Professor Chris Miller - “Chip War: The Fight for the World's Most Critical Technology”
What this covers
In terms of shaping the balance of power, microchips are the new oil—the scarce resource on which the modern world depends. Today, military, economic, and geopolitical power are built on a foundation of computer chips. Virtually everything—from missiles to microwaves, smartphones to the stock market—runs on chips. Until recently, America designed and built the fastest chips and maintained its lead as the predominant superpower. Now, America's edge is slipping, undermined by competitors in Taiwan, Korea, Europe, and, above all, China. Today China, which spends more money each year importing chips than it spends importing oil, is pouring billions into a chip-building initiative to catch up to the US, even as its growing military might makes it conceivable for Chinese leaders to dream of seizing Taiwan. At stake is America's military superiority, economic prosperity, and its technological future.
Bio Chris Miller is Associate Professor of International History, where his research focuses on technology, geopolitics, economics, international affairs, and Russia. He is author of Chip War: The Fight for the World's Most Critical Technology, a geopolitical history of the computer chip. He is the author of three other books on Russia, including Putinomics: Power and Money in Resurgent Russia; We Shall Be Masters: Russia's Pivots to East Asia from Peter the Great to Putin; and The Struggle to Save the Soviet Economy: Mikhail Gorbachev and the Collapse of the USSR. He has previously served as the Associate Director of the Brady-Johnson Program in Grand Strategy at Yale, a lecturer at the New Economic School in Moscow, a visiting researcher at the Carnegie Moscow Center, a research associate at the Brookings Institution, and as a fellow at the German Marshall Fund's Transatlantic Academy. He received his PhD and MA from Yale University and his BA in history from Harvard University. For more information, see www.christophermiller.net.
Source description (no synthesized summary yet).
Miller argues that semiconductor manufacturing is uniquely concentrated in a small number of companies and countries, creating extraordinary geopolitical leverage, and that this concentration stems from the extreme complexity and capital intensity of chip production, which makes catch-up nearly impossible even for well-resourced states like China.
- TSMC alone produces 90% of the world's most advanced processor chips, and Taiwan produces one-third of all new computing power added globally each year
- The extreme ultraviolet lithography machines required for advanced chip manufacturing took 30 years to develop, cost $150 million each, and are monopolized by a single company
- China produces only 6-7% of the overall semiconductor supply chain despite spending more importing chips than oil, and faces structural barriers to catch-up that are fundamentally different from other industries
This asset isn't compiled yet
You're seeing its claims, ranked. Compile it to build the argument threads, weight them, and check each claim against your library — the full view.
The Soviet Union could not produce advanced computing power during the Cold War despite having smart physicists, a large defense budget, and an economy oriented toward military technology production, creating a production gap rather than a knowledge gap that puzzled historians of Soviet technological capabilities.
“There wasn't a knowledge gap in the Soviet Union that explained why they didn't acquire the relevant technologies and then deployment them. There was a production gap”
The first semiconductors were invented for missile-guidance systems in the late 1950s and early 1960s, with the Pentagon being the only entity willing to pay for miniaturization of computing power to fit into missile nose cones, and this military demand set off technological shifts that created modern computing.
“The first semiconductors were invented for missile-guidance systems. The desire to miniaturize computing power so that it could fit in the nose cone of a missile was something that only the Pentagon was willing to pay for in the late 1950s and early 1960s.”
Soviet military theorists understood by the 1970s-1980s, before the US Military realized it, that American application of computing, sensing, and communications technologies to military systems was transforming US military capabilities.
“Soviet defense theorists had understood at least a decade earlier”
The chip design software market is an oligopoly with three companies controlling chip design tools, and these three companies have an oligopolistic position where you simply cannot design an advanced chip without buying software from each of these three companies.
“in the chip, for example, in a new iPhone, it will be designed using software produced by three companies that specialize in chip design, all of which are based in the United States. And these three companies have basically oligopolistic position on chip design. You simply cannot design an advanced chip without buying software from each of these three companies.”
Extreme ultraviolet lithography machines are the most complex machines needed to manufacture advanced chips, took 30 years to develop, cost $150 million apiece, require multiple 747s to move, and are monopolized by a single company that is the only company in the world that knows how to produce them.
“These machines took 30 years to develop. They cost $150 million apiece. They require multiple 747s to move, and they're monopolized by a single company, the only company in the world that knows how to produce them.”
