The Current - Will the A.I. bubble burst?

Episode Date: October 24, 2025

There are growing concerns from economists, tech industry insiders and investors that artificial intelligence might be a bubble about to burst. Data centres are a rapidly growing part of the U.S. so big that some observers like MIT fellow Paul Kedrosky believe it's warping the North American economy. Murad Hemmadi, a reporter with the Logic, argues that we only know about bubbles in hindsight, and until it bursts, we're going to be waiting to find out.

This ascent isn't for everyone. You need grit to climb this high this often. You've got to be an underdog that always over-delivers. You've got to be 6,500 hospital staff, 1,000 doctors, all doing so much with so little. You've got to be Scarborough. Defined by our uphill battle and always striving towards new heights. And you can help us keep climbing.

Donate at lovescarbro.cairbo. This is a CBC podcast. Hello, I'm Matt Galloway, and this is the current podcast. I think it's quite possible. Most other major infrastructure buildups and history, whether it's railroads or fiber for the internet in the dot-com bubble, these things were all chasing something that ended up being fundamentally very valuable. The infrastructure gets built out, people take on too much debt,

then a lot of the companies end up going out of business. That's the founder of Meta, Mark Zuckerberg, speculating on the possibility, of an artificial intelligence bubble. This is a hop topic among those watching AI companies balloon in size. One company in particular, NVIDIA, has an evaluation of $4 trillion. The company's market capitalization is now bigger than the entire stock market capitalizations of Britain, France, and Germany. Nvidia supplies much of the world's graphic processing units known as GPUs.

AI companies need these GPUs to power things like ChatGPT. And with more AI companies, comes more need for more GPUs, but are they meeting demand, or are they helping to pump up a bubble that is ready to perhaps burst? Paul Kodroski is a research fellow at MIT's Institute for the Digital Economy. He studies artificial intelligence, economic disruption, and the future of work. Paul, good morning. Hey, Matt, how are you? I'm well, thanks. It was interesting.

Forbes magazine asked seven chatbots if there is an AI bubble. Grock said yes. ChatGPT said yes and no. Is there an AI bubble? Yeah, of course there is. I mean, it's the only way you get out of saying that there's an AI bubble, if you redefine bubble to be something completely opposite from what we have historically, which is to say the rapid appreciation in price, a couple of key assets, you mentioned Nvidia, but I just

say more broadly the seven largest AI-related companies, which are euphemistically called the Meg 7, the Magnificent 7, which are now something like 30% of the S&P 500, half of the S&P 500's gains over the last four years. AI is now in excess of 50%, building on data centers, that is, is something in excess of 50% of GDP growth in the United States in the first half of this year. I mean, I can go on and on, but the numbers are just staggering in terms of both the growth and the size of this, of this one sector and its contribution to the economy that's so far in excess of anything so, you know, mundane as profits or anything else, which is, which is, I suppose, as Mark said earlier, is, you know, not that unusual at this stage of a build-out of something like AI or the dot-com bubble, but on the other hand, the wreckage on the other side is going to be extreme. All right. We'll talk about the wreckage in a moment. But right now, I mean, you see the stock market soaring. You talked about the magnificent seven. These

companies that are apparently with extraordinary amounts of money. How is that warping the broader U.S. economy? Oh, in a bunch of ways. The one I point to most often is I sort of alluded to earlier, which is that almost half, a little more than half, depending on how you do the math, I guess, of GDP growth in the U.S. is tied to spending on these things called data centers, which are these giant facilities where you end up parking all of these NVIDIA GPUs for the purposes of training AI or responding, to these prompts to chatbots, but it's much more insidious than that. Because of this

confounding nature of how large it is, what's actually happening is that it confuses trade policy, and you were doing it the spot before me talking about tariffs. So the contribution of AI data centers is so large that if you think that the reason why the U.S. GDP is growing as quickly as it is, is because of tariff policy, you're confused because at least half of recent GDP growth is related to AI data centers, and while tariffs or any other economic policy, my name, has a role, the outsized contribution of the spending on AI gets in the way of actually having effective policy because you're confusing the effects of one thing with the effects of another, and that means that you're actually the policies you're enacting aren't having the effects

