Will US debt burst the AI bubble? With Ruchir Sharma

0:00Everyone's looking for the pin that could pop the AI bubble. And if you think this is just a spectator sport for Silicon Valley insiders, well, think again. The tech giants driving this AI boom dominate global stock markets. And if they stumble, the shockwaves will rip straight through your pension, your savings and your investments. But what if the pin that brings this all crashing down has nothing to do with AI at all? What if that pin is actually America's enormous, ballooning national debt? The AI boom needs vast amounts of borrowed money. The problem? Well, so does the US government, and increasingly they're competing for the same pool of cash.

0:42So could America's debt problem be the thing that finally pops the AI bubble?

0:52Hello and welcome to The Economic Show. I'm Chris Giles, the FT's economics commentator, standing in for Samaya Keynes. And to talk us through how the AI boom and US debt are perhaps fatally connected, my guest is Rishi Sharma, chair of Rockefeller International. His latest book is What Went Wrong with Capitalism. Welcome to the show, Rishi.

1:15Ruchir Sharma: Thanks, Chris. Great to be back with you again. So we always start the economic show with a question asking you to rate something on a scale of 1 to 10. On that scale, how close do you think we are to the bursting of the AI bubble, where one is not at all and 10 is it's a highly pressing near-term threat? Well, I'd say on that scale, we're pretty much at seven to eight, I would say, Chris. The first thing, of course, is that we have to first agree that this is a bubble or not, right? Because when it comes to bubbles, there is no scientific definition. By some accounts, a bubble is seen to be something which crashes by 40 % at least.

1:57Ruchir Sharma: But how we know that we are at the top of that summit before that crash comes, there's no scientific definition of this. So what I had done in the past and also written about is to identify that what are the major signs that you need to first watch for to identify a bubble. And in that regard, I came up with something called the four O's. The first being overvaluation, the second being over ownership, the third being over investment, and the fourth being over leverage. Those are the four O's I said. If you look at the history of bubbles going back hundreds of years, that you typically find that these are the four characteristics which are present when we do see a bubble.

2:44Okay, let's go to the four O's in your AI bubble. which of them are present? Right.

2:50Ruchir Sharma: So first, if you look at overvaluation, that you look at the valuation of AI stocks, or you just look at the broad valuation of the US stock market, what you find is that the US stock market in general, but the AI stocks in particular, are about as expensive as they have ever been in history. Right. From a valuation perspective, we are somewhere in what would be defined as the 90th to 100th percentile. So I'm going to put a tick by that one. How about the other three? Right. So if we go to the other three, the other is over-ownership. And here, I think that in America, again, the amount of equity ownership we have seen is quite unprecedented.

3:32Ruchir Sharma: In fact, this is for the first time that I know of any country in history where its population owns more of its wealth in the stock market than it owns in the property market. So this is quite unprecedented how exposed the American public is to the stock market. And a large chunk of that is towards AI plays, the hyperscalers and other sort of AI plays. About 40 % or so are direct AI plays in the US stock market. So from an ownership perspective, you can argue that this is really a new high. Okay, I'll put a tick by that one as well. And if briefly you could go through the other two. Yes. So the other one I speak is overinvestment.

4:18Ruchir Sharma: Now, here again, the acceleration that we have seen in investment on the AI front, if you measure it as tech investment as a share of GDP, or just the acceleration in overall AI investment in the economy, that is pretty sizable. So those ratios now are, again, approaching the levels that we have seen in past very big cycles, the housing cycle in 2006, 2007 in the US, the commodity cycle globally that we saw at the end of that decade. So we're pretty much approaching those kind of levels where the spend on one particular thing as a share of GDP is somewhere around 5 % of GDP and some aspects even bigger.

5:02Ruchir Sharma: So I'd say on these three O's, overvaluation, over ownership, and over investment, I'd put a tick. And on leverage, does it come here too? This is where the difference gets interesting. If you look at the amount of leverage that the AI companies, the hyperscalers in particular, have today, that yes, the debt is increasing quite sharply, that these companies used to be flush with cash and now they've become net debt companies. So that is changing pretty quickly for the likes of Amazon and Meta. But overall, for companies of that size, the amount of debt that they're carrying relative to their profits or relative to the cash that they are generating is still relatively low.

