can't produce AI chips above a certain threshold and sell them to China, or TSMC can't produce for certain Chinese firms.

• 17:49Speaker 01

  I could imagine a future administration saying, TSMC or any advanced fab can't produce below seven nanometers for any Chinese customer, for example. And that would be a logical next step if you wanted to move further towards blanket bans. Obviously, there are downsides to that because the US has also said, we're only targeting AI chips because of their dual-use applications. And the broader you go, the more you also hit smartphone chips and autonomous driving chips and other types of chips that don't obviously have any sort of dual-use application.

• 18:20Demetri Kofinas?

  So this might seem like an obvious question, but actually, I think when most people ask themselves, they'll find that they lack specificity in answering it to themselves, which is, what is the real risk of China dominating certain segments of the market? I think it's been made very clear what the consequences are of it dominating the leading edge, but what about this lagging edge that we're talking about now?

• 18:41Speaker 01

  So I think if you put that question to US security officials, you'll get the following answer. They'll say, look at the chip shortages in 2021, 2022. It showed that if you were a car manufacturer, for example, you rely on 1,000 chips per car. And if you're missing just one of those chips, your cars could be stranded in parking lots for months, costing you tens of billions of dollars. If China was the sole producer of one of those chips. Wouldn't that give it extraordinary influence over Western firms and therefore Western politics? And it's not hard to imagine that because China's been doing that with rare earths, with gallium and germanium, with other minerals. And so the US, as well as I think other Western leaders, are worried if China becomes dominant

• 19:25Speaker 01

  in certain components that are really critical for the rest of the industrial base, that will give it leverage that in the past China's had no concerns about using. And so it's not so much, I don't think, a concern that this is going to hurt Western firms. That dynamic is present. I think it's really China's going to gain leverage over the West in an economic and a political sense.

• 19:47Demetri Kofinas?

  So if existing trends continue, what does that mean for China's share of the global market for semiconductor production?

• 19:53Speaker 01

  So if you look at the market in dollar terms, the most expensive chips are the high-end chips, and that's where China's going to face the largest issue. So in pure dollar terms, China's share of the market is going to stay, I think, relatively contained despite the extraordinary investment that's underway in China because it's going to focus on the low end and therefore low-cost portion of the market. But if you look at the more commoditized segments of the low end, I think you can already see China's market share increasing pretty dramatically. And that's the concern. Not that China's going to take over the chip market, that it can take over certain segments of the commoditized low end. Western investors won't tolerate Western firms building new capacity.

• 20:35Speaker 01

  They might even encourage them to close capacity that already exists and that China will become a potentially monopolistic player in the low end of the market. And that's something that is eminently technologically feasible in China. It's just a question of how many dollars do you want to deploy? And I think President Xi's track record suggests he's willing to deploy a lot of dollars in this segment.

• 20:54Demetri Kofinas?

  All right. So let's actually, before we get into a conversation about the CHIPS Act. Let's talk about the export controls. I want to talk about both how they're going, China's response to them, as well as this thing that I keep reading everywhere, which is that we can't afford to have China reach AGI before we do. Let's actually start with that question because I'm not entirely clear on whether that analogy is correct. The analogy seems to be drawing a parallel with the nuclear bomb. Right? Like, obviously, in the case of the nuclear bomb, it makes sense. You didn't want the Nazis to obtain the nuclear bomb before the Americans because they could immediately use it and wreak havoc. AGI doesn't exactly work that way. So, one, what do you think about that argument that we can't afford to have China achieve

• 21:36Demetri Kofinas?

  AGI, given the fact that we don't quite know what AGI is, what it looks like, and it's not as binary in its application?

