Into the Impossible With Brian Keating - New COLD WAR Over Computer Chips? w/ Chris Miller (#391)

Episode Date: February 2, 2024

Please join my mailing list here 👉 https://briankeating.com/list to win a meteorite 💥 If there's one key factor securing America's economic prosperity and military superiority, it's not oil; it's chips. No, not the chips we all love to snack on during movies, but highly efficient computer chips. Chips that power pretty much everything from our military machinery to our iPhones. The United States held the top spot in crafting the fastest chips on Earth for a long time. But countries like Taiwan and China are catching up fast. This unfolding race that will determine our future has been thoroughly chronicled by today's guest, Chris Miller, in his book Chip War.  Wanna know what he found out during his research? Then tune in to this exciting episode of Into the Impossible!  Key Takeaways: 00:00:00 Intro 00:00:58 Judging a book by its cover  00:03:36 Moore’s law and the history of semiconductors 00:11:28 The challenges of developing chips 00:17:50 Communism vs. capitalism  00:23:48 The importance of military funding  00:25:52 Chip manufacturing and geopolitics  00:36:16 China’s demographic and economic challenges  00:41:21 Bitcoin mining and energy efficiency  00:48:02 Outro — Additional resources: 👉 Are you hungry for knowledge? Visit https://www.brilliant.org/DrBrianKeating and try everything Brilliant has to offer —free—for a full 30 days. The first 200 of you will get 20% off Brilliant's annual premium subscription! 🤓 📝 Get one month of Snipd Premium for free with this link: https://get.snipd.com/Cx7S/brianSnipd Snipd lets you take Smart Notes 🧠 with AI 💡 — it’s my favorite podcast player 😀 ! 📢 Ownership of your health starts with AG1. Try AG1 and get a FREE 1-year supply of Vitamin D3K2 and 5 FREE AG1 Travel Packs with your first purchase 👉 https://drinkag1.com/impossible ➡️ Check out Chris Miller: 📚 Chip War by Chris Miller: https://a.co/d/5hh45IW  💻 Website: https://www.christophermiller.net/ ➡️ Follow me on your favorite platforms: ✖️ Twitter: https://twitter.com/DrBrianKeating 🔔 YouTube: https://www.youtube.com/DrBrianKeating?sub_confirmation=1 📝 Join my mailing list: https://briankeating.com/mailing_list ✍️ Check out my blog: https://briankeating.com/blog.php 🎙️ Follow my podcast: https://briankeating.com/podcast Into the Impossible with Brian Keating is a podcast dedicated to all those who want to explore the universe within and beyond the known. Make sure to subscribe so you never miss an episode! Learn more about your ad choices. Visit megaphone.fm/adchoices

There's one key factor securing American economic prosperity and military superiority. It's not oil. It's not even weapons. It's chips. Chips power everything from our military machinery to our smartphones that are glued to our hands. Until recently, the U.S. had dominance in crafting the fastest chips in the world. But countries like Taiwan and more importantly China are catching up fast. This unfolding race will determine our future. And it's been thoroughly chronicled by today's exceptionally interesting guest, Chris Miller, and his book aptly titled Chip War. Chris is a professor of international history with a focus on the Cold War. He's devoted much of his life to the study. Let's get into it. Any sufficiently advanced technology is indistinguishable from magic.

Open the pod bay doors, hell. Everybody out there, welcome to a very, very special episode of the Into the Impossible podcast requested by my brother-in-law because he loves this book. Uh, by Chris Miller called Chip War. So we're going to go to war today. We're going to talk about this wonderful, really incredible runaway bestseller of 2023 by Chris Miller. Or maybe it came out in 2022. Did it come out in 22 or 23? That's right. Late 2022. Late 2022. Okay. It's been a year of just accolades and encomia from all over the world. And we're going to get into deep, deep dive into the technology into the physics. You know, Chris, I might be the first person that you've ever talked to on a podcast that's actually done lithography that's made actual semiconductor and superconduct.

These are superconducting detectors. We used in my bicep experiment. This is a lithograph proto-circuit for an antenna that we used in the polar bear two experiment, which is ongoing in the out-a-dicombed desert. We're going to talk about diffraction lithography, some of the cool stuff being done here in San Diego and in Europe. But first, we're going to do what you're never ever, under any circumstance supposed to do, which is to judge a book by its cover, especially one as delightful as this. Now, we're going to hold up a fake book. I'm going to hold up my iPhone that has, you know, I listen to the book on audiobook form. Chris, tell us what was the genesis, the origin of the wonderful cover and the title, especially the subtitle of your fabulous new book?

The title, Chipp War emerged. actually, it wasn't my first choice for title that emerged after discussions with my editor, who was brilliant, I think, at encapsulating a book's argument. I was initially actually planning to call the book The Switch, with a reference to transistors, turning circuits on and off. But my editor said, no, that doesn't get the thesis of the book, and the thesis is about the contest for controlling this technology. And indeed, one of the themes that emerged from the research was the role of the defense industrial base

in pushing forward semiconductor technology. And so that's how we settled on ship war, referencing the military origins, but also the battle between companies and countries for this technology. And so the cover, at least the U.S. cover, the international edition is all of different covers. But the U.S. cover has an American flag juxtaposed next to a semiconductor, which speaks to the fact that this is technology that everyone uses, but it's technology that governments have played a major role in shaping its development. And the international copying that I saw of a globe or a map or something like that.

