Chris Miller, Author of “Chip War: The Fight for the World’s Most Critical Technology”
The microchip powers everything we do — from our iPhones, to the cars we drive, to missile guidance systems. Do we take for granted how much the world depends on such a scarce resource?
For our Season 2 finale, our host David Stiepleman is joined by Chris Miller, Professor at the Fletcher School at Tufts University and author of the Financial Times 2022 Book of the Year, “Chip War,” to learn how a group of visionaries brought this world-changing technology into existence, the incredibly delicate supply chain that holds together the most complex machinery humans have ever made, and how America’s perch as the world’s chip superpower may be in doubt.
Professor Chris Miller is the Jeane Kirkpatrick visiting fellow at the American Enterprise Institute and the Eurasia director at the Foreign Policy Research Institute. He has previously taught and studied at the Brady-Johnson Program in History and Grand Strategy at Yale, the New Economic School in Moscow on the international financial system, the Carnegie Moscow Center on Russian history, politics, and economics, the Brookings Institution on U.S. foreign policy, and at the German Marshall Fund’s Transatlantic Academy in DC.
Thank you to Professor Miller for sharing his knowledge of this fascinating part of all of our lives. We hope you enjoy the conversation.
Listen to full episode:
More from this episode:
- Chip War: The Fight for the World’s Most Critical Technology, Amazon Book
- Chris Miller About Page
- Battered by Covid, China Hits Pause on Giant Chip Spending Aimed at Rivaling US, Bloomberg, January 3, 2023
- Best Business Books, Financial Times, 2022
- TSMC homepage
- DARPA homepage
- Background on Moore’s Law
David Stiepleman: Hello. Welcome to It's Not Magic, a podcast from Sixth Street about business building that strips away the pretense and gets right to the useful stuff. I'm your host, David Stiepleman. We use this show to talk to founders and industry leaders to get them to explain in plain English what they set out to do and specifically how they do it. This is our last episode of Season Two. I know you're used to every episode being special and substantive and fun. This is no exception. Today we're talking about the most important technological development of the past 100 years.
Chris Miller: Historically, the chip industry wasn't so capital intensive. It's only in the past couple of decades where capital intensivity has grown as the manufacturing's gotten more complex. Because when it was Bob Noyce buying camera lenses from the local camera shop to make his lithography systems, that wasn't very expensive. But today, a new fab in advanced semiconductor production facility can cost $25 billion. So it's arguably the most expensive factory in human history. And so that's capital intensive.
David Stiepleman: That's Chris Miller. He wrote Chip War, which everyone's been talking about. It's a fascinating book, a page turner, and among its accolades, it's the Financial Times Book of the year. Chris Miller is a professor at the famous Fletcher School at Tufts University. He's an expert on the collapse of the Soviet Union, and now on the vital role the global production system for chips plays in modern geopolitics. We've weighed into geopolitics before on this podcast, including with Ambassador Michael McFall in Season One. You'll find this conversation as informative and interesting on the birth and amazing growth of arguably the most important supply chain in the world. And maybe you'll come away with a lesson that pushing yourself to think across subject matters and to have and understand a lot of experiences is essential to navigating a complicated world. We recorded this conversation with Professor Miller live at a Sixth Street Gathering. Don't be alarmed. It's not a shy bunch. So you'll hear some of the Q and A towards the end with folks across our business. How did we get to 15 billion individual transistors on just one of the chips in your iPhone? Until this conversation, I kind of thought it was magic, but it's not. Let's jump right in.
David Stiepleman: So we agreed that we're not going to spend the full time defining terms for people and arguing over what Moore's law is or if it's a law at all. So I'm going to try and show that you were such a good writer by summarizing very quickly, and you'll tell me if I got it right. Chips, semiconductors, integrated circuits: all the same thing.
Chris Miller: Correct.
David Stiepleman: They are in everything. They're in our iPhones. They're in missile guidance systems. They do everything. They run the world.
Chris Miller: Correct.
David Stiepleman: The more circuits, which are on - off switches, that you can pack on to a chip, the more sophisticated the instrument will be because you can put more computing power into the chip.
Chris Miller: That's right.
David Stiepleman: Developing and making chips, it's super complicated. It's always changing. And there are very few firms and it's highly interdependent. And depending on your geopolitical kind of mood, it's either an ultra-efficient, globalized division of labor or it's really bad, because there's a lot of choke points.
Chris Miller: That's right.
David Stiepleman: Fair. Okay. One of the, one of the themes in the book is, holy macro, humans are amazing and have incredible ingenuity. So I'd love it if you could just describe to us what extreme ultraviolet light is and how it gets used and what's that all about? Because I think that will unlock a couple of things.
Chris Miller: So if you go to an Apple store and buy a new iPhone, the primary semiconductor in an iPhone will have 15 billion transistors carved into it. Each one of those is a size of a coronavirus, actually slightly smaller than the size of a coronavirus. Patterning these takes the most complex machinery humans have ever made. And one of the machine tools you need to pattern 15 billion transistors on a single silicon ship is called an extreme ultraviolet lithography tool, which shoots rays of light, photons, at a wavelength of 13.5 nanometers, which is important because visual light has a wavelength of several hundred nanometers, which is far too large to carve transistors. Visual light is too big. You need small wavelength light, but 13.5 nanometer light is hard to produce. You need to have a ball of tin 13 microns wide. That's a millionth of a meter falling through a vacuum, pulverize it with one of the most powerful lasers commercially produced. It explodes into a plasma, forty times hotter than the surface of a sun. This emits light at a wavelength of 13.5 nanometers, which is then collected by a set of the flattest mirrors humans have ever made. Then it's directed via these mirrors at a silicon chip and carved 15 billion transistors into your iPhone chip.