• 21:44Speaker 01

  Yeah. You know, I think I would phrase the argument slightly differently. I would say, first off, is there a strategic interest, in addition to an economic interest, in having US firms lead in AI? AGI is not just one thing, it's a spectrum continuity. And I think, yes, we want US, or at least Western firms, to be leaders there. And so, access to compute is one of the things you need to be a leader. The second thing that's probably more important is that the US wants to make sure it's Western firms dominating the infrastructure that undergirds AI. And I think the analogy here is you don't want to find yourself in a situation like in 5G, where everyone knew 5G was important and there was no compelling Western player

• 22:28Speaker 01

  because Huawei was able to produce equipment that was just as high quality as Ericsson and Nokia at a meaningful discount in price. And so, the US would much prefer the situation like today in AI, where if you want to build an AI data center, there are no non-US options, which means that you either don't build AI infrastructure, or you use US suppliers, and often not just US suppliers for the components, you use Amazon, or Microsoft, or Google Cloud to provide you the cloud computing services. And that creates a lock-in effect that will guarantee not just that the US has this generation of AI infrastructure, but I think that the US is betting that it's going to have the next decades' worth of infrastructure provision because its firm

• 23:09Speaker 01

  will be in the lead. And I think when you think about the ability to influence future regulation of AI, think about future espionage capabilities and locking out Chinese espionage capabilities, when you think about being able to influence which countries get privileged access and which don't, making sure your adversaries face a harder time than your friends do, these are all areas where having a major role in infrastructure provision really matters. And I think that's the primary thing that's driving US policymakers more than a debate about who gets to AGI first, acknowledging, as you said, that nobody really knows what exactly that means. And it's probably not a single destination. It's something that has developed over time.

• 23:51Demetri Kofinas?

  That's a good answer. How closely and efficiently are the big private sector firms working with the US government currently on this issue? And especially if we compare it to the American companies in the telecom industry during the period where Huawei became a dominant player?

• 24:08Speaker 01

  I think if you look at the personnel that the big AI labs, as well as the big cloud computing companies have hired, they've all got a large number of former officials on their staffs that are both working with government and managing government relations. I think there's a shared interest in seeing US firms play a leading role. But I think there's also a lot of back and forth. Companies, they want support, they don't want restrictions. Government wants to be supportive, is good at supporting in some ways, less good in other ways, but also wants restrictions. And so I think a lot of companies, for example, have found themselves in the position of wanting to partner with countries in the Middle East, like the UAE or Saudi. And the US government's been publicly nervous

• 24:53Speaker 01

  about what the shape of those partnerships are. And so that's been an area where the US government has actively restrained or constrained US tech and AI firms from doing what they otherwise like to do.

• 25:04Demetri Kofinas?

  Yeah, this brings us to a conversation that happens quite a bit within the AI safety space, which is that the United States is no longer a unipolar power, working within an international rules-based framework, where the public interest can be, can have more leverage relative to national security concerns. In this environment, there are competitive pressures, because the United States, as a world power, wants to outcompete China. And in the context of that, you could see a situation where the public interest falls to the wayside, as private firms work with the government for advancing not just your political interest, but the private interests of these large tech companies. Is that a concern that you have, when you see how this space is evolving?

• 25:48Speaker 01

  Yeah, I think it is a concern. I think there's, there's always a question of what is the public interest, and which public's interest? And it's in the public interest of everyone in Nigeria to get access to as much AI infrastructure as possible. It might not be in the public interest of the US public, to have Nigeria get as much access as possible. And so I think there are conflicts between different definitions of the public. I think there's also real debate as to what AI safety means, and what it implies for regulation. And so, your definition of the public interest in that sphere will be shaped by your views on the need for AI safety regulation, or the lack thereof. And so I think this is a space where it's easy to say, we should be devising policies in the public interest.

• 26:32Speaker 01

  And then when you begin to ask, which public, and what does that mean, it becomes much less obvious, and much more contested.

• 26:38Demetri Kofinas?

  Yeah, that's true. I was actually meaning the US public's interest, but you can see where that has been a challenge, in the supranational case of the European Union, and their implementation of GDPR, which now gets a lot of blame for Europe falling behind in innovation on the AI front. So it's obviously a very difficult challenge, but there are sort of interests to balance there. So I'm just curious your thoughts on it. Let's switch to export controls, before we get into a conversation about the CHIPS Act. How are they going? How are they working? How are they not working? What do we know about how things are going on that front?