That's right. Yeah, it's a picture of a globe with kind of circuitry carved into it. Yes. And this was very reminiscent. That's why I brought my own circuit board here. Reminiscent of what we do in the laboratory and how we use these sensors. And I want to take us back to maybe the OG of all things. Digital and switches, as you mentioned, transistors. and that is, of course, none of than Gordon Moore, who passed away not too long ago, sadly, because he was an incredible supporter, not just of applied science, but of basic science, pure research. I want to get into that with you. But I want to get your reaction to this quote from Gordon Moore. He said, allegedly he said, if the automobile industry had set a pace similar to that of the semiconductor industry, a Rolls-Royce would get a half a million miles per gallon, and it would be changed. cheaper to throw it away than to park it. Is that still true, Chris? Are we still progressing at a rate

anticipated by Moore's law, even if the exponent has changed or the base period has changed? Pretty much. There's different ways you can define and assess Moore's law, but the basic intuition that we're getting dramatically more computing power every couple of years at dramatically reduced cost holds true, which is why you can go on Amazon and buy a thumb drive with a billion transistors for not that much money. That would have been unthinkable not just 50 years ago, but 30 years ago, 20 years ago, that was a mind-bogglingly large number. And that still basically holds, which is why everyone is racing to get their hands on the next generation in Vivida chips to train bigger and bigger AI systems is because the next generation is better and better at a really

extraordinary rate. And I think if you think that the rest of the economy, the rest of society, the rest of technology, nothing else comes close to that system. sustained rate of improvement over time. The only thing that really compares is the cost and speed of gene sequencing, which has fallen really dramatically the past couple of decades. But Moore's law has been in operation now for over half a century, and it's really unparalleled in all of human history. Hey, Thala Wanderers into the Impossible. I hope this deep dive into the semiconductor world is as fascinating to you as it was to me. But there's an important conversation to be had that goes deep beyond this, and I want you to be a part of it. So please,

do me the favor of subscribing to the channel and or following it if you're listening on audio platforms. And for extra credit homework, give me a thumbs up, review, like, comment, share it. It really helps with the algorithm that these computer chips are running in the background to really boost our mission to connect millions of minds around the world of STEM, science, technology, engineering, and math. And by doing so, you'll help me help you get the greatest possible guest on Into the Impossible. Now back to the episode of Chris. Yeah, and the connection between gene sequencing and also the microchip and semiconductor revolution is really highlighted as well, especially here in San Diego. We have Illumina. I just interviewed Craig Bentner, who is a venter, who's the one of the, I would say he did sequence the human genome, as team did it and kind of shared the credit with Francis Collins and their team at NIH. But we had a raucous interview live in person.

here and he talked about, you know, the foundational assets of having computing power, you know, that was really unrival and how that dominated, allowed them to dominate. I want to take us to the kind of mythology, and some of it mentioned in the book. And I didn't know if he, I don't know if he know this, but here in San Diego, we have one of the founding fathers of the transistor revolution. In fact, the Nobel Prize winner, who, you know, shared the Nobel Prize for the invention of the transistor. We don't have his body, but we apparently have some of his, I'll say, genetic material. here in what's called the genius factory or perhaps the repository for germinal choice, also known as the Nobel Prize sperm bank. And Shockley is a character in this book, obviously. Talk about him.

I find him very fascinating. Of course, he was a genius inventor, although he may not have been as bright. Allegedly he kind of failed this determined study of IQ by Nobel laureates. But tell me about Shockley in his role. And this mythology that we have of putting these, you know, kind of of contributors to the technology revolution up on a pedestal, whether financially or intellectually. Well, Shaphth, certainly was an important figure in the development of some of the theory, as well as some of the initial inventions that led to the first transistor and then the ability to produce them at larger scale. But he's also, I think, a great case study in how people are focused on getting attention and taking credit for inventions, they can write the history

in a way that gives them an outsized role. And so he was actually one of three scientists, that devise the first transistor at Bell Labs. He actually didn't play the major role in devising it. He played the major role in theorizing it. But there's a very famous photo that was taken right when the first transistor was being announced. And he put himself at the center of the photo looking through the microscope and had his two colleagues standing behind him as though they were his lab aides. And so he's gotten the lion's share of the credit for the invention. But I think what's interesting about Shockley's, and although the history is generally written with him at the center,

In fact, he was a failed business person. He tried to mass-produced transistors, couldn't do it. And it was people like Gordon Moore who turned the transistor from an interesting science experiment into a mass-market product. And so if you want to look at the technology trend, it's actually less about the scientist who invented it and more about the person, the people who learned how to mass-produce them. Here at UC San Diego, we had two graduates of our university that went on to found. incredible business here called Seimer, and that's Robert Aiken's and Richard Sandstrom. Talk about, and now they're part of ASML, and you talk a lot about ASML in the book and how crucial and really monopolistic and hopefully benevolently monopolistically so.

Talk about AMSL and what is so fascinating to you as a, you know, not a physicist, but as a writer, as a technologist. talk about what fascinates you so much about the technology used in this exquisite production of semiconductors that drive the modern world. ASML is the company that makes the tools that make chips or makes some of the tools that you need to make a cutting edge chip. And so its job is to produce these machines that can manipulate materials at almost the atomic level, do so at a cost and at volume that is financially viable. That's actually the hard part. The hard part is not the physics. That's very, very, very, very hard.