David Stiepleman: It's hard to comprehend. It's unbelievable. Can you describe like how you take little balls of tin, you increase their speed and that's what you shoot into the laser? Like how did people figure that out?
Chris Miller: So it actually emerged from experiments in nuclear weapons. That how they first began to develop the type of UV technology. But for nuclear weapons’ technology, you need to do something once or twice or three times. For chip making, you need to do it a million times a day. Because there's lots of chips being produced and lots of transistors on each chip. And so the process of turning that science experiment into high volume manufacturing that we all rely on took thirty years. And today, only one company can do it.
David Stiepleman: And that's ASML in the Netherlands. How many parts in one of those machines, roughly?
Chris Miller: So ASML doesn't know. Several hundred thousand at least. The laser component alone in an ASML lithography machine has 457,000 components. It's produced by a German company, and so the Germans know exactly how many components are in their machines, but that's just one of the main systems. So overall, at least a million components, probably more. And all of those components have to work basically all the time, because if your mean time to failure is once a year, a machine with a million components never works.
David Stiepleman: Right. And how much does machine cost?
Chris Miller: Around $150 million a piece that takes three or four 747s to transport. They've only sold like slightly over a hundred of them at this point because there are only a couple customers in the world.
David Stiepleman: Right, so when TSMC is making, actually, fabricating, chips in Taiwan, they're getting the machine from the Netherlands. They're the only people who make this machine. Just to think about all the different steps along the way and the process of making these chips is just unbelievably incredible. And while there's obviously these risky choke points, just to think about the amount of human ingenuity that went into that is incredible. Let's take a bunch of steps back, decades back, and talk about how we started making chips and the locating that in Silicon Valley, some of it in Dallas, Texas, Texas Instruments, some of the personalities that were there. How did this industry kind of form?
Chris Miller: So if you think back to computers in the 1940s, they were the size of a large room. Computers were powered by vacuum tubes, which are a little light bulb, like devices that turn on and off, which were large, produced a lot of heat, also attracted moths and so required constant debugging, which is a big problem in early computing, still exists today in different forms. And these early computers were great for their purposes, but they couldn't be miniaturized. They were far too large. And so there was a race in the 1940s and fifties to find a way to miniaturize computing, and the transistor was the way to do that. In the late fifties at Texas Instruments and at Fairchild Semiconductor in California, two scientists began to combine multiple transistors on the same chip of silicon or germanium. And that produced the first semiconductor that we know it. And so in the early sixties, they began to commercialize these devices. The first commercially available semiconductor had four transistors on it, and then thanks to Moore's Law, we've had tremendous growth since then.
David Stiepleman: Maybe you should describe – I said we wouldn't do this but describe Moore's law quickly. And it's not a law.
Chris Miller: It's not a law, it's a prediction. Gordon Moore was one of the co-founders of Fairchild, and then later of Intel. And in 1965, he was asked to write an article for Electronics Magazine about the semiconductor industry. And he noticed that the number of transistors per chip was doubling roughly annually. And he predicted that that would happen all the way through 1975, at which point there would be 65,000 transistors per chip. And that was true. In fact, it's doubled once every year or two since then. It's all the way up to the present. We've had a doubling, an exponential growth rate. Where else in the economy do we see exponential growth rates? An example is hardly anywhere, right? Like, imagine if airplanes flew twice as fast every single year.
David Stiepleman: That's interesting to think about. I'm going to ask you at the end – I think I've heard you say you're a Moore's Law optimist. I'm going to ask you why, but let's talk about the businesses. We're here as a partner group. We're talking about our business, our cohesion as a group, how you work together. I'd love your observations because there are some stories about the Traitorous eight, is that right? Who left Shockley to form Fairchild? Then Moore and Noyce go to form Intel? Why are they moving around? Do you have any thoughts on that?
Chris Miller: In Silicon Valley, there was an extraordinary willingness to leave your founding firm, even if it was founded by a Nobel winning physicist and start something new. At this time in the 1960s and seventies, there was a sense, which was correct, that semiconductors were going to transform the world, and Moore, and noise and people around them had this vision that, and Moore wrote this in his 1965 article, that in the future there would be personal computing devices, mobile communications. And at the time, people thought he was a lunatic. But in fact, he and people around him understood the vision that they were all buying into, and they weren't going to let an individual set of managers that disagreed with them stop them. And so they repeatedly left firms, and their colleagues repeatedly left their companies, to found new startups that they thought had a better pathway to achieving this vision.
David Stiepleman: I'm really interested in this because another theme of your book, is that this generalist view kind of wins out. There's the physics – pretty important. There's the “how do you manufacture efficiently,” obviously very important. How do you manufacture it in an error- free way? And then there's this ability to see the field and understand that this is not going to be for military applications forever, and there's going to be this market that doesn't exist. Then there's, how do you price it? How do you sell it? What do you think are common characteristics of those firms? I mean, this is the golden question for us. The characteristics of those firms that get that right. That see all of those things and are able to kind of execute on that.
