Samo Burja on Growth, Energy, and AI

0:05So I actually think a big macro story that we've been exploring at Bismarck Analysis and Bismarck Brief is that the demands of AI are so massive that for the first time in decades, the economies of scale necessary to supply them require industrial revolutions in everything. So this demand shock, and we could almost call it a demand shock, arriving from the future in AI is going through silicon, but eventually it's going to reach things like mirrors and steel and natural gas. And once you're doing those things, oh, buddy, you've reignited the industrial revolution. Most conversations about AI focus on software.

0:45Samuel Boria thinks the bigger story may be physical. As AI systems become more powerful, the demand for energy, compute, data centers, chips, and industrial capacity is growing at an extraordinary pace. That demand could reshape not just the technology sector, but manufacturing, construction, energy production, and the broader global economy. Theo Jaffe speaks with Samo Burya about AI industrial revolutions, demographic decline, state capacity, and the institutions that determine whether societies can successfully adapt to technological change. We're live with Samo Burya. Welcome back to The Situation Room.

1:27Happy to be here. So we just finished our last segment talking with Ray Ma from China Buzz about, was it I don't want to get that wrong it was tech buzz China that's what it was and Sophia made the point that the U.S. has these sort of like grand apocalyptic eschatological narratives around AI because we're a Christian society and China doesn't because they're secular society you have any takes on this? I'm not convinced. Well, I think that people over index on long civilizational narratives and under index on just the concrete material conditions in a country. To give an example, there was the so-called Yellow Turban Rebellion in China in the third century.

2:17You could easily make the case that China has eschatological thinking and has end of the world cults, right? The so-called Ill-Turban Rebellion was basically a Taoist cult. And it was a bunch of brothers that wanted to overthrow an imperial dynasty. They believed the end of the world would come, that, you know, the heavens themselves would change, and that there would be a massive transformation into an unprecedented utopia. This was actually believed by many, many people, And it was a very bloody civil war and likely one that ended up destroying, you know, destroying the Han Dynasty. I think that people wanted to equate deep religious traditions with basically the beliefs that a society holds.

3:08But they underrate how much those religious beliefs change on the century timescale or the millennial timescale. I mean, certainly the, you know, religious landscape of the U.S. has changed enormously in the last 50 years or so. I think that the United States is still innovating in its practical religious sense. You know, you could say that the rise of this kind of non-denominational Christian-derived ethic where basically, you know, just be a hecking good person, whatever that means, right? That vibe, that ethic seems to be the actual religious belief of most people. And then some sort of vague belief in karma, some sort of vague belief in like an inactive God that doesn't interfere in the world.

3:57I think that actually probably describes most people who are nominally Christian in the United States today. How do you think like the Puritanism of the early U.S. manifests today? You know, Puritanism is Calvinist. They were very sort of against the pleasures of the flesh. They believed in predestination. So you were born predestined to go to either heaven or hell. Well, there's a straightforward interpretation where, you know, simply because you're doing well, that doesn't mean your salvation is guaranteed. It's not clear that you were one of the elect. But it is a hint. It is something of a possibility.

4:37So I think the classic arguments about the Protestant work ethic endure. However, I think they endure because there was first a religious belief that, okay, it's important to do your work well. It's vocation from God. If you're doing really well materially while doing the vocation that, you know, God selected for you, that might be a sign that you are elect. You're very much nervous about whether you are one of the elect. So one of the people that are predestined, you know, to be saved. But then, even as society secularized, the material rewards of working very hard started compounding. So I sort of think that in a way, the fact that, you know, America was started as a more Protestant, almost Calvinist country, or a Puritan country to, you know, take a slightly different strain, that was reinforced.

5:31It was important for the emergence of capitalism. But I think capitalism itself prolonged it, right, and developed it further. I think today people work hard on the vague assumption that, you know, it's going to work out for you, right? The whole land of opportunity mythos, which was also true for most of the 20th century, arguably still true in the 21st century here in San Francisco, I think that has long ago secularized and it's sustained by the facts of economic growth. Well, the conclusion from this that I would draw regarding China is that if you want to think about China today, maybe you should think about 1950s America.

6:15For most Chinese people alive today, their life has just gotten a little bit better and they've become a little bit richer every single year for the last 45 years. And many families are now buying their first car. The living space allotted has increased. I think recently they surpassed some of the lower ranking European countries and, you know, average urban living space, which again makes sense, right? We talk about China's ghost cities, but they built a lot of apartments recently. Meanwhile, if you're, you know, living in Paris, even if you're very rich, your apartment was probably built in the 19th century.