Harper is doing the physics in a way that is viable to mass production. And so ASML has been involved in the lithography industry for now around half a century. But in the early 1990s, when this new type of lithography called EUV, Extreme Ultraviolet, was being invented. ASML was the only company willing to bet that you could commercialize this technology and turn it into a mass production product. And they spent three decades, billions of dollars, took up. on huge financial and technical risk to produce these machines.

And today, they're the most complex machines humans have ever made. They've got inside of them the flattest mirrors ever made, one of the strongest lasers ever deployed in a commercial device, and an explosion happening inside of the chamber that Seimer developed at a temperature of 40 or 50 times the temperature of the surface of the sun. And this isn't a one-off. This is a machine that works 24 hours a day, seven days a week, 365 days a year, and it has to work because each one of them costs $150 million a piece.

So it's an extraordinary combination of, on the one hand, physics and engineering, the other hand, making it match the business case that's necessary to produce them by the dozens and eventually by the hundreds. And talk about the challenges of scaling up. I mean, you talk about the feature size on these exquisitely manufactured devices. You know, this one here you could do literally with optical light and then so forth. They don't need extreme ultraviolet, advanced diffraction gratings and so forth. But talk about limitations.

The size scale is approaching fractions of a coronavirus for scale. I love that that's become a figure of scale nowadays. Tell me, what are some of the challenges of going beyond these already literally microscopic and beyond microscopic levels? I think one of the things that surprised me was that to make smaller and smaller devices. And as you say, the transistors in the chips on our phones and in PCs are now virus-sized. You need bigger and bigger and bigger and more complex supply chains. So when the chip industry was founded, companies would do everything themselves.

They'd make their own production machines. They'd refine their own chemicals. They'd produce their own silicon wafers. They'd try to do it all. Whereas today, every step is so mind-bogglingly complex. and the level of specificity and purification and precision is so high that there's not a single company that can do anything on their own. And so the number of firms, the number of countries involved in making a chip inside your smartphone now measures vastly larger than it would have in the past. And just the machines themselves take a cutting-edge lithography machine.

They require multiple airplanes to move because they're so large. And then the next generation lithography machine that will make even smaller circuitry will be substantially larger. Because to make smaller and smaller devices, you need bigger and bigger machines. And a quote that I also want to get your reaction to goes like this. What Gordon Giveth, Bill taketh away. That refers to Gordon Moore, Gordon Moore's law, but also Bill Gates, with Bill being a proxy for the software, you know, the kind of voracious appetite that software has to chew up all these computing resources. In fact, we use in our cosmic data analysis, we use Department of Energy machines, that are, you know, benchmark top five in the world or maybe even better.

They're used primarily for nuclear fusion simulations and research in nuclear weapons, but they give us some spare cycles. We found that, at least I found, and informally talking to my colleagues, Moore's Law might keep increasing, but what I really care about is how many papers, you know, come out of it, right? I don't really care how many flops are done. I want to know, how did you distill this, you know, this technology into floating point operations that then do analysis on the district.

of dark matter in the universe, for example. That kind of benchmark, you know, the ultimate, what I do is to convert money, you know, into papers. You know, some funding agency gives me a grant and we work and we get a paper. Maybe we get 10 papers out for every $10 million. It's about a million dollars to pay. Anyway, long-winded way of asking you, is it not true that the better the hardware gets, the software developers take advantage of it or maybe don't and don't optimize?

And so therefore, Moore's law effect, the effective Moore's law for output, which is what you care about scientific discovery, gene discovery, all sorts of protein folded. It's actually saturating and not growing anywhere near the power law exponent that Moore's law is. So talk about that, the bill taking away from what Gordon gave us. Well, you know, I think it's certainly true that we've got an almost limitless appetite for compute. And the more compute we get, the less efficient we have to be with it. The less compute costs, the more incentive we have to be inefficient because we can spend our time on other things. And so it's absolutely right that in times when Moore's law is improving at a rapid pace,

software engineers try less hard to be efficient. I would not agree with the thesis that we're getting less paying for our buck out of Moore's law. I think, you know, in certain spheres you can make that argument. But I guess when I look at the explosion of interest in applications of AI over just the past couple of years, it seems to me that there's a lot of pretty fantastic products coming down the pipeline. And to me, that's a sign that we're finding new applications for Moore's Law. And it's, it's only because of Moore's Law that you're able to train big AI systems because they're just so data intensive and therefore just so compute intensive to train. And so there's a big debate

constantly underway in the chip industry. Is Moore's Law finally dying? What does it mean? How would you define that? But I guess when I look across the use cases for compute, which today, the biggest growth driver is AI and autonomous systems. It seems to me there's a whole lot of excitement for the next generation of products and concepts that are being developed. Yeah, I think that certainly is true. And I guess, you know, the next question is, you know, is the next war, maybe not a chip war, but a qubit war? What are your thoughts there? Are we going to witness, you know, kind of really truly exponential growth, even of the exponent in Morris Law, not just that it's exponential, but it's increasing in an exponential rate. Will quantum computing come to the rest?