Chris Miller: I think if you look at the person who did that the best in early Silicon Valley, it was Bob Noyce, who founded Fairchild, and then Intel. He was an MIT-trained physicist, very brilliant in terms of his technical expertise. But what he was able to do is take the technical and envision a market that didn't yet exist. Not only that, he was able to explain it to everyone else who couldn't understand the technical, because he was smarter than all of his customers. He was smarter than all the people he wanted to sell products to. They weren't MIT physicists, but –
David Stiepleman: That doesn't always translate into being a good explainer or a good marketer.
Chris Miller: That's right. But he was uniquely skilled at communicating his vision, and he employed lots of people who were not uniquely skilled at communicating but he rose to the top because he was able to connect the business vision, the communication of that vision, with the fact that no one challenged him on his technical capabilities.
David Stiepleman: And, whether it was Fairchild and then Intel, were they doing things inside of those firms to make those conditions kind of succeed?
Chris Miller: In the 1950s and sixties, it was a bunch of 28-year-old guys who just graduated MIT. So they didn't have a lot of business strategy and Noyce himself was a bad manager. He was a visionary, but once Intel started growing into a sizable firm, he wasn't interested in management. People thought he was pretty disorganized. Gordon Moore actually provided some of the initial rigor and then Andy Grove, who had become Intel's most important CEO, was the one who actually made sure the organization ran in an efficient manner. Noyce had the ability to articulate the vision that was able to get everyone to buy in. The, the workforce, the customers, the suppliers, the funders. That set him out from Pierce.
David Stiepleman: I want to talk about the funders in a second. You mentioned Andy Grove, and I think you group him with a couple of different people in the book. Morita, who founded Sony, Morris Chang, who I think is still around and I think you did talk to him. I'd love to hear about that. I also would add to that list, Lynn Conway, who transitioned from being a man to a woman in 1968 and got kicked out of Xerox and ended up in Silicon Valley. I'm going to group those four people together as outsiders. Andy Grove was a Hungarian Jew, grew up under the Nazi regime and the Soviet regime, and had some terrible experiences, then similar experiences in Asia, for those two guys in World War II in Japan and in China, respectively. Like, is there something to that, are they outsiders who just don't – I don't want to put words in your mouth. Is there some kind of common characteristic there that made them effective visionaries?
Chris Miller: I think it's certainly the case that there were a ton of immigrants and in the founding days of Silicon Valley, two out of the eight founders of Fairchild were not born in the U.S. You look through the firms that made up early Silicon Valley, and that was definitely true. Partly, that was because there was nowhere better in the world to work than Silicon Valley at the time. It wasn't called Silicon Valley then, we're being ahistorical, it was called the Bay Area. Not until 1971 was it actually given the name Silicon Valley. So partly it was because everyone wanted to work in Silicon Valley, but I also think if you were a very bright scientist, it was a place where you could take your scientific skills, have a unique vision and deploy it into a business. And that's why people like Grove, Morris Chang ended up working in this industry. Lynn Conway is different, I think. She started her career as a computer architect at IBM, and so had a background in how to actually design computers, not chips themselves, computers. After being fired from IBM very unfairly, she found her way to Silicon Valley and was shocked to find that semiconductors were still designed by hand at that point. She’d had tens of thousands resistors designed by hand. And she said, well, computers are very efficient. Let's use computing to design semiconductors. And it set off a revolution in chip design that continues to this day.
David Stiepleman: Right. I mean, it has to be a major driver of Moore's law that you're able to do those designs you couldn't possibly do by hand.
Chris Miller: And Moore's Law begets Moore's Law.
David Stiepleman: Yeah, geometrical, you would expect that. You know a character I really liked in the book, Jack Kilby. I love that story of him. Tell the story of him showing up in the summer. No one's there in Dallas, it's hot. We have some people from Dallas here. It's hot in the summer, they tell us. What'd he do?
Chris Miller: So at the time, Texas Instruments had a policy of everyone got July off for vacation, perhaps because it's so hot. But you only got your July off if you'd passed a certain tenure. And Kilby arrived for work in Texas too soon to get his July off. So he was free in the lab and spent his summer tinkering away, trying to find ways to produce transistors that were more effective and had lower failure.
David Stiepleman: And this is the mid, late fifties, Right?
Chris Miller: This is 1958. Yeah. He had the idea of “let's put multiple transistors on the same chip,” which is not a radical idea, but no one had ever done it before. And so alone, basically in the lab, he started tinkering, and developed a, a prototype that he convinced his colleagues very quickly could transform the industry because you'd have failure rates rapidly declined when you didn't have to attach two separate devices. They were just carved into the same block of material.
David Stiepleman: It's a pretty good first month at work. It's kind of ironic, as I was thinking, I was reading it, the chips beget all this technology that kind of reduces the likelihood that you're ever going to be alone by yourself, tinkering ever, which is kind of sad. Let's talk about Intel, because Intel is a really interesting story. They were nimble at shifting from kind of being freewheeling to deciding, you know what, we've gotta actually like automate, routinize and just pump out chips. So how did they make that transition? And then what happened to them?