• 27:12Speaker 01

  So I would say it's clear that they're working in the sense that China has much less access to high-end AI chips than it would in the absence of export controls. It's also clear that there's major gaps or loopholes in their enforcement. And I think both can be true simultaneously and both are true simultaneously. First, on are they having an impact? I think there is no doubt that China wishes it had more access to AI compute than it has. I'll just give you a couple of data points. Data point number one, Chinese firms will import hundreds of thousands of the Nvidia H20 chips that are specifically downgraded for the Chinese market. You would only do that. You'd only buy second-rate chips if you can't buy first-rate chips.

• 27:54Speaker 01

  And the fact that this is happening at such scale means that there is a need for AI compute that China can't satisfy domestically. I think the second data point is that I see regular anecdotes, I think generally true, of H100s, for example, being smuggled into China in suitcases. You only smuggle in via suitcases when you don't have access to better supply chains and you still need high volumes of semiconductors. And so I think when you look at all of the evidence of smuggling that is taking place, plus you look at the evidence of very high-volume purchases of specifically downgraded chips, it's clear that China does face a deficit of compute relative to what it would like. Now, the smuggling that I've cited is also evidence that the export controls aren't being

• 28:37Speaker 01

  enforced nearly as effectively as one theoretically could. I think that's partly based on the inherent difficulties in enforcing controls. It's always hard to limit the sale of high-value products because people are smart and creative when there's money to be made in smuggling. And I think it's also evidence of the fact that the US government has not historically had to deal with this type of export control problem. Historically, it's been trying to prevent the sale of missile parts to Iran and North Korea with mixed success, but that's a different enforcement challenge.

• 29:11Demetri Kofinas?

  Because it's not dual-use.

• 29:13Speaker 01

  It's not dual-use, exactly. Exactly. And so I think we don't yet have the capabilities at the level we'd like to actually enforce controls on dual-use products.

• 29:22Demetri Kofinas?

  So my next question is going to be where and how are export controls working better or worse? But before I ask that question, in my mind, when I think about this race towards AGI, towards artificial general intelligence, there are three vectors that seem to be the most important for training these models. One is data, not just quantity, but quality. And there, it seems theoretically speaking, though I don't know if someone's in a position to actually be able to measure this, it seems that China has an advantage, not just because they have more people, I don't know that that's necessarily relevant or as relevant, but their regulations are different. There just isn't as much emphasis on privacy. And so that's also reflected in infrastructure build outs and how much data is being collected.

• 30:07Demetri Kofinas?

  The other one is computing power, and obviously the chip restrictions are meant to diminish that, at least on a chip by chip basis. But then China can also work really hard to subsidize the build out of larger data centers, maybe even more clustered. At the moment, I think there's more clustering happening in the United States. Maybe you can explain why that is. But that also seems to be an important vector, and it's not entirely clear who's in a better position there. And then there's energy, the energy to actually run these facilities. And also, I'm not entirely clear on who has the advantage. Maybe on the short term, it's China, but sort of long term, it's not clear either. Do you generally agree with that? And how do you view the challenges to both countries and both ecosystems in trying to win that race?

• 30:48Speaker 01

  Yeah, I think those are the key potential bottlenecks. I think, start with data. I think there is certain spheres where China does have advantages due to its privacy regulations. Facial recognition is the obvious example here, where I think most people would say China is ahead in facial recognition. But if you look at the LLMs, they're largely, at least thus far, they've largely been trained on text on the internet. And no one has an advantage in text on the internet. It's all on the internet. Now, I think the next generation, and increasingly going forward, it'll be about your ability to generate synthetic data and to use AI to create better data to train on and to filter your data. And that's a sphere where no one has a guaranteed advantage.

• 31:30Speaker 01

  It's all about your ability to innovate and provide better mechanisms for creating quality synthetic data. And so I would guess that the US is going to be the leader in synthetic data, just because the US is where most of the leading AI investment is happening. But that's not guaranteed. That's a hypothesis about the future. But I don't think China's got an inbuilt advantage, except in specific spheres where privacy regulation really matters, like facial recognition.

• 31:57Demetri Kofinas?