6:53Or California, for that matter. Unfortunately, yeah. Not the 19th century, but like the 1950s. 1950s, yeah. Or if you're in San Francisco, like a lot of the housing stock of San Francisco, you know, all of our historic Victorians were built right after the earthquake in 1906. So rather old housing stock. one that could be greatly expanded. You know, maybe this is just a long-winded, you know, maybe this is just a long play by the automation lobby. It's not the landlords that are in charge, constricting supply. It's the robotics companies, right? Secretly, they're in control, I could joke, because if you make housing very expensive, then you're going to see$50 lattes eventually.

7:36I mean, you're already seeing$20 lattes in the city. I think 50 is not far away. So, you know, eventually the robot makes economic sense. I don't know about 20. I've seen 10. $20 smoothies at Erebon maybe. Perhaps, perhaps. I feel like I always mentally count two coffees. I think it's always a good idea to just buy. You know, if you're meeting someone, it's just such a nice gesture to always like buy for the other person. And if they're Eastern European, then they fight you for the privilege and, you know, you relent. Yeah. Yeah. So you also tweeted recently, Sam Hammond was talking about nationalizing AI companies is a terrible idea in reference to the proposal floated by Bernie Sanders and Donald Trump and OpenAI.

8:21And then you said printing a trillion dollars and paying it to AI companies exchange for equity might be the best economic stimulus the U.S. government can carry out and might yet prove a vastly better investment than, say, the infrastructure bill was. Well, you know, the U.S. government has been printing trillion dollar bills like it's no tomorrow, practically, right? We're not talking just about the infrastructure bill. There's been the COVID stimulus. There have been many, many cases where the U.S. that, of course, has been spending with a massive deficit, has a massive debt burden, has injected trillions of dollars of cash into the economy, mainly to make sure that the stock market doesn't go down.

8:59You know, at this point, I feel basically anyone that holds a 401k, anyone that owns stock, you're basically the beneficiary of the high printing system, right? It's not money that retains its value, it's stocks. And politically, it's far too costly for the sitting president, no matter his party affiliation, to allow the stock market to go down for too long. And, you know, I think this political incentive is massive. I don't think it's going away. And I think because of that, we will be printing trillion dollar bills. And then the question is, well, who gets them? And I sort of feel it's actually better the AI companies get them.

9:39If I told you that the U.S. government is investing a trillion dollars into data centers, that would seem good industrial policy. If I tell you it prints a trillion dollars and buys a trillion dollars of stock at like basically the value, the market value of, say, OpenAI or possibly Anthropic, you know, I'm pretty sure they're going to use it for employee compensation and for data centers. So not that bad, actually. Hmm. So I'm taking a pessimistic view on sort of fiscal sanity of the U.S. and an optimistic view of the institutional health of the frontier AI labs. Sure. How do you think this would compare as a stimulus to the average person compared to just directly giving them money, though, or giving them money through the welfare state?

10:29I mean, I think it obviously is going to be much longer term as a positive consequence. You know, the best asset for a welfare state, if you want it to be sustainable, is a booming economy. Countries like Sweden and Denmark used to understand this. This is why they had an export-driven system. That's why Sweden isn't just the land of welfare. It's the land of capitalism, right? It's as much the country of IKEA as it is the country of, you know, free health care. And IKEA is like, you know, globally innovative logistics company. I think that if we want the United States welfare state to continue, I think this would be a better long term decision.

11:11I don't think politicians are good at making long term decisions, but maybe something pettier, maybe something like, you know, the emerging bipartisan corruption consensus will incentivize office holders to, you know, give corporate handouts to companies that are likelier to be profitable. rather than just give it to, you know, the dole. I don't know. It seems like rather the opposite of the economic incentives that we do for profitability. You know, like the way we do corporate income taxes is by taxing the profits of companies. I mean, again... Why would we reward profitability in the tax system?

11:52I mean, we always do punish profitability. But I do think that investments in infrastructure build-out are a positive. So let's just be very frank. The reason I think that perhaps Sam Altman is floating this is that not only does it position his company as the company that's more benevolent to the general public than Anthropic, making an offer Anthropic has not made, it sort of improves their PR on the margin, right? It sort of vies for the position of least disfavored AI company. There's a way in which there's like a big AI wave and a big anti wave of sentiment, right? Sentiment that is opposed to AI.

12:36And in a way, Anthropik has positioned itself as the least bad company, right? That was also related to the papal statements we saw about two weeks ago, that, you know, picture of Chris Ola talking with the Pope and commiserating how, you know, possibly AI is bad for inequality, right? That makes the bleeding hearts like you. But then you outflank them from the left in a way. And you say, oh, I want half of my company to be owned by the government. But also current U.S. law says that probably the government will have to buy you out. So what does that do to the Open Eye IPO if the U.S. government's putting in a trillion dollars into it?