I think it's hard to talk about quantum computing with certainty because the industry doesn't exist yet. We've got companies that are in the very, very early stages of trying to create products. And so we're sort of like in the position of being in 1952, a half a decade after the first transistor was developed and trying to project forward when the iPhone would be created. We just don't even know, nor can we really conceptualize what the iPhone of quantum computing will be half a century in the future. So we're speculating pretty blindly here. I think you can say, though, is that if you look at the history of classical computing, what you find is that even the really revolutionary changes, they happened incrementally. And so it was a decade between the first invention of the transistor of the integrated circuit,

another decade between then and when big mainframe computers became standard in large corporations. Another decade after that, before the pocket calculator, which was the first real consumer product using integrated circuits was invented. Another decade after that for the PC, decade later for the iPhone. And so it was actually a very long process by which classical computing was applied to all different facets of our economy and our society.

And so I think that quantum computing will be similar. It's going to take a long time to sort out once we've got the technology sorted. What's the economic use case? What's the business use case? And how does that disseminate across the site? You said this place was steps from the water. We just haven't

the steps yet. How much did we save? Enough. Enough to get lost. Or you could book a stay with Hilton. Welcome to your oceanfront room. Just steps from the water. The Hilton sale is on now.

Book on Hilton.com or the Hilton app and save up to 20% to get the stay you expected. When you want savings, not surprises. It matters where you stay. Hilton for the stay. And when we look at kind of the historical parallel, will happen, as you said, you know, 70 years ago now and fast forward to today, we see parallels playing out, whereas the U.S. was engaged in a Cold War with the Soviet Union. You make the case that a decisive element of our victory, if you can call it that, was our preeminence in

semiconductor manufacturer and industrialization. Is that all attributable to, you know, capitalism versus communism? In other words, that there was such a profit incentive that, you know, the capitalists were destined to win. And if so, does that not bode well? And I'm not, you know, obviously I'm American, so I'm going to take America's side, America first in this instance. So does that bode well for China versus America? Because China is still a communist country. What are your thoughts there? How decisive was communism versus capitalism? Are there historical rhyming, you know, kind of analogies with what we see with China? Yeah, you know, I think you want to drill it down to its basics. That's basically accurate. It's true that the government played an amazing.

your role in funding the first inventions that set up the chip industry. The first chips that were produced were basically bought by NASA or the Pentagon. But if you ask, why was it that these initial inventions, which were purchased in very small quantities, but hundreds and thousands, turned into an industry that now sells millions and millions of units each year, the answer is because there was a huge consumer market and a huge corporate market that was opened. And so it's been, although it's been the government that has funded the initial invention, it's been the civilian market that has funded the expansion and has made it possible to lay down the capital investments that make it realistic for firms to keep investing every single year in new process technologies and new machines,

new chemicals. That's something that the communist world during the Cold War didn't have. They had the military market. They had lots of spending there. They had the best scientists in the Soviet Union who were basically conscripted into the chip industry. But they had a tiny market. And And as a result, they couldn't scale. And scaling is everything. And I think that's basically the same dilemma China finds itself in today. China is trying to build a self-sufficient chip industry because it doesn't want to rely on the United States and Taiwan and Japan for understandable reasons.

The problem is that self-sufficiency means you don't have scale. China's a big country, but it's only 20% of world GDP, whereas the rest of the chip industry is selling to 80% of world GDP. And if you're on the wrong side of that equation, it's just very, very hard to keep up. I never like it when an author, I've been in your, in your, shoes being interviewed on the podcast host. As you, can you summarize the entire book, you know, so that people don't have to buy it or even listen to the audio book or even buy the short form?

No, no, no. I am not short form. I refuse to do that. But TSM plays a big role in the book. And I want to see if you can maybe approach it from, you know, maybe I'm being naive. But my first, you know, kind of stance at this is that we have about 200 top universities in the in the world. And many of them are in the United States.

I mean, there's thousands of great universities, but 200 or so have eminent, you know, preeminent, not just photolithography, you know, fabrication facilities, but they also have the greatest scientists in the world, in my opinion. My colleagues are, you know, extremely brilliant. Why is it so hard to have, you know, to offshore TSMC and make it, you know, Arizona semiconductor, you know, and it seemed to be to de-risk a lot of the, you know, kind of existential threats that the chip universe is being threatened by with a potential conflict between China and the U.S. Why not take, you know, just make a Manhattan project like effort and onshore into, you know, maybe NATO, I don't know, you can say whatever you like, but why not? Why is it so hard?

You know, when my dean gets asked for, you know, $12 million startup, you know, to fund a new lithography machine and facility here, clean room class, you know, 300 clean room, they don't bulk at it. So why is it so hard? especially since Symer is located in America. Well, it all goes back to the question of scale, really. So TSM is the world's largest chipmaker. Because it's the world's largest, it has lower unit costs because it can drive down the cost of its purchases of wafers and purchase to chemicals.

But more importantly than that, it gets economies of learning because it learns for each wafer it fabricates. And so it learns more and has more chances to learn than any competitor. And so as a result, it's both got really impressive economics, but it's also got the best technology of any other company producing processor chips. And it's been in that position for some time. And what that means is that all of the world's chip designers want to work with it. And so it's got an ecosystem that's been built up around it,

whereby all of the companies that design ships want to work with TSM, because it's got the best cost and best technology, all of the companies that make the machine tools realize that TSM will be their biggest customer since it's got the scale. All the companies that design the software that's used to design chips need to make sure their software lines up with the SMC's practices. And so it's become the platform, so to speak, of the entire chip industry. It's like the App Store.

