Today’s Stratechery Interview is with Chris Miller, the Associate Professor of International History at the Fletcher School of Law and Diplomacy at Tufts University, and Jeane Kirkpatrick Visiting Fellow at the American Enterprise Institute. Miller’s focus is on economic history, particularly in Russia from the time of Peter the Great, through the USSR, and onto the current day.
Last year Miller published his fourth book entitled Chip Wars: The Fight for the World’s Most Critical Technology. I highly recommend this book: as I note below, I believe this history of the chip industry benefits greatly from Miller’s perspective as a historian, and not as a technologist, and it gives very valuable context for understanding why chips are so important in the context of China and Taiwan in particular.
In this interview we discuss why the USSR failed to develop a chip industry even as Japan succeeded, and why globalization has been a feature of chip-making from the beginning and how that led to Taiwan and South Korean dominance. We also discuss the surprising relationship between civilian and military applications, whether or not China can catch up in the face of the chip ban, and end with a philosophical discussion of structural forces versus the Great Man Theory.
To listen to this interview as a podcast, click the link at the top of this email to add Stratechery to your podcast player.
On to the interview:
An Interview with Chip War Author Chris Miller
This interview is lightly edited for clarity.
The USSR and Chips | Japan’s Success and Failure | Chips and Globalization | Silicon Valley and the Pentagon | Taiwan and South Korea | China’s Chip Strategy | Taiwan and Great Man Theory
The USSR and Chips
Chris Miller, welcome to Stratechery.
CM: Thanks for having me.
So the obvious reason to have you on Stratechery is your recent book Chip War, but I actually have a meta question about book publishing which is, on one hand the timing of the release of Chip War could not have been better, I think it came out the same week as the Biden chip ban on China, which is about the best publicity you could ask for. Like, “Hey, you want to know what’s going on? This book just came out, go check it out.” On the other hand, some of the biggest happenings in chips and politics is not in the book! I guess that’s what second editions are for, but what was that experience of this publishing and everything happening all at the same time?
CM: Well it’s funny, the book was published the week of the export controls, but it was also published two months after the CHIPS Act was finally passed through Congress.
That’s right. I forgot about that part too because of the long lead time. This is why I stick to blogging, it can be very timely.
CM: But no, it certainly was a good publicity for the book and I think helped sensitize people outside of the semiconductor industry as to why chips matter and why they ought to pick up a book and understand the role that chips play in their lives.
Well, I do endorse the book. What I enjoy about it has to do with your background, which is you’re not a technologist and I actually think that makes a lot of what you write about the chip industry very approachable for people that don’t necessarily have a lot of background about it. It’s a good primer, not just about the current issues, but about how the whole thing works and how we got to where we are. To that end though, I want to know more about your background. I think it’s quite interesting and not just your experience as a professor, but backing up, how’d you end up where you are, where’d you grow up? I want the whole kit and caboodle to lay the groundwork here.
CM: So I grew up outside of Chicago and both of my parents worked in tech, sort of, at a company called Comdisco, which did computer leasing in the 1990s and 2000s so I was always cognizant of the tech industry, but never worked in tech, never studied anything tech related. I did a PhD in economic history at Yale, which is how I came to be a historian. Most of my academic work focused on Russian economic history, and I wrote three books on different aspects of Russia, and I was planning to write a fourth book on the history of Russian technology and about what went wrong in the Russian computer industry given its importance to defense technology. When I started writing that I came to learn a lot about the relationship between the chip industry and the defense industrial base, both in Russia but also in the Soviet Union. That actually is the kernel that led to Chip War.
To me that was one of the most — I mean there was a lot of interesting parts — but that was one of the most interesting parts of the book was your discussion about the Soviet Union and their attempts to compete in the semiconductor industry. It’s always tough because this is the part where you’ve been immersed in it sort of your entire life, so it’s always hard to summarize. But what’s the big picture history and lesson from Russia, I should say USSR, and its attempts to compete with the US in particular?
CM: The puzzle to me was the following: we knew the Soviets could produce a lot of impressive technology because they did it during the early Cold War. From atomic weapons — which granted they stole some of the designs, but nevertheless, they were the second country in the world to test an atomic bomb — to satellites, they were the first in the world to go into space largely thanks to indigenous innovation, the first person in space, Yuri Gagarin. So in the 1950s the Soviets weren’t seen as technologically backwards, they were seen as, if anything, overtaking the United States, and that made sense because if you had to ask what are some of the key ingredients to technological success of a country, you’d say, well, you probably want a pretty well-educated workforce, Soviets had that. Capital investment, Soviets had that. You want to focus on the industry, Soviets had that. And so the puzzle to me was why, given all these clear ingredients that were present in the Soviet Union, plus the pressure of Cold War competition to produce the next best defense technology, why was it that the Soviets couldn’t produce computing technology basically at all, and the entire Cold War they were copying IBM computers? That was the puzzle I initially started out wanting to answer and there’s a number of different ways you can answer the question.
I think this is super interesting, it’s super relevant. So walk me through them — what was it that was fundamentally different about, to your point, putting a man in space versus building a semiconductor?
CM: I think the common answer in the Western literature is “Well, they were an essentially planned economy, or they were dictatorship or both, and those societies can’t innovate”. I think that just doesn’t fit the historical facts. In fact, they did a whole lot of innovation in certain spheres at certain times, but there’s nothing about dictatorships that make them non-innovative, they innovate for their own reasons. But I think the problems the Soviets face were the following: First they didn’t have a consumer economy, hardly at all.
Why did that matter?
CM: That mattered because from almost the earliest days, the chip industry in the US, the computer industry in the US, grew thanks to sales to civilian markets and sales to consumers. The first chips that were produced were deployed in government systems, NASA and the Defense Department. But by the end of the 1960s, a decade or so after the first chips had been produced, it was civilian sales that were driving the industry. Today it’s 97% of chips produced that go to civilian uses, and so if you don’t have a civilian market, you can’t scale, simple as that.