The US military was beginning to adopt a semiconductor-dependent military strategy called the third offset strategy, which relied on artificial intelligence and autonomous military systems, just as it seemed China was going to substantially narrow the gap between American and Chinese technological capabilities.
“around five to seven years ago, the US launched what's called the third offset strategy, which was a strategy that was intended to match growing Chinese capabilities in terms of the number of systems China was deploying with more capable US systems that relied evermore heavily on technology... using artificial intelligence capabilities and thinking about semiautonomous and, eventually, potentially fully-autonomous military systems.”
Five years ago every foreign chip firm (US, Japanese, Korean, Taiwanese) was investing money in mainland China, but today that investment has basically completely stopped, representing decoupling in action in the semiconductor industry.
“Five years ago, every foreign chip firm, US, Japanese, Korean, Taiwanese was investing money in mainland China. Today, that investment has basically completely stopped... we're seeing decoupling in action. The industries are being decoupled at least at the cutting edge.”
The US government's future willingness to escalate export controls is driven by the belief that future military power and intelligence capabilities will depend on computing, making semiconductor control a national security imperative.
“the US government believes that the future of military power and the future of intelligence capabilities will depend on computing. Therefore, the struggle to control access to the world's most advanced semiconductors is not just a question of relevance to your smartphone or to your dishwasher or your car, but also a question of great relevance to the future balance of military power and geopolitical influence in the modern world.”
TSMC started as the only company in the world that was not designing chips, only manufacturing them, a new business model that let it scale far beyond rivals, gaining economies of scale in both financial and technological terms because producing more chips let it hone its production processes.
“TSMC started, it was the only company in the world that was not designing chips, only manufacturing. They had a new business model, and that business model let it scale far beyond its rivals, gaining economies of scale in financial terms, but also technological terms. Because the more chips you produce, the more you hone your production processes”
The international specialization of the semiconductor supply chain, with Dutch companies focusing on lithography, American firms on deposition and etching equipment, California firms on chip design, and Taiwanese firms on manufacturing, has created extraordinary efficiencies that enable Moore's Law to continue.
“This specialization has worked, in economic terms, absolutely brilliant. But in the past seven or eight years, politics has begun to intrude into the chip supply chain in ways that were unexpected by many key players.”
DARPA has funded literally every aspect of semiconductor technology including all tools, chip designs, and the transition from flat transistors to 3D transistors that happened half a decade ago.
“If you look at a semiconductor, every aspect of semiconductor has been funded by DARPA at some point. All the tools. All the chip designs. My book, I have a line that DARPA has literally shaped the transistors on your iPhone because the transition from flat transistors to 3D transistors, which happened half a decade ago, was funded by-- was DARPA-funded research”
In the chip industry, there are oligopolies and deeply-entrenched market positions in key segments of the supply chain where capital intensity means startups are nearly impossible—you cannot raise $25 billion for a manufacturing startup even with government backing (unless you're in China).
“in key segments of the supply chain, you actually have oligopoly and deeply-entrenched market positions. And in a lot of segments of the supply chain, the capital passivity means that startups are really hard to imagine. You're not going to raise $25 billion for your startup. There's just no way, even if you have government backing. Unless you're in China, you're not going to raise $25 billion.”
China spends as much money importing semiconductors as it spends importing oil, with semiconductors being the most important product China imports, because China cannot make advanced chips domestically, creating both an economic and strategic vulnerability.
“Today, the Chinese government believes that semiconductors are their most important vulnerability because today, China spends as much money importing semiconductors as it spends importing oil. There's no product that China spends more money importing.”
If there is a disruption in Taiwan's semiconductor production, it will be impossible to produce almost any smartphones anywhere in the world the next year because the chips will be unavailable, affecting not only smartphones but also PCs, data centers, and all sorts of goods including cars with 1,000 semiconductors each.
“if something goes wrong in Taiwan, the implications will be felt far beyond that island, far beyond China. In fact, it will be felt by the entire world economy because it will be impossible to produce almost any smartphones anywhere in the world the next year. You won't have the chips.”
Downstream customers in the chip industry (Apple, AMD, Nvidia, Qualcomm, and Mediatek) are largely US firms, which gives them some influence over the supply chain, but this influence is limited by the need to keep ASML happy because there are no alternatives to ASML for critical lithography tools.
