Hi and welcome to Ones and Twos, FP's economics podcast. Every week we take a couple data points, use them to try to explain the world. I'm Cameron Dati, FP's deputy editor with you in the outskirts of Washington DC, the suburbs of Northern Virginia to be exact. Joining us as always is Adam Tooze, FP's economics columnist and Columbia University professor, this week in Singapore. Hi Adam.
So, we have one data point for you all this week. That is 285.9 billion. That is the estimated amount of private AI investment in the United States in 2025. The companies doing that investment, Microsoft, Google, Amazon, Meta among others, have committed hundreds of billions of dollars more to AI infrastructure in the next few years alone. And yeah, that would have seemed potentially crazy just a few years ago, perhaps even just the last time we discussed AI economics on this podcast, which was even just a few months ago. But what has come into clear view is that this spending is coming in the context of what could be called an AI arms race.
US private spending is far outpacing that of its major rival China right now. But China could be making up for that in other ways, with the state taking a leading role and its companies maybe even finding ways of innovating around the constraints they face. But if this is really all analogous to the 20th century atomic arms race, that raises the question of how the arms race logic will play out. Will there be a Sputnik moment? How can we tell who's winning? How can we tell what winning even means? So yeah, we thought we'd unpack all this with another segment on AI economics and the arms race in particular.
Adam, to start off with, all that private investment that I mentioned at the outset, economists have recently been flagging that amount of investment as a potential problem all its own. I'm curious why they think that could be the case and are they justified in doing so?
Yeah, I mean, speaking to us from where you are, Cam, you're really in the middle of the data center boom, aren't you? That's — Virginia is where an awful lot of it's happening or was happening until all sorts of supply chain bottlenecks started to emerge which are one of the reasons why economists are beginning to get concerned. There was a very substantial report by the BIS, Bank of International Settlements, in its annual economic report which is a kind of flagship publication. So if you think about the sequence of publications, there's the World Economic Outlook from the IMF and the World Bank in the spring of every year and then the summer the BIS puts out its report and one of its chapters, a big part of the report, is on financial risks from AI and economic policy implications.
And it's I think increasingly clear that in places like Virginia, AI is an inflationary pressure. So that's one reason why people are worried about this. So there's, you know, if there's a big deflationary pressure in the world economy, it's China shock 2.0 and the huge surge in Chinese manufacturing exports. If there's an inflationary pressure, it's the private sector investment boom endogenously financed. So the more it grows, the more money is pumped in. So it doesn't appear to be a tight budget constraint on private investment in the US. And from the economist point of view, that's a kind of worrying phenomena because it has a kind of runaway property where as more enthusiasm builds, more money's pumped in. So there's more enthusiasm and so on and so forth. And then people start worrying about bubbles. So those are the sort of reasons why economists are beginning to be concerned.
And then you can break it down a little bit further. So the knife edge kind of logic, which our show, in fact our conversations, were amongst the earliest ones I think to really outline this, is that you know it's a kind of two-way bet on disaster either way. Either the big money flowing into AI is vindicated and so then this isn't a bubble but in that case the labor market is shocked to its foundations and the BIS considers this possibility with a chart showing the labor share of income falling to 20%. So one of those kind of extraordinary apocalyptic scenarios. Or on the other hand, the labor market payoff is not there in which case the companies are overvalued and there's going to be losses somewhere in the chain and then the questions begin to be asked about the financial stability of the money that's piled into the supply chain.
And there's a series of concerns about the financial aspects of this which are ramifying in their implications. So the most obvious is that American households, not just American but above all American households, especially those with large quantities of wealth, are holding an ever larger percentage of their portfolios in stock markets which are more and more concentrated on the firms which are in the AI pack and those shift all the time. Classically, it's the big platforms. But increasingly in the last six months the focus has been on Nvidia and the chip makers Micron and players like that. So there's concentrated ownership risk from people with equity. Now equity is risk-bearing. So if the prices go down for the stock then people make losses and that will hurt a lot of big portfolios and that could cause a shock to the US economy because rich households will feel less wealthy, will spend less and the economy could cycle down. Of course if people are losing their jobs too, nightmare and you know double hit to the economy.