Chris Miller: So Intel was founded in the late sixties to make memory chips. At the time, they saw themselves as a tech-focused firm, not a manufacturing-focused firm. Their technology was quite good. But memory chips were a business that was commoditized pretty quickly in the 1970s. So their margins fell, and they faced a pressure on the market share. So by the eighties, they no longer had a really functional business, so Andy Grove, who was an extraordinarily hard-charging CEO, convinced Gordon Moore and Bob Noyce to leave the memory chip business in the mid-1980s and pivot entirely towards microprocessors, which at the time was a wild decision because the first PC had just been invented and the market was tiny for PCs. But Grove bet that this was a product that Intel could dominate, as it did. And so he transitioned the company, completely, to focus on microprocessors, which was a great decision in the 1980s, and set the company on its growth all the way up to the present where they've had a dominant share in microprocessors since then. But the problem is they were too successful.
David Stiepleman: Explain.
Chris Miller: The microprocessor business for PCs is a duopoly between AMD and Intel. Intel has historically done better than AMD and so it's had basically a guaranteed income every single year, because everyone needs to buy a new PC once every couple years, and every PC needs a processor. And so it was remarkably profitable for a very long time, but like many monopolies, it lost its focus. And over the last decade, it's had really severe issues in missing deadlines from new technologies and in missing key technological shifts. So, the smartphone. Steve Jobs went to Intel and said, will you build a, a chip for a smartphone? And Intel said, well, that seems like a little pretty low volume product. Said no. That was an error. Second, AI. Intel was behind the curve on AI, and still is in the process of trying to catch up to companies like Nvidia. And so it's a great success story of American capitalism, but it also shows the dangers of being too successful.
David Stiepleman: Can we talk about the funding models? Because we were talking about ASCM before and it's $150 million to build this machine, took them 30 years. Let's talk about the CHIPS Act also. What's the right model? Because these are incredibly capital-intensive things. They're strategically super important. Well, we also were talking about, if the government is your main customer, that's probably not a position you want to be in either. The fact that the industry grew so much because of the vision that there was going be a private consumer market was super helpful. What's the right model? What do you think the government ought to be doing, if anything? I don't wanna presume that the government should be doing anything.
Chris Miller: So historically the chip industry wasn't so capital intensive. It's only in the past couple of decades where capital intensivity has grown as the manufacturing's gotten more complex, because when it was Bob Noyce buying camera lenses from the local camera shop to make his lithography systems, that wasn't very expensive. But today, a new fab, an advanced semiconductor production facility can cost $25 billion. So it's arguably the most expensive factory in human history. And so that's capital-intensive. You've got a tiny number of companies that can afford that. They all basically benefit from either oligopolistic market that they sell to, or government backing, or both. It's almost impossible to imagine new entrants that aren't backed by governments.
David Stiepleman: You made some comment that I wanted to follow up on, and I couldn't believe I was reading this book and I was going to get to ask you this question, or any questions. Awesome. You talk about DARPA, right? The R&D arm of the Pentagon, and you make a nuance point, I think, which was, you want DARPA, pumping money into the system and trying things out and encouraging innovation and funding it, but you don't want it to have too much oversight, because what was that point? Can you elaborate on that?
Chris Miller: Well, I think there are two interesting things about DARPA. The first is that DARPA has no career employees. People come to DARPA for a five-year stint or so, from VC firms, from academia, and then leave. They've got a set amount of time to accomplish their goals. And so that lights a fire under them, and has them focused on accomplishing some sort of unique technological goal. And they've got a specific program they're trying to produce some sort of new capabilities in. And that's a really unique model. Very few organizations work that way. But second is that DARPA's got a big budget, but it's hidden inside of the world's largest budget of the Pentagon. So Congress doesn't really know what goes on in DARPA.
David Stiepleman: I like how your voice got very low there.
Chris Miller: Don't tell Congress.
David Stiepleman: Okay. We'll talk about whether this ends up in the recording.
Chris Miller: If you did a study of their success or failure rates, their success rate is not that high, because they're taking wild bets on untested technologies, but that's their point. Nowhere else in the government could you get away with that, probably justifiably so. But in the Defense Department, you can, because you can call it a security investment. As a result of that, DARPA has failed repeatedly, but had some pretty big successes as well. A lot of the interesting successes of DARPA are programs that looked like failures at first. And then 10 years down the road, somebody realized, oh, this failed project would be pretty interesting in a different use case.
David Stiepleman: It sounds like a Batman movie. But what's an example of that?
Chris Miller: If you look at an iPhone, for example, a lot of the technologies in an iPhone have their origins in DARPA programs. Now, the DARPA program manager was not thinking, let's create an iPhone, but from GPS to the displays, to the chips inside, it's hard to find a consumer electronics device that doesn't have part of its origin in a DARPA program.
David Stiepleman: Got it. Russian history, Russian intellectual history, economic history. Is this a departure for you, or did this, was this an outgrowth? This project on working on chips.
Chris Miller: I started planning for the book on the Cold War Arms race. I wanted to understand what explains why the Soviet Union and the U.S. were both able to produce the key military technologies of the early Cold War – nuclear weapons and long range delivery systems. But by the end of the Cold War, the Soviets weren't able to keep up. Which sort of seemed puzzling actually. Smart physicists. Well, they had those, lots of capital investment, they had that. Good education system, massive defense budget. What went wrong? It became clear as you start looking through defense technologies, that actually the key transformative component in all military systems over the past half century has been computing power. It's not a coincidence that semiconductors emerged for missile guidance systems during the Cold War.
David Stiepleman: You talk about the Soviet efforts to steal this technology, and they were remarkably successful at reconstructing things. But why didn't that help?