“the downstream customers in the chip industry-- the companies that are buying chips, the companies that are paying to have chips manufactured for themselves are largely US firms. So Apple, AMD, Nvidia, and Qualcomm, etc... you need to keep ASML happy, so it limits the tools that you need. There are no other options.”
The US first began restricting machine tool exports to certain Chinese firms approximately 4-5 years ago, and in that time, China has not made dramatic progress in catching up to cutting-edge technology - the gap between Chinese capabilities and the cutting edge has remained basically the same.
“The US first began restricting machine tool exports to certain Chinese firms around four years ago, five years ago. And in that time, it's not obvious that Chinese firms have made dramatic progress in catching up to cutting edge or really actually any progress at catching up to the cutting edge. They made progress, but the cutting edge has moved forward too, so the gap so far as any sort of analysis can show has remained basically the same.”
When the chip shortage occurred in 2020-2021 (despite world chip production increasing 8% in 2020 and double-digit growth in 2021 due to demand surge), car companies faced over $200 billion in global disruptions, showing the vulnerability created by concentration.
“the world produced 8% more chips in 2020, a double-digit increase of rate in chip production in 2021. We had a chip shortage because demand surged higher. And as demand surged higher, car companies faced over $200 billion of disruptions globally in a year where we were producing more chips than previously.”
Designing an advanced chip requires paying for software tools (hundreds of thousands or low millions of dollars) rather than billions, creating much lower barriers to entry in chip design relative to manufacturing.
“In chip design, the capital costs are a lot lower. So chip design, you need to pay for software tools, which are expensive, but you're talking in the hundreds of thousands or low millions, rather than in the billions for your first get-off-the-ground months or year.”
If China is n-minus-3 or n-minus-6 in AI chip capabilities in 10 years is an important question for military power, as being n-minus-5 is a real disadvantage for AI training, especially across an entire society.
“if you ask the efficacy of the export controls on AI chips for military systems in 10 years' time, I think it's going to be really important, actually, is China n minus 3 or is it n minus 6. Because... if you want to train AI systems, being n minus 5 is a real disadvantage, especially if it's not just one facility, but across your society.”
TSMC, the Taiwan Semiconductor Manufacturing Company, today has 95% of its production, including all of its most advanced production, in Taiwan, and Taiwan produces 90% of the world's most advanced processor chips that power almost all smartphones, most PCs, data centers, and cell phone towers.
“That company is called TSMC, the Taiwan Semiconductor Manufacturing Company, which today has 95% of its production, including all of its most advanced production, in Taiwan. Taiwan today produces 90% of the world's most advanced processor chips, the type of chips that power almost all smartphones, most PCs, data centers, cell phone towers.”
In 2014, Xi Jinping and the Chinese leadership launched policy programs designed to wean China off reliance on imported chips and build domestic chipping capabilities, setting up investment funds at national, provincial, and local levels to pour billions of dollars per year into China's chip industry.
“in 2014, Xi Jinping and the Chinese leadership launched a number of policy programs designed to wean China off reliance on imported chips and build up domestic chipping capabilities. And so there were a number of investment funds set up at the national level, but also provincial and local levels, designed to pour billions of dollars-- probably on the low tens of billions of dollars a year into China's chip industry.”
An iPhone contains approximately 15 billion transistors carved into silicon, with each transistor smaller than the size of a coronavirus, making semiconductor manufacturing the most complex and sophisticated manufacturing humans have ever undertaken.
“inside that iPhone, there will be a dozen or so semiconductors. And the most important chip inside of an iPhone, the main application processor, will have carved into the silicon 15 billion transistors. 15 billion transistors, each one smaller than the size of a coronavirus.”
Creating a version of an extreme ultraviolet lithography machine is very difficult, no one has done it, and no one is even trying to do it, so controls on the software tools and machine tools would have a meaningful impact in slowing down China's semiconductor capabilities.
“creating your own version of an extreme ultraviolet tool is not easy. In fact, it's very, very hard. No one's done it. No one's even trying to do it, so far as we're really aware. And that suggested to the US government that controls on the software tools and the machine tools would have a meaningful impact in slowing down China's capabilities.”
Moore's Law has demonstrated sustained exponential growth in transistor density for over half a century, with no other good in the economy showing exactly the same exponential growth, with airplanes, for example, not being able to fly twice as fast every other year.
“there is nothing else in our economy that has demonstrated Moore's Law-like growth for over half a century. Imagine if airplanes flew twice as fast every other year. We can't imagine it. Nowhere else do you get those sustained productivity improvements at such an extraordinary rate.”