Another option for risk here is that somewhere in the credit chain, so not in equity direct risk-bearing investment but in form of loans, there could be problems which would then ripple through the financial system and quite a lot of the loans come from private equity which is very untransparent. We don't really know what the risks are. And then a further thing that's really concerning, people have been working hard to outline the risks here, right? A further thing that's concerning is there appears to be circular financing within the AI bubble. So companies notoriously are, you know, placing big orders for Nvidia chips, which are boosting Nvidia, which is then investing in the people that buy its chips. So you have a circular flow and on many of the balance sheets of the participants in this boom, profits are being booked which are essentially improvements in the book value of firms whose book value is increasing because people have bought stakes in them which are part of this chain and whose profits have been increased.
So you've got this circular flow and again you raised a bunch of questions about this last year and I downplayed them but more and more emphasis being put on this dimension. So from every side, from the labor market potential to the immediate macro effects to the potential risks in wealth portfolios to financing, there is a huge cluster of concerns building up here. All of which are shrouded in well huge uncertainty and dramatic dynamism. So it's a pretty intoxicating combination. And wherever I go in the world right now, there's really only two conversations and one is the China shock story and the other is AI.
So if one were to try to combine those two stories in some ways as you were just suggesting, you were recently in China, we were recently in China as well, recording the podcast in Dalian at the World Economic Forum conference there. I'm curious what the view from China so far as you can tell is on the subject of this investment boom in the United States. Is China observing it from afar? Is it participating in it? What is the dynamic involved there?
With respect to China and this AI investment in the United States, it's a really strange combination of motives and factors. I mean at the technical level, it's I think consensus that the only players that can compete with the US are the Chinese AI players. And it's only the big Chinese platform businesses that can compete directly with the US AI powerhouses who have all of the different elements, chip development, model development, the whole works. So that's a given. On the other hand, it's also clear that the Chinese AI players are not engaging in the same huge hyperscaling investment as we're seeing in the US. And that's in part because they're being discouraged from doing it, but also because their business models don't point in this direction. Also, their approach to the AI game is more focused on intelligent, smart, relatively lightweight models which are more open than the proprietary, massive, hardware-intensive, compute-intensive American model.
It's compounded by the fact that it's really hard for the Chinese to get the top chips. The very best chips from Nvidia are not entirely absent from China because of various workarounds but hard to get hold of such that the Chinese are doing interesting workarounds with clusters of their own lower performing chips but also increasingly heavy investment in Southeast Asia. Malaysia close to where I am right now has become a major center for AI data center investment.
But then on the other hand, so on this side there's kind of parallelism within divergence but when you get to the finances the story is really very different. There's a huge disparity between the valuation of the major AI players in the US which is increasingly driven by the chip producers who are much more present in Korea and Taiwan than they are in China, with huge gains being made by the Korean producers and enormous sales by TSMC out of Taiwan. Like this entire cluster ultimately depends on chip supplies from Taiwan.
And then furthermore, and this is something that really came into view for me with conversations in Beijing, thinking about the situation in Beijing, and now actually in conversations I've had in Singapore, is that you know the rest of the world other than the US, other than China, is vicariously making a big play either directly or indirectly in the American stock market and participating in the run-up in valuations in the US, right? The whole world funds America's trade and government deficits by buying American private assets and they're buying those private assets on the American stock market. So this party in America is everyone's party but the Chinese.
So because of the capital controls, the exchange controls that were put in place in 2015, it's not impossible to get money out and billions and billions, indeed over a hundred billion do flow out, but it's relatively difficult. And if you look at the overall portfolio, the Chinese are the other great lump of giant wealth in the world alongside the Americans. They've massively underweighted the American AI bubble. So there is at the level of global wealth allocation a big gap between China and the rest of the world.