I think this fits in because if you’re trying to get a man into space, you’re trying to get one man into space one time. Whereas the entire economics of chips and of the tech industry generally is 100% about scale. You have to put such massive investment upfront, and then the cost of goods sold for a chip is basically zero, and so to justify and to get a return on that investment and to provide the space for iteration, you need that massive demand to make it all worth it. If you just try to do a single shot, it’s probably not going to work out.
CM: Yeah, that’s absolutely right. The second thing that I didn’t realize is that I was under the impression when I started that nuclear bombs were hard to make, but computers were easy to make because there were a few nuclear bombs in the world and a lot of computers, and actually it’s the exact opposite. Nuclear bombs are so easy to make, even the North Koreans can do it.
(laughing) I don’t think I have any new North Korean subscribers, so no problem with that statement.
CM: I’m safe, okay. Whereas actually it’s the things that are the most widely produced, like chips, that are the hardest ones to make because you’ve got to drive down the cost, you’ve got to scale down the components on them, and that is the most complex manufacturing we undertake. I hadn’t really thought that through and I think most of us haven’t really thought through that dynamic and as a result, it has us focusing on the wrong types of complexity and the wrong types of technology and we, I think too often, overestimate the complexity in things that are done once and underestimate the complexity involved in scaling.
Yeah, this is catnip for me. This is in some respects an articulation of the entire reason why Stratechery exists, which is I feel no one understands the implications of goods that require massive upfront investment and then basically zero marginal costs on the backend because you only see the backend in consumer markets. I mean you are talking about how it applies to national security and things along those nature, but it applies to business, it applies to markets.
The example I think is a great one is the auto industry in the COVID era where “Hey, a $1 chip, whatever” all the suppliers were putting freezes on all of them without appreciating that $1 chip was actually the most complicated thing in their car, and if they couldn’t get it, they were screwed and you couldn’t just ramp up on demand like you could a transmission, which would seem much more difficult to make.
CM: Yeah, I think that’s absolutely right. And that gets to a third thing that I hadn’t fully appreciated and that’s Moore’s Law, because it’s the scaling of the industry that makes possible the scaling down of the transistors. Like everyone else I had heard of Moore’s law, but I hadn’t really made sense of the fact that the entire rest of our economy doesn’t work with Moore’s Law dynamics. Doubling every two years is something that is not seen anywhere else in human economies, nothing else has come remotely closed. The example I like to use to think through it is what would aviation be like if planes flew twice as fast every two years? It’s mind-boggling, it’s just inconceivable, and yet we’ve all taken for granted that the chip industry will give us 2x the performance every two years without even thinking about it.
That’s an interesting combination with the infinite malleability of software, where those two things go hand in hand, and I guess this leads to a question which is in the long run, did you circuitously in some respects come back to top-down economies can’t innovate, because isn’t that sort of the takeaway here? What builds that consumer economy, because Moore’s Law is a choice, it’s not like an inevitable sort of thing. What gives the impetus to do that and to build up that consumer economy is the fact that software is infinitely malleable and you can make basically anything you want, which is at the end of the day, innovation actually is the ultimate incentive.
CM: Yeah, I think that’s right in the sense that Moore’s Law is the result not of an innovation but a half century or now longer of a series of innovations piled on top of each other that were all made for their own reasons, but all collectively ended up being Moore’s Law. So it’s not, “Make this missile system”, which you do once and you succeed or you fail. Moore’s Law emerged because you had thousands of different engineers at dozens of different companies, all for their own kind of logic and market rationales trying to race forward and collectively producing these advances in computing.
Yeah, it kind of is a technology that is intrinsically opposed to central control, it has to be the opposite.
CM: Has to be.
Japan’s Success and Failure
Tell me about the contrast between the Japanese approach to chips versus the Soviet approach. Why was Japan so much more successful in entering this US-dominated industry relative to the Soviet Union?
CM: Well, the Japanese entered the chip industry not by trying to copy illegally, which the Soviets did, but by licensing technology. They were among the earliest licensees of the transistor after it was first produced, early licensees of the first integrated circuits, and they produced them better. The first chips began to be commercialized in the early ’60s, and just 15 years later, the late ’70s, Japanese firms by all accounts were producing at much higher levels of quality than US firms.
The complaint about dumping was never really quite right. People bought the Japanese chips because they failed much less and performed much better.
CM: That’s absolutely right. You had US CEOs at the time saying, “Well, we’ve got the real technology, we’re the most advanced in terms of this and that criteria”, but actually the technology that mattered again was the scaling. Japanese firms could scale with quality to a much greater degree. But that’s also what did them in, because they didn’t do a good job of managing their capability of scaling with market dynamics and they weren’t guided by profitability or guided by market share as their goal. So Japanese firms took over the market for DRAM chips, the type of memory chip that was the most prominent chip at the time, and never made any money. Kind of shockingly they dominated the market for a decade and hardly any of them ever posted a profit.
Well, I guess to just speak about Japan for a moment, because I think it’s interesting, first, why did South Korea and then also Micron in the US surpass Japan in memory, and second why did Japan never build any strength in logic? They peaked with memory and that was sort of it.
CM: So I think on the second question, Japanese firms did try to move into microprocessors at a time when they were still a niche good in the late ’70s and early ’80s, but they were doing so well in memory or it seemed like they were doing so well in memory that it was an Innovator’s Dilemma type situation. They had huge market share in memory, they had just defeated TI and Intel in then DRAM business, so why would you switch your business model to produce this low volume type of chip that seemed pretty niche? Whereas if you were Intel in the early ’80s you had no choice, you’d just been knocked out of your primary market.