The difference between advanced nodes (n) and one generation behind (n minus 1) is roughly like the difference between a new iPhone and a two-year-old iPhone—slightly better battery life and features, but still an iPhone, not a fundamental difference.
“the simplest way to conceptualize n versus n minus 1 is like a new iPhone versus a two-year-old iPhone. Because that basically gets you the one node difference. So one differential is not huge. You get slightly better battery life with the cutting edge. You get a couple of new features, but it's not-- it's still an iPhone.”
The laser system in an extreme ultraviolet lithography machine is just one component that has 457,000 components, and in aggregate the machines have approximately a million components, so if each component breaks once a year the machine never works, requiring nearly perfect reliability across hundreds of thousands of parts.
“the laser system, which is just one of the components in the UV machine, has 457,000 components. That's just one of the key components. So in aggregate, the machines have, call it, a million components. If each one of those components breaks once a year, the machine never works.”
The extreme ultraviolet lithography machines work by pulverizing a ball of tin falling through a vacuum twice with one of the most powerful lasers ever produced commercially, causing the tin to explode into a plasma measuring 40 times hotter than the surface of the sun, which emits extreme ultraviolet light at exactly 13.5 nanometers that is collected by the flattest mirrors humans have ever made.
“inside the machines that produce extreme ultraviolet light, there is a ball of tin falling through a vacuum that's pulverized twice by one of the most powerful lasers ever produced for a commercial device. The tin explodes into a plasma measuring 40 times hotter than the surface of the sun. The plasma emits this extreme ultraviolet light at exactly the precise wavelength of 13.5 nanometers. It's collected by a dozen of the flattest mirrors humans have ever made”
The difference between having a 10-year-old iPhone and modern iPhone shows the importance of continuous chip improvements, as older iPhones couldn't stream video while modern devices can.
“If you're 10 years behind, it's a different story. Remember iPhones were-- when they were first released, streaming video wasn't a thing. It was impossible to stream video on the first iPhones.”
Data centers don't scale linearly without losing efficiency when using less advanced chips, creating inefficiencies if China tries to compensate for technology gaps by building larger data centers.
“Data centers don't scale linearly without losing efficiency, so there's some issues there. But could we see China trying to find ways around it? That's interesting and possible to ask.”
TSMC four years ago began opening a facility in Arizona that will produce n minus 1 technology (one generation behind cutting-edge), not the most advanced technology, with all R&D for next-generation processors continuing to happen in Taiwan.
“TSMC, four years ago really began a process of opening a facility in Arizona. It's a medium-sized facility and per TSMC's latest announcements, it will produce n minus 1 technology. So not the most cutting edge, but one generation behind... all that R&D today happens in Taiwan. And under current plans, all that will continue happening in Taiwan.”
The historical trend in semiconductors is not that countries are catching up to Taiwan, but that all other countries have fallen behind Taiwan - Japan, the US, and others are falling behind the technological frontier rather than catching up.
“The trend has been to fall behind Taiwan, and so the assumption that China will catch up to the cutting edge in 10-years time seems like a very bold assumption to make, given that the trend of the last five, ten years is that other countries-- Japan, the US, et cetera-- have fallen behind the cutting edge rather than caught up.”
The US government is also restricting the transfer of certain tools to Chinese chip firms, focusing on software tools that design chips and machine tools that manufacture chips, because all software tools are produced by US firms and all critical machine tools are produced in the US, Japan, or the Netherlands.
“second, to restrict the transfer of certain tools to Chinese chip firms. And, in particular, the US is focused on two types of tools, the software tools that design chips and the machine tools that manufacture chips. Because all the software tools are used by US firms, and because all of the critical machine tools are produced either in the US or Japan or the Netherlands, the West, in aggregate, has a lot of influence over whether or not you can acquire the relevant tools.”
The $39 billion in direct incentives in the CHIPS Act is not a huge sum given that one new facility can cost $20-25 billion, and to fully replicate Taiwan's capacity would cost hundreds of billions of dollars, far beyond what Congress is willing to spend.
“the CHIPS Act has $39 billion for direct incentives. As I mentioned, one new facility can cost $20 or $25 billion, so it's just-- it's not a huge sum of money. But if you wanted to buy all of Taiwan's capacity and build in the United States, it would cost hundreds of billions of dollars, and Congress is not willing to spend the money.”