And so a lot depends in the sort of, you know, in the race not of technology companies or of global GDP but between rich people if you like, where the richest in the world are, the huge fortunes, the centibillionaire fortunes are out of reach for the Chinese wealthy at this point. Right? So this is a structural break. When we were talking about the Bill Gateses of the world, then a Jack Ma who's worth tens of billions of dollars by any reckoning was kind of at a level, was kind of in the game with the wealthiest folks in the United States. But as American wealth breaks away to these stratospheric levels with Elon Musk being the most radical of all, China's wealth is sort of left behind in this historic gear shift from huge wealth to crazy ballistic out of space kind of wealth. They can't get there.
And that I think is in the end quite a profound tension. And it's emblematic also of course of the power of the Chinese regime to set the terms. We argue a lot about whether China's capitalist. Large parts of Chinese society clearly are organized around market principles, around private accumulation. But when it comes to this most basic element of portfolio allocation, like where really wealthy people can put their money right now, there's a huge gap opening up between China and the rest of the world.
Yeah, that also does seem to underscore the ways that these parallel, you know, technological processes in China and the US are structured ultimately as a geopolitical competition. I mean, there's not a kind of cooperation on the level of finance, on the level of wealth accumulation, on the level of technology and sharing it. Which is again why the analogy of the arms race comes up. And I'm curious then how China manages to keep up. I guess you alluded to that a bit technologically, the different approaches it might be taking. But if this is an arms race, in what ways exactly does it parallel the nuclear arms race of the 20th century? What lessons perhaps does that earlier arms race offer for us today?
I've been thinking about this a lot. On a panel the other day in Beijing, I was getting a lot of questions that really gave me pause about AI and I tried talking about Marx and AI in Beijing. I thought maybe people would have a basic Marx education. Total blank, absolute stony face blank. So I was thinking about like films that had been in popular culture recently and for some reason Oppenheimer popped into my head and we did an episode about it. And I can see, you know, in the conversation about AI, the Manhattan Project has come up a lot and Niall Ferguson did a piece recently for the Hoover Institution about you know AI global competition pointing out how dangerous the competition had become, how suddenly it had exploded and it was in the context of the Mythos program which that Anthropic had launched.
But what I realized on this panel was you know that the fundamental difference here is private ownership. I mean, imagine if the atomic arms race had been run competitively by five or six eager maniacal — well, imagine if Wernher von Braun had been in charge of his own rocket, nuclear-tipped rocket program. After all, not very far away from that reality. And he was competing with four or five other similarly wired individuals to control the destiny of humanity in terms of this technology and that's how bonkers our situation is right? Because you know you can say a lot about the craziness of the nuclear arms race but it was at least conventional in the sense that it was state controlled.
I mean clearly private businesses were involved in the Manhattan Project, it was a giant engineering, chemical engineering, heavy engineering project, right? And clearly government money has gone into Silicon Valley in various ways as Mazzucato and people like that will remind us. But nevertheless, the contrast between the Manhattan Project, which is a wartime project centrally driven by the American military machine, indeed the modern Pentagon in large part comes out of this project, and the realities of the current moment spearheaded by a figure like Elon Musk who literally has eviscerated the state apparatus of that first race. Right? NASA being the victim of SpaceX which has now made him a trillionaire basically on the back of an AI promise, right? Not so much the rockets but of the AI that's supposedly kind of integrated in that program.
It strikes me that it's really a regression away from the mid-century arms race model to something that's much more like the merchants of death vision of arms races, which is late 19th century, early 20th century, in which Marxist theories of imperialism and left liberal theorists of imperialism saw the arms race as essentially driven by private capital. The Krupps, the Armstrongs, the Schneider-Creusots of this world. In other words, the private arms manufacturers they saw as driving into imperialist competition that then led to war. And there were congressional inquiries and inquiries around the world in the 20s and 30s about the private arms business as the ultimate profiteer of war. And that discourse really shrinks back in the 30s, 40s and 50s, not because the private businesses aren't involved in the military-industrial complex, but because the state has expanded to such an extent.