  So I'm probably already a bit out of my depth here, but I'm just thinking extemporaneously, when it comes to synthetic data, how important is the prompt? And is that where a lot of the innovation comes from on the prompt side? How do you actually create prompts that encourage AI to generate better data?

• 32:14Speaker 01

  Yeah, you know, synthetic data means a bunch of different things. And so it's not just asking an AI system to create, you know, random strings of data. It's asking AI to help sift through your data, for example, to check your data, to generate certain types of data. And so the engineering around that is actually really important. And that's where talent, I think, is going to matter much more than the raw data that exists in the real environment. One other sphere here I think is worth mentioning is is data about the physical world. And this is when you think of autonomous driving, for example. This is a sphere where it's not clear yet who's got an advantage in terms of data. But that I do worry that if China is going to be much more supportive in terms of its autonomous driving regulation than US regulators,

• 32:58Speaker 01

  you could imagine a scenario where there are just more autonomous cars gathering more data and therefore building better autonomous systems in China. So I think we should take this seriously, even though it's not clear right now that Tesla is not in the lead, for example, in autonomous driving.

• 33:11Demetri Kofinas?

  Is that also true in robotics? So for example, we saw recently with the longshoreman strike, they, one of their demands was putting curbs on autonomy at the docking facilities for fear of being replaced by robots at some stage. That's not the case in China, which have, from what I've been told, the leading edge of automation at these docks. Is that an example where you could also fall behind? And is there a concern within the robotics industry of that actually happening?

• 33:35Speaker 01

  Yeah, I think that is a justifiable concern. I'm not sure I've got a high confidence view as to who's got the better dynamic there. I think if you're in China, the primary deterrent to using robots is that labor costs are lower. Whereas in advanced economies, you've got higher labor costs. And so you need robots a lot more than you do in China. And so that would make me think that advanced economies are going to try harder on robots because we've got no choice.

• 34:01Demetri Kofinas?

  That's a great counter to what I said. I love that, especially given some of the economic challenges they're having. And if we're going to move, you know, beyond the rest of the economy. Although China does have demographic issues, so who knows? Anyway, that's an interesting thing to think about. Maybe we'll go back to it. So another question I wanted to ask you is where and how are sanctions working better in some parts than in others of the supply chain? Where are they working most effectively when it comes to China?

• 34:24Speaker 01

  Do you want to go to the question of compute and energy access first?

• 34:26Demetri Kofinas?

  Absolutely. I forgot to bring, I forgot to talk about that. You're right. So we talked about data. We talked about compute and then we were gonna talk. I think we talked about compute, but maybe if you wanted to flesh it out more. And then we definitely did not talk about energy. So, no, go ahead. Please answer that.

• 34:40Speaker 01

  Well, on compute, I think we know who's got access to the better chips at scale. I think the interesting question is why we haven't seen clusters in China at the scale one might expect. As you alluded to, the Eric Schmidt critique of export controls was that if you've got worse chips, you just build larger clusters. And so far as we can tell, China's not doing that right now. In fact, I think one of the most interesting charts you can make is the chart of capex at big US tech firms since, for example, the release of ChatGPT versus the chart of capex at big Chinese tech firms, Baidu, Tencent, Alibaba. And what you find is that US firms have increased capex because they're building these massive clusters and Chinese firms really haven't.

• 35:21Speaker 01

  There's not a clear trend in capital expenditure and the size of clusters being built does seem to be meaningfully lower in China and the quantity being built is lower. Why is that is a really interesting question. My best guess, my best explanation is that the Chinese private sector doesn't think it can make money off of big clusters like that. The Chinese government has been much more skeptical of AI for consumer applications than the US government has. You've got to make sure your AI doesn't say the wrong thing about Chairman Xi, for example. And anyway, the consumer internet in China has had a really bad five years since the tech crackdown. And so your willingness to invest billions or tens of billions of dollars in a cutting edge cluster, I think is going to be lower in China as a result of that.

• 36:05Speaker 01

  And we haven't seen the state step in to fill that gap to the degree that would be necessary to make up for the private sector's limited investment relative to the US. And so it's not just that China doesn't have high-end chips. It's also that China doesn't appear to be building clusters at the scale of the US. I think for its own internal political economy reasons. And that's, that's also a key dynamic.