5:46Ruchir Sharma: So from a pure leverage standpoint, I would say that the balance sheets, at least of these big AI companies and the AI ecosystem is not that leveraged, even though it's changing very quickly. And maybe they'll get there in the next year or two at the pace they are currently borrowing. So that's a tick in pretty much all four boxes. And as we'll unpick in this episode, the reason you believe that the AI bubble is close to bursting is because of what's happening to US public debt, national debt, the amount the government owns. Let's try and put it into some perspective. First of all, how big is the US national debt?

6:25Ruchir Sharma: The number now has approached 100 % of GDP. And by some metrics, if you just take the overall gross debt, it is well over 100 % of GDP already. What does this mean, right? Because as you know, Chris, that people have been worrying about the U.S. debt for a long period of time, ever really since America ran persistent budget deficits from the 1970s. And I think that here's some historical perspective might be in order. Since the 1970s, except for four or five years, the U.S., for example, has run a budget deficit every single year since then. And so obviously, when that started to happen in the 1970s, people got concerned.

7:08Ruchir Sharma: In the 1980s, the concerns really ramped up. And as is famously known, that in Times Square in New York, a real estate developer even set up this clock to monitor the rapid pace at which the US debt had been increasing. So these concerns have been there for a while. Okay, so what that clock shows at the moment is that US government debt has hit in the last month or so$40 trillion. That's not necessarily the best figure because some of that is government money owed to other bits of government. Don't need to get into that. It's all quite complicated. But about$32 trillion is, I think, what most economists would say is the real level of government debt that's owed to other people outside government.

7:58And as you said, that's 100 % of GDP. So that's the same size as the US economy produces every year. So I think that's the best comparison. In comparison with GDP, how much has that risen, let's say, this century?

8:13Ruchir Sharma: Yeah. So at the top of the century, that number was closer to about 37 % of GDP or so, So we've gone from 37 % of GDP to 100 % of GDP. So that's the scale of increase which has happened. But I think that the most important thing here is, why is this a concern now? Technically, these numbers have sounded large for a long period of time, that even at 10 trillion, that number sounded very big back then. So why is this number today at 40 trillion, much more concerning? And that's because I think there are two or three other things that have come together at the same point in time, which is what's concerning.

8:56Ruchir Sharma: One is that the amount of interest that the U.S. is paying every year now to service this debt is more than 3 % of GDP, or nearly a trillion dollars a year. America spends about$800 billion or so on defense. So now it is paying much more on servicing its debt through interest payments than it is spending on its mighty military every year. The only thing on which America is spending more is Social Security every year. But at nearly a trillion dollars, it's now beginning to even overtake how much it spends on Medicare. So this is how much America is spending on servicing its debt every year just through interest payments.

9:41Ruchir Sharma: And it's rising because the average interest on the debt lags, right? Because the government takes debt with an average maturity of five to seven years or so. But now, as interest rates have been rising and some of the debt rolls off, this share is going to keep going up. So that's the point, which is that the amount that America will soon be spending on just interest will be larger than any other component of its budget. Yes, that's an amazing figure, how much money the government is spending on servicing the debt. One of my favorite numbers is that at the start of this century,$1 in every 10 that the government raised in taxation went on debt servicing is now one in five.

10:27So the government is turning more and more into just a debt servicing machine. But now we've been talking a lot about debt. But why has the amount of interest on the debt been rising this year? At the start of the year, mid-February, on average for 10-year borrowing, the US government had to pay up just under 4%. Now it's about 4.8%. Why is that?

10:55Ruchir Sharma: Well, there's a couple of reasons, Chris. I think that they've all come together. One, of course, has been that you've seen a big increase in oil prices, and that is contributing more to inflation as well. So you've seen sort of inflation go up on the back of higher oil prices. And I think that this is one of those mysteries about the market, which is hard to understand, which is that the market all of a sudden has turned its attention to now the debt after ignoring it for a long period of time. And that leads to this also this expectation cycle where people begin to get more concerned about the debt, which leads to higher interest rates.