Chris Miller: Well, the problem in the chip industry is that there's two problems. One is that if you buy a cake, you don't necessarily know how to bake it. Same thing if you steal a chip, you don't necessarily know how to make the chip. The second is that even if you find out how to make the chip five years later, Moore's law has raced ahead and you're far behind. The Soviet Union was focused on copying from day one. They actually had, students in physics at Stanford in the late 1950s and early 1960s, studying with Bill Shockley, who just won the Nobel Prize for inventing the transistor. And several of these students came back to the Soviet Union, handed chips that they had pocketed to the Soviet radio electronics minister who was in charge of the electronics industry. At one point, he handed it to back them and said, copy it. And that was the origin of the Soviet chip industry. Copying seemed like a good strategy because it worked really well with the atomic bomb.
David Stiepleman Yeah.
Chris Miller: But it worked very, very badly with computing.
David Stiepleman: You're just always going to be behind. Did anyone try to innovate? Do you have any knowledge of that inside the Soviet Union?
Chris Miller: Certainly, people were trying to innovate, but the structure of their system – super secretive, no consumer market, meant that incentives just weren't there. The person who devised the first integrated circuit of the Soviet Union was not known until after the Cold War ended, because it was done in a military lab.
David Stiepleman: Oh, interesting. It raises the question of your process because my perception is that as an academic who's looking at governments and you're looking at government records, they're either well-kept or they're not well-kept, but they're public, they get released. Very different doing research. I should ask this as a question. Is it very different doing research on companies and these personalities and records that are kind of self-serving, they're not necessarily organizing. Like what was the challenge of that?
Chris Miller: Well, government records can be self-serving too.
David Stiepleman: Yeah, for sure.
Chris Miller: Some companies have great archives, like Texas Instruments has an extraordinary archive at Southern Methodist University. You can go into and look at what Morris Chang was writing about in the 1970s. Others do not, and so much of the research for this book was interviews of people who were in the industry and whether scientists, CEOs, government officials, were engaging it. That let me understand that not only the technologies changed, but also the personalities in the businesses, which is great fun. So Morris Chang, for example, someone I had a chance to speak with. It's one thing to talk to his colleagues and read about his accomplishments, but actually hearing his narration of how it was that he rose up the ranks at TI, and what he was doing on the production lines of the early transistors and semiconductors, gives you a whole different flavor for his success, for example.
David Stiepleman: His going to Taiwan, right. You should tell that story. And what was his take on why they passed him over at TI?
Chris Miller: His background, maybe. Morris Chang, born in mainland China, spent his childhood fleeing to Japanese armies during World War II. His father was a banker and a nationalist government official. Uh, so they fled after the communists took power in ’49. He enrolled at Harvard in 1950, the only Chinese student in his class, sort of an extraordinary data point in how America has changed since then. He wanted to study Shakespeare and loved English literature until his uncle told him to do something useful with his life. So he transferred to MIT, which ended up being good. So after MIT, thankfully for his uncle, we've got transistors with high capabilities because he focused on something besides Shakespeare. So he got a job at Texas Instruments at a time when the industry was just taking off, and he had a reputation for a unique capability to have the right intuition for how to improve manufacturing processes.
Chris Miller: He was a manufacturing guy fundamentally, and he spent his career in manufacturing, and he rose up the ranks at TI and was passed over in the 1980s for the CEO job. Now, there's two explanations why. He didn't tell me which he believes. One story is that Texas Instruments was a company in Texas. There weren't a lot of Chinese people in Texas either. And that it was seemed to be unacceptable to have a Chinese CEO. That's one possibility. Two is that there were some personnel struggles, personality disputes. And so that's the explanation. Maybe both are true, hard to know.
David Stiepleman: How would you weight them? Do you have a view?
Chris Miller: I asked Morris Chang what it was like to move from Cambridge, Massachusetts to Texas, with probably some preconceived notions as someone who lives right outside of Cambridge, Massachusetts. And he said that he found Texas much more, more welcoming than Cambridge. So he had nothing but good things to say about the people of Texas.
David Stiepleman: Got it. We have a couple of Texans here who are very happy to hear you say that. There you go. What was it like to talk with him?
Chris Miller: He's over 90 right now. He's ostensibly retired, but still does interviews from his office at TSMC, so not really retired. He's one of the most powerful people in Taiwan because TSMC makes up over a third of Taiwan's exports. It's the most valuable publicly traded company in Asia. So if there's a tycoon in Taiwan, it's him. And he's regularly interviewed now about his views on China and risk but actually hardly ever asked about his career in the 1950s. He was very generous with his time in talking about his early career, and explaining how it is that he got involved in this to begin with. You know, I think his career is fascinating because there's no one whose trajectory tracks the chip industry from the earliest days up to the present. Not only in terms of the way the technology has changed – he was there on the production lines when TI was basically making transistors by hand, all the way up to the present where his facilities are producing a quintillion chips a year for Apple's iPhones. But also the geography he maps on to as well, from China, lived in the U.S., moved to Taiwan, which is exactly what the chip industry has done.
David Stiepleman: That's incredible. Any other good stories in terms of who you interviewed? Who did you like spending time with?
Chris Miller: The other one that stood out was Carver Mead, the Caltech chemist, who also was present at the creation of the chip industry. He has a story of Gordon Moore walking into his office before they'd had a chance to meet in the late 1950s and pulling a sock out of his briefcase with a bunch of early transistors in it, and giving it to Carver Mead to play with in his electronics classes.