• 36:28Demetri Kofinas?

  Is that also a concern from a national security standpoint that large data clusters would be a target? You know, we've all seen that meme circulating on the internet about bombing the data centers.

• 36:38Speaker 01

  You know, I would think if I were devising China's national security strategy, I would be much more worried about not having a large cluster than I would be worried about the large cluster being target, just because it seems like we know that large clusters are pretty critical for building next generation AI systems.

• 36:55Demetri Kofinas?

  Because of economies of scale.

• 36:57Speaker 01

  Economies of scale, that's right, that's right. And China seems to have concluded that that's not how it's going to deploy its capital, which is a very big difference from the US strategy.

• 37:04Demetri Kofinas?

  Do you expect that policy to change?

• 37:06Speaker 01

  I'm surprised that it's now been two years since ChatGPT. It's been two years since all of the big US tech firms have been pushing this direction, and we haven't seen this move in China yet. So I don't know what would make it change now if it hasn't been happening the last two years.

• 37:22Demetri Kofinas?

  So I had Brian Janis on the podcast sometime over the summer, can't quite remember when, maybe it was in the spring, talking about the increased energy demand usage, reformatting the grid, expanding the grid, getting more energy to these data centers. And part of that also included a discussion about Amazon's recent purchase of a nuclear power generating facility to power one of its data centers. I think I saw news that the US government has temporarily suspended that deal or put the gaba on it, whatever term you want to use. What do we know about that? And how big of an impediment are regulations when it comes to sourcing energy in the US for data centers versus China?

• 38:01Speaker 01

  So I think that that is a key issue. I saw the same Amazon headlines. I don't know if they're repealable, those rulings or not. I'm not an expert in that area of regulation. But I also saw Mark Zuckerberg talking about how a species of bees and the Endangered Species Act was slowing one of Facebook's efforts to build, to deploy nuclear for data centers as well. I think that does speak to the regulatory burdens. You know, I think in China, on the one hand, one might assume that because it's an authoritarian government, because they've got a long track record, for example, of bulldozing housing to build new apartment blocks, there's no property rights protection anywhere near the level in the US, that you'd be able to build a lot more power.

• 38:44Speaker 01

  I'm not sure that that's true, though, because that assumes you've got Political backing for building big data centers, which doesn't really appear to be the case, given how it's not a regular topic of Xi Jinping speeches and how they're not building nearly as big a data centers as you might expect right now. So unless the party apparatus gets behind data centers and therefore, energy provision for data centers as a really core priority, I think it's possible that China faces the same regulatory roadblocks as US providers do because there's the lack of political support.

• 39:16Demetri Kofinas?

  What about the build out of nuclear facilities? We've seen a, how many? Like at least 100 nuclear power plants commissioned in China? Maybe you have the numbers you can tell me. So is that an important advantage, even when taking into account the US's natural gas and oil production capabilities? And also, does it suggest that we will inevitably see a change in policy towards nuclear in the United States?

• 39:39Speaker 01

  You know, I think on China's nuclear build out, it's happening to some degree. The extent to which the announcements actually translate into new nuclear plants is, I think, somewhat open to question. Just like in the US, no one actually wants a nuclear plant next door, or at least a new nuclear plant. That's true in the US. That's also true in China. And so I think we should be, have some skepticism that the number of nuclear announcements that we've seen in China all translate into new nuclear plants. China got plenty of coal, too. And so if you don't have nuclear, you can easily power all the data centers you want via coal. There's no intermittency problems. I think it really is the question of, can you get the power plants built next to the data centers with the grid interconnect? That's the real challenge, probably both in China and in the US.

• 40:22Speaker 01

  My guess with the US is that nuclear plays a role, but it's actually a pretty small role because the time horizons for nuclear are long, time horizons for natural gas are short. And I think we're just going to see more gas used to power data centers, even if that comes at the expense of carbon neutrality goals.

• 40:38Demetri Kofinas?