Apple's got iOS and everyone designs their apps to fit in those boxes. And that's basically what TSM is. And because it's got this immense scale, if you design a product that doesn't fit with TSM, you miss your biggest customer. And what that means is that if you want to replicate TSM, you've got to replicate the scale. And the scale is very, very expensive to replicate. So TSM is going to spend this year between $30 and $40 billion building new chipmaking facilities. That's roughly as much as the U.S. Chips Act is going to spend over the next five years.

So the U.S. government wants to build a chip industry. It's going to spend in five years what TSM spends every single year. And that just shows the level of difficulty in trying to build up that scale out instead of TSM. And TSM hasn't been investing that much for just one year. It's been investing huge sums for the past 30 years, which is why it's got more capacity to produce not only the most cutting edge ships, but also less cutting its ships than anyone else. So, you know, you convinced me in the book that the, you know, advantage in the U.S.,

even from its earliest days and probably still until now, as I say with nuclear weapons research being conducted on, you know, giant DOE computers, that the advantage was due to military funding. So, I mean, 40 billion, you know, it's a lot of money, but it's, you know, 4% of the U.S. military budget. Why isn't the U.S. military, you know, viewing this if it is so, and I agree with you, of course, strategically. Why isn't there, you know, a 4%, you know, kind of a diversion of military funding into the military, just military chips alone would seem to be a wise investment. So what are your thoughts there? Could we see that, do we need the boogeyman like we had with

the Soviet Union or is China such a, such an entity, or will we only, you know, kind of leave it up to private company? It seems like a great risk to our national security. But I think it is. And I think that's one of the key reasons why Congress a couple years ago passed the Chips Act, which is intended to revitalize chip making in the U.S. The challenge is that, although it's true the U.S. military needs cutting-edge ships. And there's chips in every F-35 fighter, for example, that today can only be made in Taiwan. The volume that the U.S. military buys is small. And it's true that $40 billion is just a small fraction of the U.S. defense budget. But you need 40 billion every couple of years because a chip making facility is cutting edge for a couple of years

and you need another one and another one after that. And so if you're sitting in the Pentagon, you can buy a new aircraft carrier for $10 billion or you can buy half of a chipmaking facility. An aircraft carrier is staying in operation for half a century where a shipmaking facility is cutting edge for just about two years. And so that's the calculus that led the Pentagon actually to decide very deliberately about two decades ago not to produce its own ships anymore. For a long time, they had this confidential fab where they could produce all sorts of chips for the military and the intelligence agencies. But they decided it was just no longer financially viable. And it was better to rely on Taiwan and save that money to buy

other types of equipment. You have a wonderful passage in the book. You talk about chips from Taiwan provide 37% of the world's new computing power each year. Two Korean companies produce 44% of the world's memory chips. The Dutch company, ASML, that we talked about, builds 100% of the world's extreme ultraviolet lithography machines without which cutting-edge chips are simply impossible to make. OPEC's 40% share of world oil production looks unimpressive by comparison. And actually, I think OPEC has even decreased now. I think U.S. is making more oil than even Saudi Arabia. Now, talk about the, let's do it all men love to do secretly, right? Let's do some wargaming in this book. So let's say Taiwan is attacked by China. I've used these machines about 30% of the time I've tried to use

you know, a stepper, you know, deep ultraviolet etcher or whatever, it's down for maintenance, and there's a tech in there and a bunny suit and they're working on it. These things do, you know, just like an F-35. I think we had on Hazard Lee, who is a stealth fighter pilot, wrote a book about that, you know, a fighter pilot. We had him on the show earlier this year. And, you know, he told me that there's about 20 hours of maintenance for every one hour of flight. And I don't, they're not, you know, maintaining the chips, but just all the system. These machines need an incredible loving, tender care. Could it even be practical in a war game scenario for China to take out, you know,

or take over the island of Taiwan in about a month they'd be out of raw materials. The machines would be down. And there's only one sole. They have to take over the Netherlands too. Well, I think that's, that is part of the dilemma. If a war were to start, Taiwan's chip fabs would shut down and they'd be basically impossible to get started up. again, unless you had the support from the Netherlands and from the U.S. and from Japan.

It's the machines, which, as you say, require all sorts of spare parts and expertise that are imported into Taiwan. It's also the chemicals, which largely come from Japan and the U.S. as well that are ultra-precise, hugely difficult to produce and need to be supplied on a regular basis. So if there's a war, Taiwan shuts down. The risk for the U.S. is that if you ask yourself, who depends more, more on ships made in Taiwan. Is it us or is it China? The answer is very clearly us. It's, you know, who are TSM's biggest customers? Apple, NVIDIA, Qualcomm, AMD. I haven't listed a

Chinese firm yet. They're all U.S. companies, mostly based in California. And so who would suffer more if TSM were shut down? Well, that's a really complex and interesting question. If you actually do the economic analysis, you find that China would suffer more because all the iPhones, for example, are largely assembled in China. So in pure GDP terms, China suffers more. But if you ask yourself whose political system is more able to sustain that type of disruption, I worry that the U.S. is less able to sustain it. And China knows it because China would be looking at the U.S. Everyone in the U.S. would be looking at their 401Ks, which are largely invested in the stock of Apple, Microsoft, Nvidia, Qualcomm, et cetera. And they'd be crater. And so I think there's some concern that