And if AI is a nuclear-analogous arms race, the terrifying thing about it is that we're back in the world of Elver. We're back in the world of Krupp and Armstrong. We're back in the world of late 19th century military-industrialism where a British naval arms manufacturer would just sell a cruiser to a Latin American state largely without government control. And we're kind of back in that space of a private, unfettered, oligarchic, global arms race where the Pentagon will suddenly realize that oh god, Anthropic's got something there that's potentially lethal. You know, that's a manifest threat of aggressive cyber warfare at the largest scale. And this isn't, you know, before we even get talking about AGI and the more visionary ideas about AI. So I think to my mind the analogy is compelling not so much because of the similarity but because of the alarming differences that it reveals.
I mean just to follow up on this in particular, I mean is this an expression perhaps also just of how capital markets have changed in the interim? I mean like if in the 20th century the money was available on capital markets to sort of pursue the atomic arms race in a private capacity. I mean this is what one hears from these AI companies. They need the money. They need this huge investment. They need to scale and this is where the money is. It's private. It's available. That's where it's available. Is there like a version of the arms race where yeah that would have been the case where you know rather than it being the result of a kind of government planning, there was just kind of more private investment?
Yeah. I mean but I think the difference is and this goes on to the broader questions. I mean Joe Weisenthal has been writing about this on Bloomberg right that he's kind of going the other way round. So the Manhattan Project, Wernher von Braun's rockets, the Soviet nuclear weapons program, they grow out of mobilized military states which are taking a huge slice of private GDP, incorporating it into the state war machine, and of course issuing it back out again in contracts to private businesses, but ultimately with a view to getting a rocket or getting a cruise missile, a V1 or V2 or getting an atomic bomb into the state's power, right? That's the project.
And you take taxes and you shrink private consumption and you conscript people and you introduce rationing and then of course you do pump the money back out again to private contractors, but also a state procurement apparatus, a state garrison system to build the weapons. And I guess the kind of question that we could be asking and that is how the entire nuclear program was funded. And the thing about nukes ultimately as the Eisenhower new look of the 50s demonstrated is that they're way cheaper compared to conventional armaments. Right? So if you're fighting the Cold War with conventional armaments, you're up at 10% of GDP plus in terms of economic mobilization. Once they adopt the new look, the American defense budget progressively shrinks from low teens in the early Korean War down to five or 6% and then down to the sort of numbers we're familiar with today, 3 or 4%, even less than 3% at times. So deterrence gets cheaper and nuclear weapons are part of that cheapening process, believe it or not, because per throw weight, per ton of explosive power, they're very, very efficient weapons.
With AI, it's almost as though we're going in the other direction where we have these private businesses which are making these huge technological promises. And as they expand, there's no doubt that there's a national security interest in them. Right now, they just use the flywheel of profit expectations to generate credit. It's not even conceivable at some point the Pentagon wants to take a bigger slice. At that point they become part of the Pentagon's budget which is three, four going on five percent of GDP in a really big year and then they become part of the crowding out kind of logic that we're familiar with with any kind of critique of government expenditure and it depends on whether the economy is at actual full employment or not.
But insofar as we recognize AI as an inflationary pressure, which is where the economists are at now, AI clearly is an inflationary pressure. If it's a national security item and you prioritize it, then you could really be in a situation, it would be entirely unsurprising if this happened, where the Fed said, 'Right, there's inflationary pressure from the AI side of things. We want to lower inflation, so what we're going to do is raise interest rates and squeeze the rest of the economy to make space for AI.' And then in a roundabout, tacit kind of way, we'd be heading towards a low-key version of a Cold War era, unacknowledged or maybe semi-acknowledged, military economic push around AI. We would back our way in tacitly to something that resembled the macroeconomic balancing acts of the 1950s where there's a certain amount of industrial spending that's just got to get done for national security reasons, military-industrial complex demands it and you shape the rest of your macroeconomic policy around that. I think we're a long way away from that.