Chris Miller: He was as someone as well who'd been there from the early days of the industry all the way up to the present, played a huge role in transistor design in the early stages, but then also devised a lot of the neural network thinking that eventually led to the use of transistors and semiconductors for AI today.
David Stiepleman: Got it. You referenced the more recent geopolitical tensions, right? This idea of responsible stakeholder theory. Is that dead? Should we not count on that? Should we not count on trade spreading knowledge and friendship? And are you a pessimist? How should we think about that?
Chris Miller: Well, I think history does not suggest that trade produces peace. It would be nice if it did, but it doesn't. And it's not just the history of the last twelve months, with Russia and Germany finding that their gas relationship did not produce peace. It’s hard to find good evidence historically that trade produces peace. Britain and Germany before World War I, there's a long list of countries that have been deeply integrated in terms of trade investment, nevertheless gone to war at extraordinary economic cost to both parties. So I'd like it to be the case that integration guarantees peace, but I don't see the evidence for it, and I don't think hope is evidence. As a result, I think there's reason to be worried that in fact, the situation is spiraling close to being to the point where's out of control. You look at the calendar the next couple of years, U.S. presidential election in 2024, Taiwanese presidential election in 2024. There's a lot of trigger points that could lead to a small crisis, and a small crisis could very easily lead to a big crisis.
David Stiepleman: In terms of bringing chip manufacturing back on shore, in terms of releasing these choke points, what should we be looking at to see whether or not our policies are working in that regard?
Chris Miller: I think if you ask the U.S. government off the record, they will tell you that the goal of the CHIPS Act is as an insurance policy in case there's a war in the Taiwan straits. And then the key question is how much chip-making capacity do you have outside of Taiwan and outside of China? We're going to get more chip-making in the US than we otherwise would have, thanks to the CHIPS Act. But also the Japanese, the Indians, the Europeans, the Koreans, a lot of countries are pouring money into semiconductors right now. So we're going to have a less Taiwan-focused semiconductor supply chain over the next decade. Not dramatically less, but meaningfully less. It's not clear we're going to have a less China-focused micro supply chain because China's spending probably as much money as all the other countries I mentioned combined, building out its own industry. So that creates a very interesting dynamic where you have risk of overcapacity because everyone is trying to pry themselves with an insurance policy.
David Stiepleman: That's interesting. I mean, we also block the sale of components to companies that would then manufacture in China. What are you looking at to see when that's working? I saw some announcement last week on Bloomberg that China announced that they were reducing funding to chip research, which I think was meant with a lot of skepticism.
Chris Miller: Yeah. That was one Bloomberg headline that seemed to contradict the last decade of Bloomberg headlines. So I don't read too much into that. I think the U.S. government is focused on controlling the transfer of advanced GPU chips, chips that you use for training AI systems and data centers to China, on the grounds that you can't train AI today without access to advanced data centers. Advanced data centers need these chips and these chips are all produced in Taiwan, so the U.S. has a choke hold over them. The goal that Jake Sullivan, the National Security advisor, very clearly and explicitly articulated in a speech last September, is to stop China from developing as advanced AI systems as the U.S. and its allies for civilian or military purposes.
David Stiepleman: Got it. Let's talk about the future a little bit. You're a Moore's law optimist. What does that mean?
Chris Miller: There's a lot of people who look at the trajectory that companies like TSMC have sketched out and they've got a five-to-seven-year roadmap as to how they're going to keep shrinking transistors smaller and smaller. And they say, after five or seven years, no one knows what's next. And that's true. No one knows what the next innovation is going to be that lets you shrink even further. But that's always been the case. Going back to the 1980s, you can find various esteemed computer experts saying Moore's law is almost dead. Gordon Moore himself in the 2000s said he couldn't imagine how it could be continued. Carver Mead said something similar around the same time. And it's always continued. So I am a Morris Law optimist. I think the amount of money going into R&D, the amount of different techniques we have to keep, it's not just shrinking transistors.
Chris Miller: We can stack transistors, we can reorganize chips to provide more floor space for transistors. There's lots of different techniques at play. And so I'm just really skeptical that given the focus, and given the track record, we're going to face a limited computing power. Now, there's one key caveat, which is that the proper articulation of Moore's Law, as he put it in 1965, was that not only would the number of transistors double every year or two, but the average cost per transistor would decline. In other words, we get a free lunch of computing power every year or two. And that historically was true. That relationship broke down about five years ago. So now the cost per transistor is not declining, and by some estimates it's even rising slightly as it gets more complex to layer transistors on top of each other. That's a big shift. We get more computing capabilities, but we finally have to pay for them for the first time, basically ever.
David Stiepleman: Got it. You made an interesting prediction in the book, or an observation about where the most sophisticated or higher end computing power innovations are going versus lesser ones. And you call it the fast lane slow lane. I think has a lot of implications for where money goes, so can you talk about that?