13:16I actually think that the transformation of an American frontier lab into this weird state-owned enterprise type of system might actually be a great business decision for the company. It's not necessarily great for capitalism or great for technological progress. But, you know, I think that, you know, being the most favorite government provider and being partially government owned, there is even a national security argument for this. Do you think capitalism can even survive full labor automation? Capitalism can certainly survive full labor automation, as long as, you know, they're still competing sort of AI entities and systems.

13:56I do think politically, it might result in basically all of the humans joining the welfare class. And then there's a puzzling question as to whether the welfare class like retains political power or not. I do think ultimately some amount of restraint in the expropriation will just happen just because people, you know, people want the robots to work, even in a situation where only the robots are working. I just don't understand why the voters would keep control of a state, right? I think people have control of states because they're either useful as conscripts, useful as taxpayers, or potentially because they can riot in the street and, you know, overthrow the government.

14:42And that's what gives powers to protests. And protests are what gives power to sort of the welfare voting class. Yes, this is the idea of the intelligence curse, if you've read it. Great essay. I have not, but I'll add it to my reading list. I do think that the political economy of the U.S. will be profoundly transformed even by partial automation. What do you mean by that? I mean, we have already had partial automation for many decades. I don't know how much the political economy of the country has been transformed. Like, is our system that different from what it was in, say, 1930, when there were a lot more people working in agriculture?

15:26I think the argument pretty much makes itself that it is, right? If we post-date the New Deal changes as the great automation of agriculture and a mass migration to the cities, note that, you know, people talk about the Dust Bowl, people talk about the Great Depression. But part of the reason there was massive population pressure of small farms that were no longer economical, that was partially the automation of labor. The mom and pop farm could not really compete. And today, the landowner class in the United States, you know, we say farmers, but really, you know, we have landowners and farm workers, right?

16:05The sort of family owned farm is such a minuscule part of, you know, the economy. And it's such a minuscule part of the voter base that anytime you look at legislation related to food quality, you should just be thinking about giant agribusinesses, right? So I would also say that another transformation of the economy was globalization itself, where, you know, you can think of China as a big robot. And basically China starting in the 1970s and 80s, I grant that first it was Japan before China and China more took off in the 90s and 2000s. But let's say East Asia, the East Asian industrial complex just outcompeted the American worker.

16:47and whatever you think of unions today, right, these sort of blue collar, industrial, Democratic Party aligned unions, I think they're weaker than ever. Only the government worker unions are more powerful. So when Joe Biden was like running on this vision of almost a revival of the American worker class and the unionized worker of the well-paying job, it just seemed like a massive anachronism. So I would say that, yes, the composition of the labor force and who is best compensated and who is most politically useful, I think that does regularly change the politics of countries, even if the Constitution on paper remains the same.

17:32Right. After all, laws can stay on the books for 200 years, but they'll be interpreted differently given the needs of the moment. for sure. You wrote, if your economy is a knowledge economy, you're worrying about mass unemployment. In contrast, if your country has an industrial economy, you're looking at 10 % year on year economic growth, Taiwan, South Korea, but eventually also less obvious ones like the Netherlands, which has ASML. How compatible is this with fertility rate decreases eventually causing a permanent recession? You seem like two kind of opposite takes on future economies? I think that the industrial economies of China and East Asia and so on, I think that they will, in fact, be experiencing in the very near future, right?

18:24So the statement you read was about, let's say, the next 10 years, right? In the next 10 years, if you are, say, doing finance in London and your company does not have access to frontier models, you're probably going to be in trouble. If you have a law firm in Geneva and your company doesn't have access to frontier models, you know, let's say mythos for law, you're in trouble. Now, if you're doing either of those in San Francisco or maybe New York, you're a little bit better off. But there's a way in which the global white collar class will see disruption at the very least from companies making use of the most advanced AI they can at scale combined with the highest prestige brand that you can.

19:09You know, if you think about it straightforwardly, if a law firm, if America's most prestigious law firm can handle a million customers rather than 500 or 5 ,000 customers, why would you ever hire anyone else, right? Even if 99 % of the time you're talking to an AI and then for five seconds, for two minutes on your screen flashes a human lawyer that signs off and just asks you, hey, did you read the statement we sent you? Telehealth for lawyers. Exactly. Right. So I think that that might still work for some white collar economies, but it'll be a winner take all. Meanwhile, when it comes to industrialization, you know, you have Taiwan reach 10 percent growth.

19:54Taiwan is the country I was referring to. South Korea might reach 10 percent growth in the short run because of memory chip production. And then when you follow this chain, you realize, actually, if ASML keeps growing to meet demand, and currently it's the only company that can, the Netherlands might see 10 % economic growth eventually as well. Now, politically, the government of the Netherlands might be kind of at war with ASML, but it's far from a given that they won't just embrace it because it might be a much needed cash infusion for the problems of aging. I think it's like$650 billion market cap, and then the next most is like in the$300s.