China might think that although in pure GDP terms, the impact on China's greater. In fact, China might be able because of its political system to withstand the disruption and the U.S. might not. You mentioned NVIDIA. Let's talk about them. As I understand it, they're not really manufacturing. Yeah, they're, you know, whatever, like 8,000, whatever they call their chips. It's their design. And then they're outsourced and they're made by TSMC, right? That's exactly right. Yeah. InVity was actually founded right after TSM was found. They wouldn't have gotten. started up, except for the fact that their investors knew they could go to TSM for manufacturing services. So, Vivida's never manufactured anything. They only design. They partnered with TSM from

the beginning. Talk about what you see exciting in the AI space. You mentioned, you know, autonomous driving and stuff like that before. What is the most interesting kind of on the horizon, not like futuristic, you know, 100-year quantum, you know, Tesla's driving around autonomous. Talk about what excites you in the near term for the chips that you write so intriguingly about? The key trend in AI has been that the most advanced AI systems require vast increases in data every six months. So there's been great empirical work done on the amount of data used to train the cutting-edge AI system of each day. And they find that every six or 12 months, there's a doubling of data used in AI training. So faster than Moore's law of the growth.

And so all of the advances in AI that we've seen have, have been possible thanks to advances in compute that have made this type of vast training possible. And so if you want to project forward trends in AI, it's been pretty clear of the last decade. You can just project forward data training volumes, which means projecting forward compute volumes. And that will give you what the next generation AI system will basically look like. And so there's a really deep interconnection between the types of chips that Nvidia makes and the types of advances in AI that we've seen. the next question is, you know, we know right now what it takes to train an AI system,

high-powered GPUs. But as we begin to deploy AI and all sorts of devices in cars and phones and industrial equipment, you probably need or want different types of chips that are optimized to that specific domain, have the right power to electricity consumption ratio, for example, for different domains. That's where a lot of focuses right now in figuring out what's the right hardware specification for AI inference in different use cases. And another, you know, kind of interesting historical, you know, kind of excursion you go through is Intel. Obviously, we talked about Gordon Moore and his contributions early on and not just technically,

but in terms of his, you know, kind of entry into the zeitgeist, our understanding of power of, you know, things that depend on power laws, of which Moore's law is just want. Talk about Intel and how they kind of lost. I mean, I remember in the 90s, you know, it was like on your computer, you got a, you know, from Dell or whatever, a big sticker Intel inside. Now it seems like they're, you know, hopelessly far behind. Are they going to be able to catch up or, you know, are they going to have to pivot into doing some other, you know, software services or data storage or something like that? Where they get away or come back to their roots. Well, you know, one of the big problems that Intel has faced is that the PC market,

which, as you say, used to be a real growth driver, has slowed. Now nobody knows what kind of chips in their PC because it doesn't matter. Chips are good enough. PCs run. They're not exciting anymore. And so Intel still dominates that market, but it's just a less exciting market not growing anymore. Then Intel in the data center, which is Intel's other big market, has lost out to companies like Nvidia. It was behind the curve when it came to AI.

And so the most recent data suggests that three out of four dollars spent in data centers past year went to Nvidia. And Intel and AMD have just been competing for the, remainder. So that's a real challenge for Intel they got to turn around. But the second issue is that they fell behind TSM when it came to manufacturing technology. They do both design and manufacturing and they've been lagging TSM for past couple of years. Now it looks like they've got a chance of catching up to TSM. We'll see if they do with their aim is to catch up by 2025. This is the 18A. Exactly. Yeah, yeah, exactly. And right now they say they're on schedule. And so it may be just as good as what

TSM hands. But then they've got to match that manufacturing capacity with designs that people want to buy. And they're right up against both AMD and Nvidia right now for both PCs and data centers. And it's going to be very tough competition. Another kind of interesting take in the book that I came away with is the importance of immigration to the all-chip story. Obviously, this involves cooperation and sometimes competition between different nations, different philosophies. I read an article recently that talked about how much the tightening of curbs on AI exports to China is going to really impact not only sales in the bottom line for NVIDIA, but also like recruiting of talent in both directions maybe. Talk about immigration. Has it been net good and with limitations on H-1B visas.

Many of my students and postdocs and so forth have benefited from that, both from China. and Southeast Asia. Talk about that. How is immigration figure into the chip war? My view is that immigration has been central to U.S. success over the past 75 years. If you look at all of the key turning points in the chip industry, you can find immigrants as part of the story. For example, the eight engineers who founded Fairchild Semiconductor, which is the first Silicon Valley startup and the chip space, two of them were foreign-born. So that's 25% right out the bat. If you look at the success of Intel, it was about Gordon Moore and Bob Noyes, but it was also about Andy Grove, who was the longtime CEO, was born as Andros Grove when he was born in Budapest before moving to New York as a refugee. And the same is true today.

Look at Jensen Huang, CEO of Nvidia, born in Taiwan before growing up in Kentucky. And so I think there's a very strong pattern. is basically as strong a correlation as we've got with Moore's law when it comes to immigrants playing a huge role in the chip industry. And it really is a challenging issue today because, as you say, as the U.S. tries to split the chip world into a China-focused world and a rest of the world sphere, it's forcing people to choose sides. And so there's actually been regulations the U.S. has passed that say neither U.S. citizens nor green card holders can work with certain Chinese firms. And there's been reports that certain green controllers have decided to give up the green card to keep working with these banned firms. And so it has had a direct impact. But I think in the long run, the U.S. is still got a pretty strong position vis-à-vis China.