Chris Miller: So over the past couple of decades, advances in microprocessors that go in PCs or data centers have been so substantial that it's hasn't made sense to devise specific chips for specific applications when it comes to basic computing. It was always better just to wait for the next Intel microprocessor. It'd be two times as good and that would provide you the performance you needed. One of the ways you can get around the fact that it's harder to eek out performance gains that Moore's law needs, is that you can devise specific architectures for chips that solve specific problems. So for example, Amazon is designing its own chips that are specifically designed for its own workloads in its data centers. Google does the same thing. Facebook. We're going to see more companies doing that. They're going to say, we're willing to pay more to design chips that will give us more performance than a general purpose microprocessor. And for companies that are willing to pay more, they're going to get performance out of it. And for those who aren't willing to pay more, are going to get slightly less impressive chips. But I think that the trend is going to be more companies being willing to pay more. Right now, it's just the big cloud computing firms, but it's going to be auto firms, it's going to be industrial firms down the road because the gains to specialize in your chip design will be substantial.
David Stiepleman: I hesitate to ask this in front of Marty and Adam, but I'm going to. What's quantum computing and does that obviate the need to talk about chips anymore?
Chris Miller: Look, I think quantum computing is going to be a big deal when it materializes. I think we're probably some way away from the first commercially viable use case of quantum computing. The history of classical computing suggests that there's going to be a long process of integration of quantum computing into economy and society. So, I don't think we need to really think that hard about quantum computing disrupting classical computing for at least the next decade and possibly longer.
David Stiepleman: Okay. I'm not going to worry about it then. At least for now.
Chris Miller: I think that’s safe.
David Stiepleman: Did you expect the popular upswell of attention to this book? You were hopeful, I'm sure.
Chris Miller: I was surprised. I mean, I was writing this book before the semiconductor shortage, which really put semiconductors on the map for most people, and then before the recent regulations that the Biden administration imposed on transferring chips to China, which has been a big deal in terms of getting semiconductors discussed in Washington and elsewhere.
David Stiepleman: You think the government was behind? I think you have that view in your book that they kind of didn't get it.
Chris Miller: Yeah. I think that's changed the last five years. I've spoken with people in the government fifteen years ago who said the U.S. is falling behind Taiwan and South Korea and there was a view in Washington that, well, our tech companies are big, and globalization is simple, and it doesn't really matter. And I think if there's peace in Asia, I think that view is correct. But I guess I'm more worried than people were ten years ago.
David Stiepleman: What's the gadget weapon thing that blows your mind when you think about the power of computing and what we're using it to possibly develop?
Chris Miller: I think the device that people don't think enough about is data centers. Because data centers are where advanced computing happens and they're increasingly where productivity improvements are going to be created. We're going to have smarter algorithms that are trained and honed in data centers. And more than that, they're increasingly where new products are going to be invented, in data centers. And so today we think of data centers as somewhere where stores are data or it's the cloud, it's out there somewhere. But actually data centers are, I think, at the core of where the economy's going. As a result, the chips that go in data centers will be more important in ten or twenty years’ time than they are today. I sort of think of data centers as the factories of the 21st century and the chips are the machine tools that are hammering out new products for us.
David Stiepleman: I want to open it up to questions if people have them, and if you're willing to do that.
Speaker 3: So by some accounts, China hasn't gotten much out of its vast investment in semiconductor technology or building their own capability. Why? And do you think that will continue or will change?
Chris Miller: So, I think first off, assessing what China's gotten, I think depends on whether you're assessing commercial viability or not commercial viability. In terms of successful companies in the chip industry, China has hardly any. And those that have been successful have only been successful thanks to really dramatic amounts of government subsidies. Why? I think the answer is because it's really hard. The trend in the chip industry has not been of catch up but of falling behind. There used to be more firms that could produce the cutting edge, and now there are fewer than there were 10 years ago. Because the cutting edge races ahead and, and it's harder to catch up. So when people say, why hasn't China caught up? I say, well, well why has the U.S., Japan and Europe fallen behind? I think that's the question we ought to be asking. So I'm not really that surprised in some ways that China hasn't caught up. But I do think that if you look at the challenges China faces now in terms of creating market-leading firms, the problem is that the way that's been done historically is by deeply integrating into supply chain. So TSMC, extraordinary capabilities, but they're made possible by buying the best components from the Netherlands, from the U.S. and by selling to the most impressive customers and learning across the supply chain. And that's worked great for TSMC, Samsung, similar story. For Chinese firms, that's very difficult to do because they're increasingly restricted from that.
David Stiepleman: Learning across the supply chain. In other words, understanding how to put stuff together from different places, but also seeing what designs people want.
Chris Miller: Yeah, exactly.
Speaker 4: This is veering into opinion territory, but if all of these other countries are taking out insurance policies and sort of devaluing Taiwan's own insurance policy, do you have thoughts on implications, other things that they or other countries could, should be doing?
Chris Miller: Well, it's an interesting question for Taiwan, because right now TSMC is the most important business by far in Taiwan. The semiconductor industry in general is Taiwan's most important sector. And so the Taiwanese are deeply worried that the U.S. and Japan and others are hollowing out their industry. You know, there's actually a little bit of truth to that in that we're directly subsidizing TSMC to build facilities here rather than there. I think it's a really complex balancing act that's got to be struck. On the one hand, the U.S. is defending Taiwan, putting the lives of Americans on the line to defend Taiwan. So the U.S., I think, is implicitly telling Taiwan, you gotta play ball. We're putting something on the line for you. You need to help us, in case something goes wrong. But it's a very complex and unpleasant discussion to be had. And you know, if you talk to U.S. officials on the record, they won't even mention Taiwan when it comes to semiconductors because it's too sensitive. Off the record, of course, everyone understands that the CHIPS Act is an insurance policy.