It’s very underrated. Everyone wants to talk about that apocryphal, or maybe I guess it was real, meeting with Andy Grove and Gordon Moore where they’re like, “We need to get out of memory”. But it’s under-appreciated that this was not a brilliant flash of insight, this was accepting reality and probably accepting it a couple years too late.
CM: Yep, I think that’s right. I think the other benefit that the US ecosystem had writ large was that it was more responsive to new trends in the PC industry, and just the emergence of the PC itself is something that — could it have happened in Japan? I think you wouldn’t say it couldn’t have happened, but it seems like all the ingredients were much more prevalent in the US. A bigger software design ecosystem, Bill Gates being the critical representative, plus companies that were willing to innovate more rapidly to produce PCs. At the time there were a couple of Japanese firms that were good at productizing new ideas, Sony being the best example, but Sony was the exception, not the rule. What really struck me about the PC industry is that IBM created the first PC, but then they were quickly out-competed by all the clones that emerged, which drove down the cost and drove up the prevalence of PCs.
For someone that started out saying, “I assumed that the story was free markets just being better at innovation and that wasn’t the case”, I don’t know, that sounds like the case that you’re kind of making right here.
CM: (laughing) Yeah, in this case, I think it was. The Japanese did a very good job at scaling, but here is the counterfactual: Suppose that Japanese firms had been disciplined by a need to make a profit, they would’ve focused less exclusively on simple scaling to win market share. They would’ve at an earlier date tried to ask themselves, can we make money in DRAM? Some of them I think would’ve exited DRAM because they didn’t make any money there and tried to do something else. So actually I still go back to the structure of the Japanese corporate and financial system as to why their chip firms just for far too long focused on producing unprofitable chips.
Yeah. It sounds like central economies, whether in a product sense or a financial sense, do bad at innovation! We will come back to that in a bit.
Chips and Globalization
One of the more interesting points you made in the book, that I think is very important in understanding things around the CHIPS Act and everything that’s going on, is that not only are chips and globalization not a new thing, but in fact, chips were at the leading edge of globalization and were arguably the first product to be globalized. This started with Fairchild Semiconductor, even pre-Intel.
CM: Yeah, that’s right. [Fairchild Semiconductor and Intel founder] Bob Noyce had an investment in a transistor radio facility in Hong Kong, and so when Fairchild was looking to set up an offshore assembly facility for assembly of its transistors and its ships, they turned to Hong Kong because Noyce already knew the area. So from 1963, Fairchild was operating its first offshore facility, so five years after the creation of the first integrated circuit, Fairchild was already offshoring and that trend continued to Hong Kong, Singapore, Taiwan.
This was actual assembly, right? Which back in the day you were literally physically connecting chips or connecting wires to chips.
CM: That’s right. It started with assembly and then slowly in places like Korea and Taiwan moved up the value chain. But the earliest stages were very basic assembly.
Where was China in all of this? You have the Soviet Union that is copying, they recognize the importance, but they’re trying to copy it, Japan licenses. Then the way we’ve seen in other areas of industry, I think autos are probably the most famous example, applying high production skills and capabilities which for semiconductors, the whole point is to have an identical chip that comes out at the end and to do it at massive scale. You have Hong Kong and then extending to Southeast Asia doing these very labor-intensive aspects of chips, which today are increasingly automated, but back then was literally, again, just like you had to physically use your hands to connect wires. What was China doing this whole time?
CM: Well, the short answer is not much because this was first, the Great Leap Forward and the Cultural Revolution were both happening in the early decades of the chip industry, and so the Chinese government was trying to experiment with some military focused chip making. But in reality, there was not much at all, and there was actually some debate in the Chinese Communist Party about whether industry could even think about moving towards electronics, whether they should focus solely on smokestacks and coal and steel and the heavy industries that had defined Mao’s view of socialism. So the Chinese were really late to the game when it came to semiconductors.
The US was spurring this expansion into Asia broadly of the chip industry, correct? Just as a economic counterweight to communism?
CM: That’s right. The US wanted to plug in its Asian allies, not only militarily via the Alliance System, but also economically. If you think of the geopolitical situation in the 60s and 70s, the fear was that there’d be dominoes falling in East Asia as countries fell to communism, so there was a goal both to integrate them, but also to provide a lot of jobs for peasants moving off the farms into the big cities in Southeast Asia. CHIP assembly was a perfect way to do that.
Yeah. Because it was small, the container revolution hadn’t happened yet — I think the container was standardized in 1968, I believe. So chips were small, they fit on a plane and they required a lot of labor. I think that that’s a very under-appreciated reality at the time.
CM: There was an interesting alliance that was de facto forged between some of the leaders of Southeast Asian countries, Lee Kuan Yew, the Taiwanese government, the Korean government and the US that had the military aspects to it, but was also about this overlapping interest. These countries wanted economic development, the US wanted a cheap workforce, and it matched up the economic and the political goals.
The US chip companies had the obvious motivation of cheap labor, that was the gating factor as far as chip production goes, so everyone’s interests were aligned and I think that’s a really important and useful observation to make today. When people talk about how did chip manufacturing end up in Taiwan? It’s like that was the goal, it’s like a mission accomplished.
CM: Yeah, that is absolutely right. If you go back to the earliest days of the chip industry, before they started doing offshore assembly, there were efforts to find cheap sources of labor in the US. There was a chip assembly plant set up on a Navajo reservation, for example. There were executives sent to scour the lowest wage parts of the US, but there were just no ways to do it that was economically viable, and so that’s why US chip firms wanted offshore labor from the outset and globalization wasn’t an error, as you say, it was the strategy from day one.