If you look at the net immigration flows, there's a whole lot of people going from China to the U.S. and a much smaller number going from the U.S. or other countries to China. But as you say, the complexity and the small numbers of H-1Bs make it more difficult than it probably should be. It's peak pollination season, and my business is scaling fast. To keep the nectar flowing, I need a phone plan with top priority data speed. That's why I chose Google Fi Wireless. My connections stay strong even when the hive is buzzing. Plus, unlimited plans started $35 a month.

Now that's a deal that doesn't stay. Explore Google Fi Wireless plans today. Plus taxes and government fees. GoogleFi Wireless is not subject to data traffic deprioritization during times of high network usage. Being China, a group famed kind of geopolitical strategist, Peter Zahan. I don't know if you're familiar with him, but he's sort of America, I wouldn't say America bull, but he basically is claiming that, you know, China's on the verge of collapse. Their demographics are horrible.

I've, you know, pointed out I had on Mo Godot, who worked at Google X for a long time and is a great writer and thinker. And we talked about, you know, how even if it's true that they're, you know, going to lose net. net replacement of 300 million people, they're still going to have over a billion people, you know, for the next 50, 60 years and there's no chance of America catching up. Perhaps India already surpasses them. But what do you make of Zahan's kind of analysis that, you know, China's essentially, you've written, you know, several books in the USSR and their collapse and Putin and Gorbachev. Do you see any parallels?

Are they imminently about to collapse, as Peter might claim? Or do you feel actually where that may be, you know, a sciop that, yeah. they're perpetrating to make us underestimate how powerful they may actually be. If you look at the history of authoritarian systems like China's, what you find is that they always look pretty strong until the moment before they collapse. And it's very, very hard to predict the moment a brittle system just pops apart. And so when I look at China right now, things look pretty stable. But I think we've got to recognize that the Soviet Union looked pretty stable in 1989, too.

And then two years later, it was gone. Now, I don't think that's a prediction. I wouldn't bet on that if I were. betting on China's future. But I do think Peter's right to say China's got a lot of real challenges, the demographics, the economics, the confrontation, not just with the U.S., but with all of its key trading partners, Japan, Korea, Taiwan, India. If you were to, you know, arrive from Mars and say, which country would you rather be, I don't think you'd pick China. And we are just to pivot back to some of the personalities in the book. So talk about Jensen Huang.

He's sort of this iconoclastic figure, shows up at a, you know, a biker. jacket has a huge tattoo on his on his body kind of like a like a an Elvis kind of cult like guru following how much of that is to sort of is to blame for their success maybe and then do you do you believe that you know kind of this this cult of personality is warranted I mean is he's that a visionary I haven't researched beyond what you provided in the book as a description which I think is wonderful but I'll talk about him as a person as a lever as a visionary where Where do you see how he fits in in the grand story of chips, but also of innovators in the kind of Silicon Valley tradition? You know what I think he deserves all of the credit he's given.

If you look at what Nvidia has done, it is a series of extraordinary bats that looked wild and reckless when they were made and turned out to be absolutely central to advances in computing. When Nvidia was founded, and Jensen was one of the three co-founders, they co-founded it in a Denny's in San Jose. So it's like this classic kind of Silicon Valley startup story. Their only product was to produce graphics chips for computers. They did that pretty successfully for 15 years. And then they heard in the early 2000 stories of PhD students at Stanford and Berkeley and elsewhere using their chips for machine learning and AI. And they realized that basically the same designs that were good for graphics were also good for training AI systems.

And so they began pouring irresponsibly large sums of money into building out a soft software ecosystem around the idea of using their graphics ships for AI. At the time, they were criticized for being hugely wasteful, betting on an industry that didn't exist. Wall Street was furious all the money they were spending. And until last year, it looked like it might not pan out. But over the past 12 months, as we've just seen what large language models in particular can do and realize that basically all of them are trained on Nvidia's ships, every tech company the world has been scrambling to get their hands on products that Nvidia created. And I think Jensen deserves a whole lot of credit for seeing that trend nearly two decades in advance of everyone else.

And the important part is he not only saw the trend, he was able to stick with it, spend a whole lot of money building it up when most people said it was a crazy idea. Hey friends, it's me back again with a request for you to win a chunk, not of a silicon chip, but of a meteorite, a real 4.3 billion-year-old fragment of the early solar system. And all you have to be entered to win each month is join my mailing list at Briancating.com slash list. When you do so, you'll be entered to win one of these chunks of space schmutz, the four billion-year-old chunk of our solar system. Maybe not as high-tech as those silicon chips that we've been talking about today with Chris Miller, but nevertheless,

unique, interesting, and fascinating. I'll send you information about meteor showers and all sorts of other goodies and how you can watch them. So join my mission. And if you have a dot edu email address you'll be automatically entered to win and you'll automatically receive a chunk of space dust. Now back to the episode. But before you forget, visit briankeating.com slash list to join my mailing unless you'll receive the hottest news from around the STEM universe once each week on Mondays. Thanks. Back to the episode. I'll finish up with a couple of questions both in some way related to cryptocurrency to Bitcoin. It's been rumored that, you know, something like half the compute power on Earth is dedicated to,

as dedicated to Bitcoin mining by itself. And then there's, you know, whatever, 20,000 other alternative coins and networks and so forth. I had on Michael Saylor, extremely bullish about it. And, you know, his basic thesis is that, you know, Bitcoin is stored energy. And that's the irreducible kind of undilutable, perfect form of money, et cetera, et cetera. I'm not getting into the Bitcoin thesis, although if you have any hot tips for our audience, let us know, Chris. But in reality, the staggering amount, if it's true, that half the world's energy is basically dedicated computer compute power and something like the equivalent of, you know, all the U.S. nuclear powered carrier battle groups worth of energy is devoted to Bitcoin mining, which is significant fraction.