US firms represent 45% of semiconductor revenue globally by far the biggest player, with Taiwan, Korea, and Japan each representing 15%, making the US the elephant in the room in terms of industry size.
“if you look across the supply chain, US firms are 45% of revenue, so by far the biggest player with Taiwan, Korea, Japan 15% each... the US is still the elephant in the room”
Most electronics are currently produced in China (PCs, smartphones, etc.), but over the last six months most big OEMs have made public costly statements about new investment in India, Southeast Asia, Mexico, etc., suggesting some shift in electronics supply chains away from China is occurring.
“Right now, most electronics are produced in China. PCs, smartphones, et cetera. It's certainly the case over the last six months most big OEMs have made public costly statements about new investment in India, Southeast Asia, Mexico, et cetera.”
China is now one of the biggest markets, in some cases the biggest market, for tool exports from tool makers in the US, Japan, and the Netherlands, despite export controls.
“China's now one of the biggest markets, in some cases, the biggest market, export markets for tool makers in the US, in Japan, and in the Netherlands.”
The chip industry emerged out of the Cold War military industrial complex and has been shaped by political choices in the US and other countries, with the industry internationalizing across countries that are close US allies or partners, as a deliberate Cold War construction combining policies by elites in East Asia and the US government.
“the process how this works that is increasingly politicized. It's been politicized from day one because the industry emerged out of the Cold War military industrial complex... since the middle of the Cold War, the US government has been very deliberate in trying to bind other countries to it by bringing them into electronics and semiconductor supply chains... the internationalization of the chip and a supply chain that stretches from the US to Japan to Korea to Taiwan was a deliberate construction in the Cold War, a combination of policies pursued by elites in East Asia and by the US government.”
If China could operate captured TSMC facilities, the likelihood is very close to zero because the facilities only work when constantly serviced by the companies that sold them, they require material imports from Japan, US and Europe, and most customers are American so would require cooperation.
“a lot of the machine tools, they only work because they're constantly serviced by the companies that sold them... you need the servicing from abroad to keep the tools working. You need the material imports from Japan and a little bit from the US and Europe to keep the facilities working. And TSMC's customers are all mostly American at this point... I think the likelihood that China 'takes' the facilities might be above zero, but the likelihood that China operates the facilities is-- I think we can-- I take probability seriously. I think we can put it pretty close to zero.”
Corporate lobbying for weakening export controls is a dynamic tension with politicians who only think about export controls when brought to their agenda, creating uncertainty about whether the controls will be sustained over time.
“the challenge of export controls is that they're super tactical and that political leaders only think about export controls when it's brought to their agenda, whereas companies think about them on a daily basis. And so one of the dynamics you have, I think, is you assume regular corporate lobbying to weaken export controls, and politicians engage with them once every six months.”
If China is deterred from military action against Taiwan, it will be because of economic interests—TSMC destruction would be catastrophic to China's own economy—rather than military calculations or willingness to take risks, suggesting economic deterrence may be more reliable than military deterrence.
“If China is deterred, it will be from this type of strategy, it will be because of economics. OK. But we can debate whether that's--... the answer to your question is if China is deterred, it will be-- from this type of strategy, it will be because of economics.”
When the United States was losing the war in Vietnam and discussions about pulling back from Asia were ramping up, Taiwanese and Singaporean elites deliberately doubled down on their bets in electronics manufacturing specifically to get more US investment in their countries and increase US commitment to their defense.
“key countries in Asia like South Korea, like Taiwan, like, Japan and Singapore have deliberately targeted participation in electronic supply chains and in semiconductor manufacturing to bind themselves to the United States. It's not a coincidence, for example, that as United States was losing the war in Vietnam and as discussion in the United States was ramping up about pulling back somewhat from Asia, that Asian elites in Taiwan and Singapore and elsewhere doubled down on their bets in electronics manufacturing, precisely to get more US investment in their countries.”
The unique barrier to entry for semiconductors compared to other technologies is that the technological frontier moves so rapidly and grows exponentially that catch-up requires hitting a moving and accelerating target, making it fundamentally different from other industries where you can reverse-engineer a design.
“The thing about the chip industry that is unique from most other industries is that the technological frontier moves so rapidly that you're not trying to catch up to a static target. You're trying to catch up to an exponential-growth target... That's why, again, as I said, the question isn't, why can't China catch up with Taiwan? It's, why is everyone else falling behind Taiwan?”