Right now it would just be a matter of the runaway boom in AI causing some inflationary pressure which the Fed would have to tamp down. But this kind of set of parallels really opens up. It's fascinating.
Yeah, this is a very useful reminder that even though the investment is all private right now, the final arbiter are these governments. The US government is there ultimately as we've seen they are willing to make decisions first of all on national security grounds to not release certain models to certain countries or to not release them outside the United States and then it's easy to imagine broader economic decisions being made for the sake of this national security technology and then drawing resources away from the rest of the economy and directing it towards this particular sector. I guess it's a sort of potential economic disaster all its own that this kind of geopolitical competition could create.
This is just, I mean, it's interesting that you would — I wouldn't sure I would refer to it as a disaster. I would just say it was a tradeoff. You know, if you push the economy to its limits, if you're at full employment, you know, if you're in Virginia and electricity is scarce and expensive, there are trade-offs and you get AI at the expense of other things which do without electricity or power chips or whatever it is that whatever real resources that AI is mobilizing.
And this is how capitalism operates, right? So credit is endogenously generated around these booming sectors because people want in and are willing to lend money to get profit-bearing or interest-bearing shares. And that is essentially incredibly elastic. It's not infinitely elastic probably because people aren't insane, but it's very elastic because people want in on the competition. And then the question becomes what you can actually buy for those tokens, those credit tokens that you've got and those money claims. And that's then a matter of an actual zero-sum tradeoff, which is not a disaster. That's literally just, you know, there's no free lunch when it comes to electricity in Virginia. Either the AI data center uses it or somebody else does. And what will decide that is purchasing power and the AI companies equipped with the resources they're equipped with have a lot of purchasing power. And this is just how growth works. This is Dutch disease if you like of a relatively undramatic and unspecific kind being played out. A rapidly growing sector squeezes the rest of the economy.
Well, I guess I can follow that logic, but I guess I'm wondering whether the tradeoffs, you know, ultimately are structured by political choices, political ideologies. This question of this competition is itself a kind of ideological choice or structure, right? And so you know whether the government itself will put its thumb on the scale of organizing these trade-offs in the sense of what is necessary or not in terms of this competition I guess is a question that comes to mind. You know it's easy to sort of imagine a different set of trade-offs given a different set of geopolitical ideologies.
Well many of us wanted the Green New Deal. Many of us wanted many of us wanted renewable energy to be this thing and the Biden administration will say like that's what we did, the IRA, the Inflation Reduction Act was this sort of thing, you reweave the economy away from fossil towards green at the margin. That means that you're trading off, you're either trading off in the government budget or debt that you're going to have to pay back over the long run or at the limit I don't know construction workers or whatever else happens to be scarce at that moment goes towards the thing you're prioritizing. This is the nature of transformative industrial growth assuming you don't have limitless abundant underemployed resources which thank God America doesn't have right now.
I mean it doesn't, you know, it has a, it's at a relatively high level of employment anyway. Plus, the immigration has been shut down. So, there's not a highly elastic supply of labor from the outside. So, yeah, it's for real. The funny thing about this is like this is what real growth looks like. You know, this is when you're actually changing the structure of the economy, these are the trade-offs. Political priorities absolutely play a key role. Like what does the government, you know, it's been a while since this has happened in America, the last big boom like this was shale right, in the 2010s where you had very elastic credit conditions and money just poured into unconventional oil and gas.
And again you had this combination of political facilitation, the Obama administration was super congenial to shale, it was part of the all of the above energy strategy and it does at the limit it warps the entire economy around this new sector which has all the purchasing power. And as you're saying, Cam, someone has kind of given it the imprimatur of political priority.