Talk about, you know, the prospects for, for, you know, savings or maybe the Picks and Shovel layer, or maybe the energy side of compute because, you know, it costs energy to compute, even to erase a bit, you know, at a fundamental physics of information level, costs energy. And so, and it contributes to waste heat and waste energy, which contributes to, you know, the use of fossil fuels primarily. So talk about how do we integrate energy into it? How should we think about things like cryptocurrencies? and they're, and they're, you know, I wouldn't say monopoly, but large demand on compute power and energy requirements. Well, it's absolutely right that this is a key challenge. The industry faces and really has spaced from the beginning.

If you look at why the first decision was made to move from computers based on discrete transistors to computers based on integrated circuits, lower power consumption was a key driver. And so the industry has been driven forward by its ability to produce more power-efficient compute. That's been a key demand from customers from the earliest days. And so if you ask yourself, why is it that every year smartphone companies pay a lot of money to DSBC to give them a better smartphone ship? It's because the power to compute ratio tends to get better on a pretty regular basis. And there's tradeoffs, more power versus more performance. And the efficiency laws that dictate that power consumption should get better over time have been performing much less well in recent decades than they did in the early stages. But there's a huge focus underway.

right now in finding ways to make it all more efficient. Because as you say, it is fundamentally a key input to compute. And if our demand for compute is basically infinite, I think it basically is our supply of power is not. And so we've got to find ways to keep economizing on electricity usage. Yeah, my favorite kind of thing to think about quantum computers is that they're extremely good at solving, you know, what's called the Grongian for our quantum computing system. They're very good solipsistically at looking at the underlying quantum mechanical equations. And those are governed by things called Lagrangians. And they're very good at finding these minimal energy surfaces and doing path integrals and all these things that Feynman talked about way back when. But they're not,

you know, yet found, as you said, scalability of product consumer, you know, viable consumer option, et cetera. But my favorite, you know, kind of future optimistic, you know, polyannish view is that, yeah, quantum computers are going to figure out of. a way to get high-TC superconductivity, which would then reduce the energy draw on compute and almost every other thing that we depend on for our modern technological world. Well, Chris, this has been a phenomenal interview. I thank you for your time. I can't resist asking you one final question. I usually ask when I'm speaking with a Nobel Prize winner, so that shows you the esteem I have for you. This podcast is called Into the Impossible, and it derives its name from Sir Arthur C. Clark,

who said the only way of determining the limits of the possible is to transcend them and go into the impossible. You're a professor. You've written bestselling books, you know, one of the FTs, finalist, best book of the year. Give me 20 seconds of advice. You have 20 seconds with your 20-year-old version of yourself. What do you tell him? What do you tell him to give him the confidence to do as you've done to go into the impossible? Well, I think, you know, it's a hard question for historians to ask, because our job is to look at things that have already been done, so we already know that they're possible. I think if there's one thing that I guess I've learned,

it's about the need to try to connect different disciplines together. When I started, I had no knowledge of electrical engineering or physics or material science or anything, just a background in economic history. But it turns out that just applying that background to an industry like semiconductors can produce a lot of pretty extraordinary stories and unlock a new way of seeing technology. And so that I think is what I hope made,

the book pretty interesting. And there's a lesson that I drew from the research. That's it. I can't resist asking one final question. It's prompted by Richard Feynman, who is in some ways founder of nanotechnology, a famous, famous paper. He probably read it called Plenty of Room at the bottom. And he starts off that he says, I imagine experimental physicists must often look with envy at a man like Cameron-Lan Amunds who discovered a field like low-temperature superconductivity, which seems to be bottomless in which one can go down and down and down. such a man is then the leader and has some temporary monopoly on a scientific adventure. I see a lot of that in this book. It's told so wonderfully. And if I could close with one last question to you, again, inspired by Sir Arthur C. Clark, the namesake generator of this podcast, after all. And he said that any sufficiently advanced technology is indistinguishable from magic. And I like to think Feynman answered that question once. He said, what statement contains the most information about the physical universe in the fewest words? And he said, basically the atomic hypothesis that everything's made of little atoms whirling around and bonding

and combining together. What do you see as the most magical form of technology? Is it the Domeconductor chip? Is there something else that we didn't get to discuss? What form of technology is most akin to magic, in your opinion? We talked a bit about the ASML lithography machines that convert tiny balls of tin and to light at just the right wavelength, 13.5 nanometers, to bounce off the flattest mirrors in the world and etch tiny circuits on the chip inside of your iPhone. To me, understanding how that machine works and understanding how every day of my life I relied on the chips that it produced, that was really a magical discovery. It's still sort of a shock that machines like that work on a regular basis. Absolutely. Well, Chris, where can people find you and learn more about your

work and your other books? Fantastic. My website is Christopher. for Miller.net. Chris, thank you so much for your valuable time. I enjoyed it very much. I hope we can speak again in the future, maybe meet up someday. Great. Thank you, Brad. Ambition comes in all shapes and sizes. At First Citizens Bank, we roll with your goals because we're built for what you're building. Fit for your ambition for Citizens Bank.