Most companies prefer to go to TSMC for manufacturing and will go to Samsung only for better pricing or when they want to avoid concentration risk, making TSMC the industry standard.
“TSMC, you can go to Samsung to produce something pretty close to what TSMC can do, but most companies, I think it's fair to say, go to Samsung so they can get better pricing than TSMC. And they prefer to go to TSMC as often as they can. I think it's the industry standard.”
The goal of the CHIPS Act is not complete self-sufficiency in chip production but rather to increase the share of advanced processors produced outside Taiwan from the current 10% to potentially 15-20%, and to reduce Taiwan's current 90% share of advanced processor production to 50% within 10 years.
“So self sufficiency is not the goal... I think although the US government hasn't stated numbers, and I don't think they'd be comfortable even stating numbers in private, I think if you look at the policies being pursued by all these different governments and ask, in 10 years time, what share of advanced processors there are produced in Taiwan? the goal is to get that number down from 90% to 50%.”
Data centers will become the weapons factories of the future because that's where AI systems are trained to operate more effectively, with drones learning to fly, sensors learning to sense, and computer vision systems learning to recognize military vs civilian targets all happening in data centers.
“in 30 years' time, we're going to realize that weapons factories in the future are data centers because data centers are where you're going to train systems to operate more effectively... That's where drones learn how to fly, and sensors learn how to sense more accurately. Computer vision systems learn how to recognize cats versus dogs or tanks versus cars. It's all in data centers.”
The US is in a position of escalation dominance in the semiconductor sphere because the US is severing relationships and China can't retaliate because any retaliatory steps will cause companies on their own to pull back, accelerating decoupling.
“the US is in a position of escalation dominance in this sphere. Because the US is severing relationships, and China can't retaliate because any retaliatory stops will cause companies on their own to pull back.”
Taiwan is simultaneously being told by the US that the US will defend Taiwan while also being pressured to move semiconductor production off the island, creating a complex and contradictory diplomatic situation.
“the US is simultaneously trying to convince Beijing that we're really going to defend Taiwan. And so all of Nancy Pelosi's visits, these are all part of a broader government-wide policy of making our deterrence more credible by political signaling while simultaneously telling Taiwan that we'd like their most important business to move some of its production off the island.”
Japan has undergone a significant shift over the past couple of years toward hawkish views on economic security in semiconductors, with METI Minister Nishimura's January speech being as hawkish as statements from any world leader in a long time, suggesting Japan independently arrived at similar conclusions to the US rather than being pressured.
“I think Japan has gone through a very big shift over the past couple of years that has paralleled the US shifts... if you read the METI Minister Nishimura's speech in Washington in January, I think you will find as hawkish a statement on what METI calls 'economic security' as from any world leader in a very long time.”
China's catch-up in semiconductors is not the base case because China is trying to pursue domestic development strategies while unable to pursue the integrated supply chain catch-up strategies that worked for South Korea and Japan, and Chinese companies prefer integration into international supply chains over domestic-only approaches.
“catch-up is not the base case. I just don't-- I don't see why it would be the base case, given the structure of the industry, given the trend of falling behind, given the unique challenges that China faces... China finds itself in now is that it's unable to pursue the strategies of catch-up that worked in South Korea... And the reality that, yes, it's a manufacturing superpower across many industries, but in the chip industry, it's a minnow. It's a small player.”
The story of allied cooperation on export controls is never as simple as 'US twists arm, allies respond'—there are lots of interesting domestic, commercial, and political dynamics at play in each country, making cooperation easier than media narratives suggest.
“But for all of the allies, if you will, the story is never as simple as, US twists arm, allies respond. There's a lot of really interesting domestic, commercial, political, et cetera, dynamics at play, which is why this has been a lot easier, I think. The media gloss six months ago was, Biden administration fails to produce multilateral control system, therefore act unilaterally. That's not the right explanation.”
When the US cut off oil exports to Japan in 1941, Japan could not drive cars in Tokyo by early 1942 without a policy change, but current US semiconductor restrictions are much less severe because they only restrict advanced AI training chips for data centers while allowing imports of smartphones, PC, and car chips to China.
“one of the questions I'm often asked is, well, the US cut off oil exports to Japan in 1941, and then they bombed Pearl Harbor months later. And that, to me, just seems like a fundamentally different situation... Whereas all the US is saying now is your data center companies can't buy the most advanced chips. You can buy less advanced chips So we're talking about of the share of chips coming to China-- iPhone chips, PC chips, car chips-- totally unaffected.”