Why Asia in particular? I mean, we’ve danced around the idea that the Japanese were just pretty good at high scale, high quality manufacturing. Was there some aspect that made Asia particularly attractive in this regard, or was it just a matter of that’s where we were worried about the dominoes falling? It was economically in the right place, what were the factors that went into how it’s Asia specifically that dominates this space?
CM: Yeah, I guess you could say, “Why wasn’t it South America” or “Why wasn’t it Mexico?” would be the counterfactual. There were a couple of assembly facilities across the Caribbean, and even today there’s a bit of packaging in Costa Rica, for example, some in Mexico.
I think there were two reasons why Asia. One is the political overlap that as you say there was a desire to push investment there for the political reasons. I think the second is that when US firms arrived, they realized that they not only had a cheap workforce, they also had in some ways a recently well-educated workforce and especially as you get to the 1960s and 70s, people would arrive in Hong Kong or Singapore and you could hire someone to be the foreman on your assembly line and they’d often been trained in a US or British University with an electrical engineering degree, and you could pay them local wages. So it wasn’t just that you were getting cheap wages, you were getting cheap wages with a skilled workforce.
Silicon Valley and the Pentagon
Switching gears just a little bit, and going back to something you mentioned earlier. One of the common critiques of tech, or people that want to bring Silicon Valley to heel, is that it’s all downstream from the government, you only exist because of the government. You made this point before and one of the takeaways I got from your book is that the Pentagon was actually a beneficiary of Silicon Valley much more than the other way around. It turns out military applications for chips are not that complicated, they could just take stuff off the shelf. Was that your broad takeaways here?
CM: Yeah, I think that’s right. Certainly, the Pentagon put a lot of money in the chip industry, especially in terms of long-run R&D, there’s no doubt about that. But it’s also clear that the key innovations in the chip industry happened when people took technologies that were developed for military purposes with broader implications that were unclear and then repurposed them in a far more important way to civilian markets.
You look at Bob Noyce, for example, before he founded Fairchild, Noyce worked at an old school electronics firm in the East coast where he’d worked on a research contract run by the Army and he thought it was a horribly bureaucratic, stifling environment. He quipped at one point that he was absolutely sure that he knew a lot more than the Army supervisor who was running the research contract, and so the Army had no idea what direction that his research ought to be taking. So when he founded Fairchild, he sold to the government, no doubt, but he was also from day one looking for how to commercialize this, how to take civilian markets. That was really key because if we’d only been selling to NASA, the Defense Department, we wouldn’t have gotten the scale of economics that we started with.
One of the other bits you talk about, an extension of this, was that the US military acceleration in the late 70s or 80s, I guess is really the decade that mattered, was about leaning into the US chip industry. Walk me through that and the role it played in ending the Cold War?
CM: So the US military in the 1970s faced a dilemma, which was that the Soviets had built up a much bigger force, especially in Europe — they had more tanks, more troops, et cetera. The US had to either decide either not to compete, that wasn’t viable, or to try to compete quantity wise, but that wasn’t financially viable, and so the third strategy that was much more attractive was trying to offset the Soviet quantitative advantage with the US qualitative advantage.
There were a number of key figures here, but the most important was Bill Perry, who was later Secretary of Defense in the Clinton administration, but he lived in Silicon Valley. He was in the same Madrigals Choir group with Bob Noyce, and so he just had a deep appreciation of what Noyce and company were delivering day after day in terms of new computing capabilities. Perry ran a small defense contractor that built intelligence equipment for the US government, the intelligence community, and he realized the ways in which the computing that Noyce was producing, better microprocessors every single year, was going to be transforming the types of equipment he could field. So when he was Under Secretary for Research and Engineering at the Pentagon in the 1970s, he bet everything on trying to develop military technologies that were more dependent on computing power and taking the best Silicon Valley could offer and putting it right into the next generation military systems. That was important because that was something the Soviet Union just couldn’t do, they didn’t have the industry, and so that set up the US for a major leap forward in capabilities that the Soviets simply couldn’t match.
Even today, a lot of the weapons that we’re witnessing in the war in Ukraine are basically from that time period. But this is an interesting point in general — you mentioned before that actually the things that we have a ton of are more difficult to make than the things we only have a few of, which is counterintuitive — it seems to also be the case that the application of technology to military, to date, is actually pretty low tech, right? I mean, the chips we had in the 70s and 80s were more than sufficient for most of the applications that we’re even witnessing today.
CM: Yeah, I think it depends where you look. So if you ask what kind of chip do you need to drive a tank, it doesn’t need to be that advanced, no doubt about it. But if you start going into intelligence systems, communication systems, and in military use, in civilian use, signals processing is absolutely critical. You’re getting tons of data, how do you make sense of it? That’s where you get a lot more demands in computing power and those are things you don’t see. You don’t see the computing behind your radar or your sonar, you don’t see what satellites are up to. But that’s where actually the more computing in intensive parts of the military are today, it’s not in the tank or even in the missile system. Missile guidance systems were the origins of the chip industry but today, your iPhone can guide as accurately as any missile.
Drastically more accurately. Didn’t you say that Texas Instruments made that chip for guided bombs and it basically hasn’t changed in 50 years?
CM: For their laser-guided bombs, there’s been very, very little change. Now most bombs have GPS guidance in them, so there’s been a bit of a improvement. But yeah, the GPS guidance is not all that different from what’s in your iPhone.
Taiwan and South Korea
So fast forwarding, one of the way things evolved today — was the die cast back in the 50s and 60s as far as Taiwanese dominance or Asian dominance in general? Obviously South Korea is a major player, Samsung is still the largest memory maker in the world, I believe, Hynix is there, and then you have all the testing and packaging is in Malaysia and in Taiwan also, more and more in China. Was that path inevitable by the time we got to the late 1970s, 1980s?