  So I'm going to ask you now about the sanctions and the loopholes. But before I do, do you ever sort of listen to yourself in these conversations? Or listen to other people that are having conversations about this problem and wonder yourself, how did we go so fast from having this mindset that global warming is the existential threat of humanity to now we're like, how do we burn more coal? How do we build more nuclear facilities? Like, how do we get more energy online? Isn't it interesting how this conversation changed so rapidly after 2022?

• 41:04Speaker 01

  Yeah, yeah, you know, I think that's, I think that's right. I guess it's a combination of energy prices rising and filtering to politics pretty much everywhere. Plus, I think the trade offs have become starker. It's one thing to say, I think we're just going to change our auto fleet. That's hard enough, but to say we're also going to potentially slow progress in AI heightens the trade offs that are involved. And so that that does explain, I think, why the politics have shifted. And it's interesting to watch the big tech firms in the US. They've got really ambitious green power and decarbonization targets that are going to get harder and harder, I think, to meet.

• 41:37Demetri Kofinas?

  Yeah, one of the books I have in my library, I don't know where it is back there, but it's Matt Simmons Twilight in the Desert. And that was, of course, written during the time when we thought we were at peak oil and oil prices were going to skyrocket, we're going to run out of energy. And it's just another example of why people should be humble about what the future holds, because what you think is certain can change just like that, just a matter of a few years. So let's talk about the export controls and the chip restrictions. Where are they working well right now?

• 42:03Speaker 01

  I think they're working well when it comes to certain types of tools that are illegal to transfer to China. So best example, EUV lithography tools, you can't transfer them to China. There is really no way to smuggle them. There's no evidence that these have been broken at all. There's a blanket ban on shipments to China. So it's super clear, these are huge tools, hard to smuggle, no circumvention. They're working less well on the chip making tool side, where you can sell tools to certain factories, but not to other factories. And then the question comes, well, which types of chips are being produced in which type of factory? And the reality is that Chinese firms have got a lot more scope to obfuscate that question than we have ability to enforce. And indeed,

• 42:47Speaker 01

  there's been some really fantastic research that some of my colleagues have done, showing how there's two fabs next to each other. One is supposedly producing less advanced chips, the other is supposedly producing more advanced chips. You can sell certain tools to one that you can't sell to the other, and there are wafer bridges. So an aerial tunnel from one to the other, where you can transit a silicon wafer between them. And we're asked to believe that only non-restricted Not advanced wafers are transferring between them. This is almost certainly evidence of the tools being used in a fashion that's illegal. And I think the US government has written the rules in a way that makes that type of obfuscation too easy. And I wouldn't be surprised if that changes in the next round of export controls,

• 43:33Speaker 01

  or at least if those rules get tightened up in a way to make that much harder. I think the third area of controls, which is the controls on GPUs themselves, I think those have been effective in the sense that China is struggling to get access to the GPUs that it needs at scale, but obviously ineffective in the sense that there's still a whole lot of smuggling around them.

• 43:53Demetri Kofinas?

  Yeah, so I think the case you're describing involved SMIC and either an affiliate company or a subsidiary company where they had a wafer bridge connecting between them. In other cases, they're also using shell companies to purchase the equipment and then actually use it. And they're restricted in the case where they're supposed to be restricted from using it. So that's one of them, this example of a wafer bridge. What are some of the other examples of how China is evading these sanctions or these export controls, and which I guess fall in the category of sanctions? And why have the loopholes remained unclosed?

• 44:26Speaker 01

  Well, I think the other big example from recent headlines is using shell companies to produce at TSMC, which on the one hand is an admission of extraordinary weakness. You're going to get around the restrictions by going to Taiwan to produce the chips for you. That is the best piece of evidence that you don't have the capacity you need at home. But on the other hand, is obviously a huge loophole. If you can get the most advanced production capabilities in the world at TSMC. I think there are three reasons why we haven't seen all these loopholes been closed. So one is is just administrative capacity. It does take time for the government to learn that these loopholes exist, probably longer than it should, but it takes time. And then, once the desire to close the loopholes becomes consensus in the government,

• 45:09Speaker 01

  then you have to undertake a really long process of negotiation about when it's okay to issue the next round of export controls. And so we had one package of export controls in October 22, a second package in October 2023. It wasn't that nothing new was learned in. That year, it was that the decision was made, we'll do one round of updates a year. And so there were six or nine months in which several loopholes were left open because you didn't have a consensus about when to issue the latest controls. October 2024 has passed, and I think because of the election, there's a new round of controls that will probably drop at some point this month. And again, that will close loopholes that everyone knows exists, but couldn't be closed for political reasons. So that's one explanation why.