11:31Ruchir Sharma: And those higher interest rates in turn further add to the debt pressure. So you have more and more bondholders asking for greater compensation to be paid for the fact that inflation is relatively high, oil prices are staying high, and also this vicious cycle is sort of developing where people get concerned more about the debt. And so the interest payment goes up because the yield on government debt goes up. And that leads to further what is about the sustainability of the debt. Because the government's interest rate, in a way, sets the floor for interest rates that everyone has to pay. But once that begins to go higher, then interest on everything else goes higher.

12:14Ruchir Sharma: Mortgage rates, the rate at which companies can borrow. So everyone has to pay more. And so it's that increase in borrowing costs across the economy which begins to have an effect and impact on everyone. Okay, so we know that there's an increase in borrowing costs for government, and that is going to go across the whole economy because government sits at the centre of financial markets. Let's talk about why this matters for tech companies. How exactly does higher yielding government debt travel down the economic pipeline to hit a tech startup or a Silicon Valley giant or a tech company investing billions in data centres?

12:59Ruchir Sharma: So if you look at the math that the AI companies today have to grapple with, the amount that they are spending every year on building out the AI infrastructure is around a trillion dollars or so. That's the global spend. If you look at how much companies like these are spending on building out the infrastructure needed to power AI. So data centers obviously being the biggest component of it, but it's about a trillion dollars. Now, by the most generous estimates, the amount that the AI companies are earning in revenue from the AI services that they're providing is about$200 billion or so. So you have a gap every year, a financing gap of at least$800 billion, right?

13:43Ruchir Sharma: So that's the gap which is there. Now, how do you fund this gap? So part of this$800 billion or so gap is funded from the companies at the heart of the AI build-out who have cash coming from other businesses. But then you still have a big gap which is left. And that has to be financed through two ways. One, by more borrowing from the market or those companies issuing more of their stock. That's the way that that gap has to be filled. So the AI story, which was like a very big technology story, has now increasingly become a capital market story, a financial market story, which is that these AI companies are reliant on the capital markets to plug this gap.

14:31Ruchir Sharma: And not all of it can be financed from these companies' internal cash flow. That's why they need to come to the market to borrow. Okay, so if they have to borrow, let's use that 800 billion number, how does that compare with the amount that the US government is also trying to borrow every year? Right. So the government's budget deficit in the US is around 6 % of GDP, at least, because if you take some of the state's deficits and stuff, it tends to be a bit higher than that, which means about$2 trillion of borrowing that the government has to do. So that's$2 ,000 billion, just so that people understand the differences.

15:12Ruchir Sharma: Yes. And the AI companies can borrow from the market. They can also do equity issuance. But that number is also rising very fast, and they're all competing for the same savings, whether it's household savings or corporate savings. And this is the big difference from the past. And so in the early days of this boom, the belief was that big tech was so rich, so flush with cash, that they could easily self-fund this out of profits. That's no longer true. Yes, because the arms race now, the AI arms race is just in a different orbit. So that's the big change which has happened, that the excitement about AI is great.

15:53Ruchir Sharma: But it's also the fact that the investment needs have gone up also dramatically as people have to build out more and spend more on compute and other AI infrastructure. So we know that the borrowing needs from the markets is going up very fast. It's very high numbers in competition with the US government. How high does the benchmark rate of interest on US government debt? 10-year debt is the benchmark. How high does that have to get before you think big tech is going to struggle to fund its AI ambitions? Yeah, so I did some work on this. And I think that the number I came up with was that if the 10-year yield decisively breaks above 5%, that could be quite problematic.