CM: I think it was not inevitable for a couple of reasons. I think there’s a small number of firms that we’re talking about that really shaped the geography — Samsung being one, TSMC being another, Intel being a third. You could easily, I think, imagine different choices made by key leaders at each of those firms that would’ve produced a entirely different geography of the CHIP industry.
What are the choices?
CM: So start with TSMC. Morris Chang could have been the CEO of Texas Instruments in the 1980s, in which case TSMC would never have been founded. Would the Taiwanese government have kept investing in the chip industry? I’m sure it would’ve. Would they have maybe eventually created a foundry? I’m sure they would’ve, but I think there was something unique in Morris Chang arriving with all the credibility he had at exactly the right time with the ability to sell immediately to some of the most important customers in the industry that put TSMC in the position it is today. So I think there’s a fair amount of contingency in that outcome and I think the best evidence is that there are other Taiwanese firms that were founded before TSMC that have done less well than TSMC like UMC. It’s not something in the drinking water in Taiwan that explains TSMC rising.
I think Intel’s a great example as well. Intel’s had a series of ups and downs over its history defined by a number of key CEOs and it’s far from inevitable that the Intel that we know today is the Intel we necessarily got. There were a bunch of decisions that could have plausibly been made differently. For example, when Intel turned down the contract to make chips for the first iPhone. Could it have said yes? It very plausibly could have said yes. And if so, the landscape would look very different.
What about Samsung?
CM: The memory business is in some ways the most commoditized of the businesses we’ve been discussing and so as a result, it’s also among the most driven by structural forces rather than individual decisions. Samsung’s dominance is partly a story of their relentless execution, partly a story of the fact that they had a cost structure, tax policy, government support from the earliest, is that also made it pretty viable. I think it wasn’t inevitable though that Japanese firms were completely out competed in memory and Korean firms basically took over, given all the advantages the Japanese firms had in the 1980s. It doesn’t seem to be inevitable that would’ve had to have reversed a decade later.
It’s interesting because the way I’ve thought about it, and correct me if I’m wrong, is that one of the things Samsung did was consistently invest through downturns. Every downturn, they would catch up and catch up and then they would pull ahead, which takes a lot of guts — you’re spending billions of dollars and losing money, but knowing you’ll come out the other side and because it’s a commodity business, if you’re first to a new process, you’re just going to take massive, massive amounts of share. But that seems to be opposed to your point before about the Japanese companies not being necessarily motivated by profit or whatever it might be. Was there a shift in Japan? Was this about the malaise of the 1990s in general, where they felt they couldn’t invest to the degree necessary? How do you resolve that paradox?
CM: I think it’s a matter of who else is in the market. If you’re in the market with 10 other people and you’re all investing through downturns, you’re all in trouble. But if it’s just a small number of players and you’re playing a game of chicken and you swerve last, then you get the market share, and I think that’s the market Samsung found itself in the 1990s. The Japanese were already in retreat because they’d spent the previous decade and a half over-investing, and now their shareholders wouldn’t put up with anymore. There were a number of smaller Taiwanese firms that couldn’t invest through downturns the way that Samsung could. In the US, there was TI for a while who was still in the business, and then Micron, which has been very successful, but has been successful because it’s so carefully deployed capital and its investors have done well as a result of that, but that also means that it hasn’t always competed for market share.
This consolidation point is a big one. There’s a phrase I think from the book that, “diffusion and globalization” — I might be summarizing, I can’t remember, it was in my notes — “Diffusion and globalization were false, what had happened was consolidation and Taiwanization”. What do you mean by that, “Consolidation and Taiwanization”?
CM: Well, people use the phrase globalization to suggest that everyone’s got access to key technologies, and we all do, in the sense we can all buy a smartphone or buy a PC anywhere in the world, but the production is immensely concentrated. TSMC is producing 90% of the most advanced processor chips, we can get to ASML I’m sure at some point, and the extent to which that we’ve had these, often a couple of firms, in some cases, just one firm defining the market, is not what we think of when we think of globalization. But the scale economics point exactly in that direction.
The fact that it’s a worldwide market lends itself to this, combined with this unique nature of chips, where you want to spend a massive amount to get the perfect chip, and then produce millions of them, it works better if it’s one company doing that.
One thing you’ve said about TSMC and ASML is that, “The way to understand them is less about them being manufacturers and more about them being integrators.” So, what do you mean by that?
CM: If you want to turn to ASML, I think they’re the best example of this. They’re a company that on the one hand, manufactures the most complex tools humans have ever made, hands down, and we can dig into them. On the other hand, they’ll openly tell you that, “Their expertise is not in the tools themselves, but in bringing together such a complex supply chain.” At first when people from ASML tell me this, I was shocked. I thought they would be bragging about their manufacturing capabilities, but they were more focused on their systems integration and the ability to manage suppliers all over the world. I admit, I started the project not taking the people who manage supply chains all that seriously, but I came to develop a lot more respect for them, because them doing their jobs well is an extraordinarily difficult thing to do and when you’ve got a supply chain that does involve thousands of suppliers, you’ve got to do it really, really well.
We’ll get to that in a little bit, but I want to jump ahead to the chip ban and things on those lines. First, you had this move by the Biden Administration to limit access to the most advanced chips. The question immediately was, “What about Japan and the Netherlands?” Both countries make a lot of the equipment that goes into a foundry, Tokyo Electron probably most prominently, and then you also have ASML obviously in the Netherlands. But is there a bit where, to your point, maybe we haven’t yet gone far enough? Because if ASML is ultimately a supply chain manager and a system integrator, some of the most advanced components come from Germany, for example. Is the job done, or is there a bit where there’s more companies that need to get on board if you’re going to actually meaningfully restrict Chinese development of advanced chips?