The US government had owned fabs that became too expensive, and even though it has enough money to build aircraft carriers, it doesn't have enough money to build its own semiconductor facilities, so the US has relied on trusted foundries with commercial partners.
“Right now, the US government had own fabs, and that became too expensive. We have enough money to build our own aircraft carriers, but not our own semiconductor facilities. And then, the US government had the trusted foundries that they operated with commercial partners”
The US government did not have a strong incentive to worry about geopolitical risks in semiconductors during the 1990s-2010s because 'I want more risk' was not high on the agenda, and the industry was happy with current arrangements.
“when TSMC began to grow, both in terms of volume chips produced and in terms of its role at the cutting edge, it was 1990s, the 2000s, and up to the early 2010s when 'I want more risk' was not very high on anyone's agenda. The US government, for a couple of decades, basically stopped thinking about semiconductors.”
If China is about to succeed in taking Taiwan, the likelihood that somebody destroys TSMC facilities is plausible, with multiple parties having incentives to do so (the US, Japan, Taiwan itself).
“the likelihood that somebody destroys the facilities, I think, seems plausible. I can think of multiple parties that would have an interest in doing so.”
China produces around 6-7% of the overall semiconductor supply chain, with nothing that China produces that can't be produced elsewhere, making decoupling on semiconductors relatively easy because China is a small player in semiconductors.
“China produces around 6% or 7% of the overall supply chain. There's nothing that China produces that can't be produced elsewhere. That's an interesting fact, right? And so I think probably not to be-- just trying to be objective, it seems like decoupling on semiconductors is not that hard because China's a really small player in semiconductors.”
Korea produces primarily memory chips (a more commoditized product) rather than advanced logic chips, and was just about to face Chinese competition when the latest US export controls were implemented, making Korean memory chip producers among the greatest beneficiaries of the controls.
“In Korea, I think it's quite interesting in Korea. Korea produces primarily memory chips, which are more commoditized chips... where there was a Chinese firm that was basically at the cutting edge, about to begin high-volume manufacturing this year. And so the greatest beneficiaries of the latest US export controls are a couple of Korean firms that were just about to face competition.”
Jet engines are another example of advanced manufacturing with really complex supply chains that require high precision, where stealing designs isn't enough and where catch-up is similarly difficult.
“I think second is, I think there are certain types of advanced manufacturing that have really complex supply chains, require really a lot of precision, where it does seem like stealing designs isn't enough. Jet engines is the other obvious example.”
One of the metrology tool makers knew where 143 out of 144 tools it sold three decades ago were located because you cannot operate these tools without pretty regular servicing, making the tools inherently tracked by suppliers.
“The tool that was brought online three decades ago, they sold, I think, 144 tools, and he knew, three decades later, where 143 of those tools were. Because you can't operate these tools without pretty regular servicing.”
Wherever Chinese firms can access foreign semiconductor tools, they keep doing so and make backup plans, but they continue to prioritize foreign tools because their goal is to produce and sell chips, not to develop homemade machine tools.
“wherever Chinese firms can access foreign tools, they keep doing so. They make backup plans too, but they keep accessing foreign tools... your goal is to produce chips and to sell them, not necessarily to support the production and development of homemade machine tools.”
The brain drain and exit of R&D personnel from China as US export controls restrict their ability to conduct advanced research creates an additional barrier to China's catch-up, as talent leaves the sector.
“Most of the wafers that TSMC produces are for US firms. You need the personnel there. You need them all to do their job and not try to sabotage it”
There is significant acquisition activity in machine tools and software tools by big firms buying smaller companies developing specific components or improvements, which may or may not be an optimal system for innovation given the capital intensity of manufacturing.
“There also is some happening in the machine tool space for specific components... you can create your component and sell it to one of the big firms. And so you see a lot of acquisition activity by the big machine tool companies and also by the big software tool companies buying smaller firms and getting plugged into these broader systems. Whether that's an optimal system, I think, is an interesting question.”
US military historically doesn't do as well as it could in applying cutting-edge technology quickly to military capabilities, though it does okay compared to other countries, suggesting there is room for improvement in technology application speed.
“the Pentagon historically-- I guess it does OK in comparison with other international players, but I think there's room for more.”