16:43Ruchir Sharma: Now, how did I come up with that number? There are two or three factors here. First is just, you know, this is a market sense that if you look at the 10-year interest rates on government debt, those have been capped at 5 % or so for the last 20 years. They've gone down to below half a percent. But in recent years, whenever they've tried to get to about 5%, that's been the cap. But if it breaks above 5%, in market terms, that's like entering a new regime. that something has shifted, that that 5%, which used to be the cap on interest rates, has now been broken. We are above 5 % in nominal terms, this is.

17:26So we're at 4.8 % right now, roughly. How can you be so sure? And why specifically does the 5 % number matter for a private tech giant?

17:37Ruchir Sharma: I think there are some very fundamental reasons why getting above 5 % also matters. One is that if it persists that the rate of interest is above 5%, then the interest rate that the government is paying on its debt then gets higher than the nominal GDP growth rate of the economy. It's a very economic basic equation that sort of explains when does the debt start to get more unsustainable. It's typically when the rate of interest gets higher than what the nominal GDP growth rate in the economy is. And 5 % nominal GDP growth rate for the US economy is what's expected for the foreseeable future. And if the interest begins to stick up a 5%, the interest rate on that, that's a problem.

18:25What exactly is the mechanism for tech companies maybe struggling when government borrowing costs go above 5 %?

18:34Ruchir Sharma: The government's borrowing costs typically set the floor for what the interest rates are in the economy and from the interest rate at which other firms can borrow. So if you begin to get interest rates on the government debt, which are prohibitively high, in a way that ends up attracting more capital to that government debt as well. And so therefore the AI companies then will struggle to compete to raise money when that is being sucked away by the government more and more. I think that that is the mechanism through which AI companies struggle to keep borrowing at the pace at which they currently are.

19:17Ruchir Sharma: And so the AI companies then will have to borrow at rates which are higher than that. And also then they're competing with the savings getting channeled in that direction rather than in their own direction. And in general, the sense that the government's finances are in trouble, I think, ends up causing risk appetite for investors to get much lower. OK, so the key issue here is competition between the tech companies and the government. Let's take a break. And when we come back, Rishia, I want to ask you when you see the AI bubble bursting.

20:28Thank you.

20:37And we're back. Roushia, we've been discussing your argument that skyrocketing U.S. national debt is on a collision course with the AI boom. Your argument is that the ultimate tripwire for this boom is the yield on 10-year U.S. Treasury bonds, which is currently sitting at about 4.8%. And once it hits 5%, everything could come crashing down. So how close are we to it going over 5%, the tipping point in your view?

21:06Ruchir Sharma: Well, I think the US government knows the sensitivity about breaking up a 5 % and sustaining there. The US Treasury Secretary has been doing whatever he can to jawbone the market, to keep yields lower, to increase the buybacks. So I think that they're going to try their best to make sure that you don't get higher interest rates. Now, the competition here is really on two fronts, which is the fact that the hope is that before yields decisively break above 5%, that you begin to see productivity improvements happen on the AI front. And those productivity improvements raise economic growth so much that America is able to grow itself out of this hole that it has.

21:53Ruchir Sharma: I think that's the real hope out there, that the productivity boom from AI is so big that it leads to a huge pickup in economic growth and that America is able to grow itself out of trouble. It's a plausible scenario, just that we don't have much evidence as yet because overall productivity numbers in America have gone up a bit, but nothing very significant can be pinned down to the AI boom. So it needs to happen very quickly. And the second thing is inflation, that we need to see some of these inflationary pressures ease so that investors don't demand higher interest rates and the Fed, which controls short-term interest rates, doesn't feel compelled to increase interest rates in the near future, which pushes interest rates across the curve higher.

22:37Ruchir Sharma: So I think that those are the two things to watch out for, whether that happens very quickly. Otherwise, the natural path for interest rates is higher just because of the arguments we have made, that the financing costs are increasing, and we are getting very quickly stuck in this vicious cycle. Now, as I said, that decisively breaking above 5 % and staying there, the longer we stay there, the higher the probability that this AI boom comes to an end, partly because of what history shows us. And in this regard, I've looked at booms and bubbles going back the last 300 years or so, And every single bubble in history has ended one way, which is it's been a monetary event, that you're going to tightening in monetary conditions, which in the modern era of central banking, going back 100 years, means higher interest rates, short-term rates.