CM: Well, one of the things that the U.S. controls from last October did is they also imposed limitations on the ability of U.S. firms to supply components to Chinese makers of semiconductor manufacturing equipment. Now, that does not apply to, say, German firms supplying components to those Chinese firms, but I think we can safely assume that there are parts of the U.S. government that are watching that very carefully.
China’s Chip Strategy
Let’s reset on China a little bit. So you have the Mao era, they had other things going on I think is a way to put it. When did China start re-engaging as far as chips go? I think the story I think of about China generally is, obviously Deng Xiaoping’s reforms to the late 80s and you come in the 90s, and they joined the WTO around 2000. You have a country that came up in this free trade world, which was greatly to their benefit as far as having access to world markets and building up their manufacturing base, but also to their short term benefit in that they didn’t need to create chips, they could just buy them, but then they never actually figured out how to make them. Was this something that was top of mind or did that just happen?
CM: I think it was just the result of the fact that if you wanted to have advanced electronics, you needed to buy the best chips and they were certainly not possible to produce in China for a long time. You did have certain Chinese firms that recognized the market opportunity and tried to move up the value chain, Huawei being a great example of this in its chip design arm, HiSilicon producing pretty close to world-class design capabilities by around five years ago, maybe a bit more, five years ago before the U.S. Entity Listing.
Also companies like SMIC in its first decade, SMIC was basically a private firm trying to compete with TSMC and UMC, invested in by Goldman Sachs on its founding and was essentially a normal player in the East Asian electronics industry. So there were some private firms that were trying to just be normal private firms. But the Chinese government, more or less ignored the chip industry until 2014 and that’s when in a number of high profile speeches, Xi and a number of other senior leaders said, “This is a core technology that China has to learn to domesticate because it’s too reliant on foreigners, above all the United States.”
To what extent, even back then, was there an appreciation that having power in the chip industry, wasn’t even about SMIC level or TSMC level, it was actually down to the machines that make the chips level? Is that one of those situations where it’s easy to look around and say — I mean you think, it seems with Intel, Intel will gladly benefit from relative ignorance about chip making and say, “Oh, this is the most important thing.” Then you actually go down a level, “Oh, no, you actually need the lithography machine”, “Oh, no, you actually need the mirror, or you need the laser, or you need the metallurgy” or whatever it might be. Has this been a crash course for politicians generally, about just how deep the rabbit hole goes?
CM: I think the entire world’s political leaders have been learning a lot more about the semiconductor supply chain over the last decade. I think you can certainly find Chinese policy documents or speeches by Chinese leaders that articulate a desire to domesticate a lot of the semiconductor supply chain, but if you look at where China’s invested its dollars over the past decade, they’ve largely gone to just building out fabrication capacity. So if you just trace the money, it seems like the Chinese political system writ large, thinks that the problem is insufficient capacity and so it’s built a lot of capacity.
And that capacity is not particularly useful if you can’t get the supplies that go into it.
CM: That’s right. The way the U.S. controls and the Dutch and Japanese controls are written is that they only restrict leading edge and fairly close to leading edge tools. Low end tools are still legal to transfer and so for low end chip making, China’s still got a wide open access to foreign markets.
One of the takeaways I had from your book is that a mistake China made, and this goes back to the 50s, is they in some respects overemphasized chips, in that they felt like we have to have our own domestic industry. Whereas the difference compared to Japan and Taiwan and South Korea is they integrated themselves into the global supply chain and made a beachhead there and expanded so they were looped in as opposed to trying to build a separate stack. Is that a good summary of a mistake China made?
CM: Yeah, I think that’s right. I think the question is whether it’s a mistake China made, or whether that decision was built into the political differences between Xi’s China and the U.S.? Is it possible given the political differences that China could have taken another path? I don’t know the answer to that question.
Well, you can imagine a different scenario, where if you go back to the 80s or 90s, the Korean approach was available to China, but it didn’t seem to even be on anyone’s radar, and it feels like there’s a consistency — to your point, you talked about going back to the 50s and 60s of China saying, “We don’t want to get looped in here, we want to build it ourselves”, and building it themselves has never worked out. But now today, they have no choice but to build it themselves.
CM: Yeah, I think that’s right. I think from the perspective of the political leadership, there’s a “build it at home” mentality, that is built into the way the political ideology is structured. Just to be completely fair, I think it’s also critical to note that the Chinese firm that did the most to try to tread the alternate path that you articulated, is ironically Huawei, which we can discuss separately the espionage allegations and the links to the Chinese military, etc. But there’s no doubt that Huawei was, in some ways, the most integrated of any Chinese firm into the global tech stack, and the fact that the U.S. didn’t allow it to keep pursuing that path I think is an important data point on how viable that alternative path actually was in the long run.
I think this is a really important point because to my mind, the most important episode vis-a-vis U.S., China and chips was not what happened last year, because that was an inevitable outcome of what happened to first ZTE and then Huawei, where any rational Chinese actor would realize this is not sustainable in the long run, to be buying chips from the U.S. Looking back, at the time, I was actually quite concerned about that for this exact reason, the “keeping your enemies close” idea — keeping China looped into the U.S. supply chain is actually a good thing and this is going to spur the opposite. Is that the takeaway that you’ve had in general, or is it a thing where all this was maybe inevitable, and ZTE and Huawei and the Trump administration got the ball rolling, but that ball was going to roll no matter what?
CM: I think I’ve leaned towards the inevitability side of the equation, but I think it’s still not clear how far that’s going to go. The reality is the U.S. and Chinese tech sectors are still intertwined in any number of ways and we’ve gotten some decoupling, if you will, at the leading edge of the chip industry. But then you look further down the tech stack and actually there’s still a lot of inter-reliance and interdependence and so I think it’s actually a really kind of hard and interesting policy question for the U.S. is, well, what actually is your ideal outcome? It seems to me that the best case outcome for the U.S. is where the U.S. is capable of cutting China off certain things and China has still reliant in the U.S. for other things. But, that’s obviously not a position that China wants to stay at.