• 45:51Speaker 01

  A second explanation is that your loophole could be my market access. And for the equipment firms, ASMLs and Applied Materials and Lam Research, they have been very public in advocating their market be kept open in spheres that are deemed to not have national security importance. And that line is, of course, vague and debatable. And it's certainly the case that those companies now often have half their revenue coming from China. And so closing loopholes means also closing off their revenue. They've been very politically influential voices in the US, in Japan, and in the Netherlands. And the third reason is that the US is genuinely afraid of Chinese retaliation. The Chinese have signaled materials, for example, gallium, germanium,

• 46:33Speaker 01

  rare earths and others will be a focus of retaliation. And I think the US wants to find the right time to issue these controls so that it either minimizes the risk of Chinese retaliation or gives the US time to prepare for the retaliatory steps that Beijing might take.

• 46:48Demetri Kofinas?

  One of the parallels that comes to mind when it comes to these firms being so interconnected with the Chinese market that losing market access becomes in itself a reason to not tighten the controls. An analogy to that is the cybersecurity market and the fact that the NSA oftentimes will not inform American companies of vulnerabilities that exist in their software because those vulnerabilities can actually be used to exploit foreign systems and access to foreign governments. So the point is that we live in a deeply interconnected world. And in that sense, and more interconnected than prior to World War I. And in that sense, the threat landscape and what a global war. There's something that I've spoken to quite a number of people about is it seems that if we were to find ourselves in a World War III situation, Chris.

• 47:32Demetri Kofinas?

  That it would be something more total, a more total war in a way we really never could have conceptualized before, throw into that dual-use technology, it makes things even more complicated. So I'm going to move us to the second hour. I want to talk about this, the commercial knock-on effects on companies, because one of the concerns has been that by losing market share, these companies like ASML, for example, or Applied Materials, lose revenue. That revenue then has a cannibalizing effect on CapEx and their ability to innovate and stay ahead of the curve. And so we end up doing more damage if we don't combine those export controls or any potential tariffs with industrial policy that supports these companies. I also want to talk about the risk of escalation, because that's very real and perhaps becomes more real under a Trump administration. And then we'll have a chance to talk about the CHIPS Act,

• 48:15Demetri Kofinas?

  the allocation of funds to companies building out domestic facilities in the United States. We'll have a chance to talk about TSMC and their Arizona fabrication facility. They also made some recent claims about yield that I'm a bit suspicious about and I want to ask you about. But we're going to do, we're going to move all of that to the second hour, Chris. For anyone who is new to the podcast, Hidden Forces is listener-supported. We don't accept advertisers or commercial sponsors. The entire show is funded from top to bottom by listeners like you. If you want access to the second hour of today's conversation with Chris, head over to hiddenforces.io/subscribe and sign up to one of our three content tiers. All subscribers gain access to our premium feed, which you can

• 48:57Demetri Kofinas?

  use to listen to the rest of today's conversation on your mobile device, using your favorite podcast app, just like you're listening to this episode right now. Chris, stick around. We're going to move the rest of our conversation onto the premium feed. If you want to listen in on the rest of today's conversation, head over to hiddenforces.io/subscribe and join our premium feed. If you want to join in on the conversation and become a member of the Hidden Forces Genius Community, you can also do that through our subscriber page. Today's episode was produced by me and edited by Stiliya Nosnicolaou. For more episodes, you can check out our website at hiddenforces.io. You can follow me on Twitter

• 49:39Demetri Kofinas?

  at Kofinas, and you can email me at [email protected]. As always, thanks for listening. We'll see you next time.