23:31Ruchir Sharma: But even before that, when you didn't have properly defined interest rates from central banks, some monetary event, like in the 1870s, the gold standard causing that, or even before that some monetary event has always led to the end of bubbles. This time, the amount of interest rate increases needed to prick the bubble is less just because, as we said, the debt stock is much higher today in general in the US economy led by government debt. So the key thing that the authorities, the government, the central bank needs to do is get good growth, which government doesn't want that, and low inflation.

24:09Again, which government doesn't want that? What specific measures would you like to see either the Fed or the US administration put in place to really to minimize the risks that this bubble is going to burst in a difficult and unpleasant way?

24:27Ruchir Sharma: Well, the obvious one is you would like to do something to create the sense that this budget deficit is going to be put under some check or under some control because the path today of least resistance is that the deficit will keep going up. Because imagine we're in the midst of an economic boom, and yet you're running a budget deficit of 6 % of GDP. Imagine what happens if you get a downturn. In an average downturn, the budget deficit in the US typically goes up by 4 percentage points. So we're looking at a budget deficit if you get even a mild downturn of nearly 10 % of GDP. So something has to be done to put an end to those kind of expectations that the budget deficit will be brought under some control.

25:11Ruchir Sharma: I just don't think that there's enough attention to that, partly because we're in the midst of this bubble. Everyone's too busy partying. Why bother changing anything? So it's only for a few people, us at the margin, who are worrying about the U.S. budget deficit and yields going higher. But the tech bulls, as you yourself point out, are saying that it all doesn't matter, that we have had such high interest rates in the past, the growth rates are so incredible that we'll be able to grow itself out of this hole pretty quickly. So that's their argument. But I still feel that doing something to draw the line of the sand on the budget deficit is something which is urgently required.

25:54Okay, so raising taxes, cutting spending, all these things are very difficult for governments to do. Have you got a final piece of advice to investors who are watching the markets and watching the AI bubble with some sort of anxiety? What would you suggest they do?

26:13Ruchir Sharma: Well, I'd say that you have to diversify, which is the idea that a lot of people, one with the other, have a lot of exposure to AI in their portfolio. And it's not just through the equity market, whereas you know that the concentration has never been higher. And then there are also increasingly a large amount of debt, which investors are buying, happens to be of AI companies because they've become the largest issuers of debt as well. So it's about how do you diversify away from that? And this may initially lead to some disappointing returns, just because anything outside of AI has not been doing that well.

26:50Ruchir Sharma: So I think the whole issue here is that you need to diversify and be aware that the amount of exposure you have to the AI trade today is pretty large, just given the very concentrated nature of most indices in the US. And on that encouraging note that there are some diversifications that ordinary investors can make, Rushir, thanks very much for coming on to the show. Thanks, Chris.

27:25That's all from this episode of The Economic Show. The producer is Michel Frankel Duval. The executive producer was Manuela Saragossa. The broadcast engineer is Andrew Georgiades. Flo Phillips is the FT's head of audio. Samaya Keynes is back next week. Thanks for listening.

27:55The New Episode of The Next Five is about industrial decarbonisation. Dino Otranto, CEO at Fortescue Metals, joins me. The single-handed reliance on oil and gas is over. As does former World Bank CFO Bertrand Padre. There is more money going into the world than into decarbonising cement or steel. And you refers Director General Axel Eggert. The steel industry needs to be decarbonized by 90 % by 2035. Listen to The Next Five wherever you get your podcasts. Enjoy. The new episode of The Next Five podcast is all about sovereignty, data and control. Charlie Giancarlo at Everpure joins me. You always get a very embarrassed response if you ask them if they know where all their data is.

28:37As does Professor Sana Karaghani. There isn't any modern AI system that's entirely sovereign. And Claire Schramm-Fergus at Lloyd's Banking Group. We really need to design for safety rather than bolting it on at the end. Listen to The Next Five wherever you get your podcasts. Enjoy.