My view on what China should do geopolitically speaking if I were giving advice to Xi Jinping is — which I’m not, to be clear, I think that’s obvious — is the U.S. wants to continue to allow China to import tools and technologies as you noted, to build trailing edge chips. I think a big impetus for this is they don’t want to destroy the business of a lot of U.S. tech companies, where 30% of their sales were to China, and so it seems like the rational response for China would be to, and I think we’re seeing indicators this is happening, is to basically try to dominate that market.
In this case, use a willingness to be unprofitable as a weapon and to actually do what we accuse the Japanese of doing back in the day, of flooding the market, driving all other trailing edge capacity out, which is basically TSMC and a bit of GlobalFoundries, but there’s bits and pieces still scattered it around. Once you build a foundry, you might as well keep it and then suddenly, the actual chips that are used, to your point, in guided missiles, and are used in cars and are used in appliances were totally dependent on China. That seems like where this is going, does it not?
CM: I think I agree completely, China’s going to build out a ton of capacity. I think there’s some uncertainty as to whether we’re going to have enough demand to meet that capacity build out or not and I think there’s still uncertainty about what our demand will be for lagging edge chips. In ten years time, people who are more bullish on demand say, “Look, every year, there’s on average twenty new chips added to a car.” No one knows how long this is going to go for, but it’s gone for a long time, etc.
And the chip that controls a window going up and down never actually has to get faster.
CM: Right, exactly. So, set aside the uncertainty about the demand picture. If China built out all this capacity, will non-Chinese firms go to Chinese foundries? I think five years ago, the answer was certainly yes. Today, it’s a lot less clear. And when you have Michael Dell on the front page of the Financial Times reporting that his customers are asking him to remove Chinese-made components from PC supply chains, that’s not the political environment that I think will send non-Chinese customers racing to take advantage of cheaper funding capacity in China.
Right. But there’s an aspect of, sure, people who buy computers are cognizant of it, are people who buy a refrigerator going to really care about where the chip in that refrigerator comes from?
CM: Yeah, maybe refrigerators is the market to target, but if you set aside autos, because autos, I think, are a place where you see a lot more nationalism when it comes to component supply chains right now, set aside computing, certainly set aside the server market, suddenly, you’ve set aside half of chip demand in dollar terms right there.
So what’s your — as someone, again, you’re coming in from sort of a historical perspective, but having dived deeply into this — what do you think about the long-term Chinese prospects as far as basically rebuilding the leading edge capacity? This is a subject of much debate amongst people that are deep in the weeds about it, but as you’ve been able to talk to people all over the place, what’s your takeaway? How far behind are they? Can they even catch up?
CM: First off, what does catch up mean? I think that this is really a key question, because catch up doesn’t mean catching up to 2023 levels of technology in ten years time, then you’re five Moore’s Laws behind. So I think we’ve got to define catching up as reaching 2033 levels of technology in exactly ten years time, just as the rest of the world does. That seems to me like a really tall order, because the trend in the chip industry has not been catch up, it’s been fall behind. Everyone’s been falling behind the leading edge in every single node of the supply chain.
At basically every major lithography transition, another foundry falls off.
CM: Yep, exactly. So the Chinese government’s going to put a lot of money behind it, that’s going to help. There’s the necessity of it that Chinese firms face, that’s going to help. I think the Chinese government’s going to do more to wall off the domestic market, which will give some end market for Chinese firms that will help, at least in the short run, for Chinese chip makers. But at the end of the day, if Chinese firms are selling to 20% of global GDP and TSMC is selling to 80% of global GDP, I think I know who I’d bet on.
So what are the implications of this? I mean, again, as you noted, it doesn’t necessarily make a difference for conventional weapons, if we think about today. Is this where the question of AI systems and stuff comes to bear?
CM: Yeah, I think that’s right and right now, we’re seeing a shortage of GPUs, given all the generative AI boom underway. But I guess there’s a more complex long term question, which is — is compute a real point at which the US can try to constrain China’s AI capabilities? I think we’re seeing the US test out that strategy right now.
What’s your prognostication? I’m going to put you on the spot here.
CM: I think there are people who say, “Well if China can’t get access to the most advanced GPUs, aren’t they just going to build data centers that are four times as large or eight times as large or sixteen times as large with sixteen times as many chips, and therefore scale up that way?” You can’t scale down your transistors, you scale up your data centers, is basically the strategy, and then we have to figure out — what are the inefficiencies involved in scaling up your data center? I’m sure they’re pretty substantial.
Well, this is why it’s interesting, I was actually surprised — what the chip ban really focused on was memory interconnects, or interconnects, which is actually the limiting factor in pursuing that exact strategy.
CM: Yeah. I mean, I think you can’t accuse the US strategy of being incoherent, I think that they put their homework into it. Whether it’s going to work, we’ll see. I’ve got a lot of faith in the Chinese government’s willingness to brute force things when it comes to national security, so I think we should expect them to try really hard. But at some point, I go back to one of the more interesting anecdotes from the Soviet experience was an interview of a weapons designer in the Soviet Union, who was asked to explain why it was that he didn’t use the most advanced integrated circuits in his guidance computer in his missile. And his answer was, “Well, our computing industry, sometimes it works, sometimes it doesn’t. The state’s pretty bureaucratic. It’s just hard to work with, it’s not as easy.” The implication was it’s not as easy as buying from TSMC. So I do think if you get a situation where we’re throwing a lot of sand into the gears of the Chinese computing industry, the Chinese government’s going to respond with lots of cash in response and that’s kind of the race that we’re playing out right now, our sand in the gears versus Chinese government cash.