20:36I mean, my opinion is, and I'm curious whether you would agree, that we should assume that's going to at least double, and that that's a very conservative estimate. I think so, yeah. Yeah. You know, Broadcom, which doesn't even make fabs, is worth three times as much as ASML. Mm-hmm. So if we think of ASML then as a company where the growth is already priced in, and maybe we see 10 % growth in the Netherlands, you have to understand that the company does as much gardening of its industrial ecosystem as NVIDIA ever does, right? They rely greatly on many other Dutch companies and many German companies, especially in the optics sector.

21:18So I think Germany, of course, is a larger economy than the Netherlands and, of course, much larger even still than Taiwan. However, you know, despite the high energy costs, I think we're going to see something like three or four extra points of economic growth. And that's going to make a massive difference for Germany. I don't think Germany deserves its economy to be rescued by the headwinds, but it shows that even a failing industrial base can be bolstered by AI demand. So I actually think a big macro story that we've been exploring at Bismarck Analysis and Bismarck Brief is that the demands of AI are so massive that for the first time in decades, the economies of scale necessary to supply them require industrial revolutions in everything.

22:07Consider the energy build-out. Energy build-out directly results in higher demand for steel, higher demand for cement, higher demand for construction crews, higher demand for basically every type of construction machinery you can think of. This is working its way through the supply chain. Just as you might say that the oil shock from the Strait of Hormuz needs a while to go through the supply chain to see the price increases, so this demand shock, and we could almost call it a man shock, arriving from the future in AI is going through silicon, but eventually it's going to reach things like mirrors and steel and natural gas.

22:44And once you're doing those things, oh, buddy, you've reignited the industrial revolution. Now, the only question is, how much fuel does the industrial revolution have, right? How much fuel is around for ignition? And I think that's where the aging populations come. So there's a headwind. people sometimes misuse headwind. Headwind means that it's actually a problem, right? Because it's slowing your ship down. There's this big headwind of global grain. I do think that eventually South Korean growth and Taiwanese growth will stop. They will have to out offshore and outsource. And when countries accept partnerships like this, for example, if you go build in Indonesia, they want some of the growth to go to Indonesia, of course.

23:32the government there, or India. So eventually the growth will stop in Taiwan. It'll stop even in coastal China. It'll stop in South Korea. It'll stop in the Netherlands. Taiwan, of course, in contrast with their 10 % GDP growth rate, has the lowest fertility rate in the world. It's like 0.65 now. It's lower than Korea. The only place where that's not true is at the one company driving the growth, TSMC, right? That company has notably higher fertility rates the national average. And, you know, I think that's kind of interesting. I think that if we saw that in Silicon Valley, where if top tech executives had notably more kids than the rest of the Bay Area, I think that, you know, that might be a long-term positive sign that fertility rebounds.

24:22You know, Elon has many, many kids, and I think that... But he lives in Austin, right? Like, that's not SF. Yeah, but it's sort of like a negative signal for the culture. You know, people think it's weird. That's true. But definitely, you know, I think it would be really helpful for aspirational cultural figures, actors, musicians to have lots of kids and to have them young. I think that's true. At least it's a vote for the future. Though I know that a lot of tech executives that do have kids like don't get that much attention for them. Right. Sam Altman is a father, for example. Of one child.

24:56Yeah. So far. You know, one child better than none. That's true. So last question before we go. You did a poll. Which things should you write about in more detail next? Live players, traditions of knowledge, functional institutions. Which one of these do you think wins? Well, you know, the poll is still open. So if anyone wants to check it out, they can go to my Twitter account, Samo Buria on x.com. I just voted functional institutions. I think that the era of functional institutions is not over. And I actually think it'll be functional institutions that will best adjust to AI. If you think about it, AI provides you with the ability right now to scale white collar labor.

25:39If you have broken processes and you try to fit AI into a broken process, you are just putting the weight into all the remaining bottlenecks inside your company or organization, right? It could just as well apply to a government bureaucracy or honestly, for that matter, the Catholic Church, right? Clearly, some parts of the Nsika were written by AI. So, you know, the Vatican has AI in the loop and maybe God too. The whole bureaucratic process that has driven the 20th century produces organizations that decay rapidly and can waste immense amounts of natural intelligence. So I think the functional institutions will definitely be the winners.

26:26And it remains relevant to figure out what are those principles that allow you to run a, let's call it, hyperscaled bureaucracy effectively. Right. Well, we'll have to go into more detail on that next time because we're out of time. So thanks so much, Samo, for coming on. Pleasure to be here.

27:06at a16z.substack.com. Thanks again for listening, and I'll see you in the next episode.

27:36is from sources deemed reliable on the date of publication, but A16Z does not guarantee its accuracy.