David Stiepleman: And of course, TSMC is building a big fab in Arizona.
Chris Miller: In Arizona, yeah. As, as well as one and possibly two in Japan and probably one in Germany down the road.
David Stiepleman: Got It.
Speaker 5: Just curious, do you think the CHIPS Act goes far enough? And if not, where is it deficient? And then what about Asia, Europe, are they thinking in the same terms that we are about having to defend this technology?
Chris Miller: Yeah. My sort of simple math, the CHIPS Act is, you know, take your probability of, blockade or war in the next 10 years and that'll give you your estimate of how much you have to spend on the CHIPS Act. It seems to me that the $52 billion have been allocated, assumes a pretty low probability of war. Who knows what the right number is? We could debate it. I probably would take out a larger insurance policy. I think other countries are doing the exact same thing. Europe's going to spend 40 billion euros or so on its own CHIPS Act. Japan is having TSMC invest in Japan. India is trying to make a big play to become a player in the low end of the chip-making industry. So there's lots of other countries that are undertaking similar policies precisely with this diversification insurance policy mindset.
Speaker 6: What lessons can be drawn to chips that you have learned from all the work and study you've done on the Soviet Union in the Cold War? When you take the pattern of things that happen and apply it to here, how do you think about that?
Chris Miller: I think the key thing that I learned while doing the research is that I thought technological advances were about improving the technology, and that's sort of true, but it's actually about finding a market that's going to allow the technology to be improved. Bob Noyce, that's what he did really, really well. He found the market that made it possible to invest in this over many decades. Had he not found the market it wouldn't have been invested in. I think it's very impressive that he was one of the two people to invent the integrated circuit, but that on its own did not give us an iPhone. And to me it's the matching technology to market that is underestimated. And that's what the Soviet Union did a horrible, horrible job at it. They had great physicists, great theoretical physics, great experimental physics, applied physics across the board. But unless you've got a use case and a funding model, you can only go so far.
David Stiepleman: Yeah, I mean we did a little of this before, I just want to press on it. Is there a characteristic of Bob Noyce or a characteristic of the firm that he built that makes it more likely that you will be able to see that field?
Chris Miller: I'm struck by his, his communication ability to explain what it was he was doing. There's a great anecdote of him, I think he was at his 40th birthday. They're on a party bus out with his parents. He held up a chip and this was at a time when no one knew what a chip was and said, this will transform the world. But he was able to convince people that in fact it was something truly revolutionary. And it's sort of like, maybe this is a bad analogy, but sort of like Elon Musk and electric vehicles. People thought they were not really going to happen. They weren't really cool. And then he's found the market for them. Yeah. Maybe that's a bad analogy. I don't know how you guys feel about Elon Musk, but I think there's something to that communication ability that is critical, and it's not just the technical capabilities.
David Stiepleman: If the analogy works, it's fine. Are you going to let your students use ChatGPT? Are you going to forbid it or are you going to, put guardrails around it?
Chris Miller: I don't think I want students writing papers on ChatGPT, but I'm not someone who's afraid of it. I think it's great. I use when my Gmail suggests how to end sentences I use that, we all do. Right?
David Stiepleman: No, I don't. I refuse to. I have everything I'm representing that's composed by me. It has to be composed by me. I don't want to misrepresent.
Chris Miller: I trust Google to finish most of my sentences.
David Stiepleman: Well, thanks. I mean, it's a great book. We really appreciate your time and this has been super interesting. So we appreciate you being here. Thank you.
David Stiepleman: That was professor and bestselling author Chris Miller. We spoke with him on January 31st, 2023 when he joined us live to speak with senior leaders of Sixth Street. The history of the chip industry is full of mind-bending technological solutions, and it's one of those stories that makes me marvel at what humans can do together. But honestly, my biggest takeaway was that the generalist, the broad thinker, the multidisciplinary perspective that makes all the difference. As Chris Miller told us, the physics was a necessary condition for what the chip has meant for the modern world. But so is a free-market economy and marketing vision. The ability to communicate, the willingness to tinker and the ability to be a decent manager of humans. The practice and mindset of intellectual openness to follow paths to wherever they lead means you have a shot at navigating hard questions of when do you copy and when do you innovate? How do you know when to pivot? What will the use cases that you can't necessarily see? And who are the right people for the key seats at a given time? The story of chips that Chris helped us think about is the story of a broad knowledge base equipping us to see patterns and make better decisions. So read books and make sure Chip War is one of them. Thank you again to Chris Miller for joining us at our event and for being a part of this podcast
David Stiepleman: You've been listening to It's Not Magic, a Sixth Street podcast. You can read more about our guests on sixthstreet.com and subscribe wherever you get your podcast. If you enjoy today's podcast, please share it and follow at @Sixth Street News on Twitter. For more on the show in our firm, thanks to Sixth Street's production team, Patrick Clifford and Ritvi Shah, putting this together with sound engineering by Steven Colon. Our theme song is It's Not Magic, an Original Creation by Patrick Dyer Wolf. Once again, I'm David Stiepleman. Thanks for listening.
The views expressed in this podcast are not necessarily those of Sixth Street and Sixth Street is not providing any investing, financial, economic, legal, accounting, or tax advice or recommendations in this podcast. Please see additional disclosures on our website for more details.