Taiwan and Great Man Theory
You were in Taiwan recently, we were able to get coffee together, which was great, and I was very jealous because you got to interview Morris Chang, who I agree with you is one of the seminal figures in certainly Taiwan history, but tech history broadly, he’s on my personal Mount Rushmore. Did any of your conversations that you had, not just publicly but also privately, change any of your views on the overall landscape, whether that be chips or the prospect of military conflict?
CM: I was struck in Taiwan by the extent to which the mainstream view in Taiwan is that the chip industry is at a point of rapid change. Whereas if you look at capital investment data, you find a whole lot of capital investment happening in Taiwan relative to what Taiwanese firms like TSMC are doing abroad, which in other words, suggests a lot less change in the geographic distribution than headlines might suggest. I came away convinced that chip making in Taiwan by Taiwanese firms is going to play a very, very important role in the chip industry long into the future, despite the headlines.
Yeah, I think that’s right. I’m pretty cynical about the whole Arizona thing. I think it’s a very expensive PR escapade where TSMC gets to say, “We’re diversifying, we’re supporting the US.” Apple gets to say, “Oh, we have chips made in the US. Ignore all the assembly and components that come from China.” And at the end of the day, there’s nothing that has fundamentally changed about the Taiwan-centricity of TSMC’s business, and that sort of Asian-centricity of Apple, or whoever else is involved in this so I think your takeaway is very similar to mine.
CM: Yeah, I think there’s been a lot of focus on what’s TSMC going to do, or what’s the US going to do, or what’s the Taiwanese government going to do, and a lot more focus to be placed on what is Tim Cook going to do in terms of where he’s buying his chips, and what is Lisa Su going to do? It’s actually, TSMC’s customers play as big a role in TSMC’s decision making as whatever the leadership wants, because they’re a customer, and they have to be a customer-focused business. So if their customers say, “Build chips in Timbuktu,” I think they’d seriously consider it.
Yeah, they’ll just charge you for it, as they were clear on their last earnings call. One meta question — you’re a history professor, studied the USSR, suddenly you are the in demand chip expert, you’re a tech guy now. What has the transition been to being a tech guy?
CM: Well, I don’t really know if I feel like a tech guy. I’m still trying to get my head around ChatGPT.
That’s what I mean, that’s why I think it’s actually fairly interesting! This is your chance for introspection, like a whirlwind transformation of the Chris Miller brand.
CM: What I really loved about studying the chip industry was being able to apply economic history as a lens to understanding technology, that was what was most interesting to me. I guess my sense is that there are other facets of the tech industry, and you use some of this work yourself, but that historicizing and understanding the big, structural, historical trends is actually critical to making sense of where we are today. So that was what was most fun about the chip industry, and I feel like if the book proved anything, it’s that there’s probably scope to do more of that in different parts of the tech industry.
I agree, and like I said, I think the history is really, really interesting. I think you made a better case against centralized control and for market economies than you might have realized, but is there any aspect of your understanding of chips, and you talked about what does Moore’s Law actually mean and what does it actually mean where scale matters so much that the most complicated objects are the cheapest. Has that changed your understanding of history in a way that you didn’t expect?
CM: Yeah, that’s a really interesting question. I think it’s definitely shifted the way I think about structure and agency, which is one of the big debates historians have, is it individuals or it’s a structure. And the chip industry gives you a lot of both. You’ve got agency in these larger than life personalities, Morris Chang-
You have Bob Noyce. You said Bob Noyce multiple times, Morris Chang multiple times. Sorry, I’m excited about this point, but I’ll let you go first.
CM: Those are two great examples of people that you could really argue that personal decisions that they made and could have made otherwise shaped the industry. On the other hand, you’ve got the ubiquity of computing power as just a structural force all around us and the more I studied the industry, the more I came to realize how every facet of modern life has just been profoundly shaped by the ubiquity of computing power and so I end up, I think, being a bit more of a structuralist, just because everywhere I look, I see ones and zeros flying through the air in a way that I think we all take for granted, but would have been shocking to anyone who kind of beamed in from a century ago.
This debate is one that I’ve thought about a ton. When I started Stratechery, it’s been 10 years now, in my initial article, I was all about, “Look, people overvalue the individual, it’s all about structural forces, I’m going to write about these structural forces”.
It’s funny, because I feel in some respects, I’m with you, it’s a little bit of column A and it’s a little bit of column B. I think I’ve come to appreciate and value the importance of individuals and how it matters. There may be an aspect of we were going to get computers, we were going to get phones, someone was going to make chips. But the timing of when it happened, where it happened, and the way there’s a general debate about technology which is it still matters the way people actually use it. Look at AI, right? Is the issue about AI, or is the issue about the application of AI and the people making decisions about it? That other part is probably undervalued by the structuralists, just as the systemic aspects are undervalued by the people who focus on agency. And the reality is, it’s both. That’s been my takeaway, and it sounds like it’s been yours as well.
CM: Yeah, I think that sounds absolutely right.
It’s kind of like how someone is going to write the definitive book for this moment in time, and you’re the one that had to take the volition to do it.
CM: Yeah, I guess structure versus agency. Hopefully, a bit more agency than structure.
Well, Chris Miller, it was good to meet you a couple of weeks ago, it is great to have you on for an interview, and I do recommend the book, I think particularly for people that don’t know anything about chips — I think it’s so valuable the perspective you came in with where it’s not in the weeds because you were not someone who came out of the weeds. Highly recommended, and it is great to have you on Stratechery, and I look forward to staying in touch.
CM: Thanks for having me.
This Daily Update Interview is also available as a podcast. To receive it in your podcast player, visit Stratechery.
The Daily Update is intended for a single recipient, but occasional forwarding is totally fine! If you would like to order multiple subscriptions for your team with a group discount (minimum 5), please contact me directly.
Thanks for being a supporter, and have a great day!