you want. So it has all of these broader and narrow or insidious effects that, for whatever reason, largely just go sailing past people. What do you make of the fact that, I mean, there's an argument, being made by some people that even if this is a bubble, all of this money or a lot of this money is going into a good place. It's helping to build new companies and helping to build this technology that people keep saying, and maybe they have a vested interest in this, but they keep saying that it's going to change our world. Yeah, as sort of almost a religious mantra from the technology community. So historically, there's an element of truth to that, obviously, you know, the railroad

build out in the 19th century, while railroads eventually became useful, fiber optic build out in the 2000 period, well, we used that to stream Netflix and so on. The trouble is in the interim between when you build those things out and they become useful, many, many business fail, many, many businesses fail, many people lose a great deal of money as markets reverse. Almost every major capital expenditure bubble you can think of back to the 18th century in canals was followed by a financial crisis. So the notion that you can wave your arms and say, you know, one day this will all be useful and glossed right over the period after the bubble when markets, Markets have a crisis, businesses fail, people lose significant amounts of stock markets investment.

This is incredibly glib, but as I said, it's a mantra from the technology community that you hear over and over and over again to justify not caring about overspending and the eventual collapse of many of the things on the other side. Those of us of a certain age will remember the dot-com bubble. And I mean, everything was dot pets.com, whatever, shelf.com. many of those companies evaporated. Is this different than that, do you think? Is it just larger than that? No, it's not different at all. It's exactly the same.

As a matter of fact, one of the more dangerous aspects of this particular bubble is that unlike in prior ones, you'll hear people saying things like, well, we now know that in the aftermath of these kinds of bubbles that everything works out. And what you get then is this kind of circular justification for spending even more than we would have in the past because we've convinced ourselves that it really doesn't matter because by the magic of, you know, so Shumpitarian capitalism, everything's going to be okay. And that's fine. But as, you know, the other line I like to point to is in the long run, we're all dead, this has, we have to have time for all of these things to work out. And you can't just, you know, wave your arms and

say, eventually these GPUs in the, or eventually these prompt engines or eventually all of the training will do and will work out. And in the interim, all of these data centers we're building, it looks increasingly like we're probably going to be vastly oversupplied by data centers, which are populated by these GPUs. And so, you know, you're going to see massive amounts of those data centers essentially become white elephants. As we're talking, the U.S. President posts the stock market is stronger than ever before because of tariffs. I'll let you go, but how do you expect this to end? If this is a bubble, what's going to happen?

The way they always do with a bad landing on a runway and a market decline. and many, many business failures, unfortunately. A crash or something less dramatic than that? Oh, no. It's almost inevitably the crash. My model for this, but sadly, is something closer to the 1920s where we have a market crisis and then it depends what policymakers do after that to tell you whether or not becomes much deeper than that.

Yikes. Paul, thank you very much. Sure, you're welcome. Paul Kodroski is a research fellow at MIT's Institute for the Digital Economy. He studies artificial intelligence, economic disruption, and the future of work. has a daily newsletter that you can find at Paul Kodroski.com. Now, not everyone thinks these big investments, as we were saying,

will add up to an AI bubble. Here's Michael and Trader, the CEO of CoreWeave, an AI cloud computing company in the United States, speaking on CNBC International earlier this week. When you step back and you realize that it's Microsoft and Open AI and Google and AWS, coming in, purchasing the infrastructure because they need to serve their client.

that's not really what a bubble is going to look like. When you talk about a bubble, you're talking about a systemic misallocation of money. And I do not believe that when the largest companies in the world are serving compute to their clients, that is what a bubble looks like. Marad Hamadi is a reporter who covers artificial intelligence for the logic and has been looking into this. Marad, good morning to you. Good morning. How many people in the tech space, like Michael and Trader, are skeptical that what this industry is seeing the enormous amounts of money being shoveled around here that this will lead to a bubble i think the idea that this is a bubble um you know paul stole a lot of my lines this morning um but i think the uh the the counter argument to this is essentially um what we just

heard which is that uh the companies that are using this you mentioned pet got bats dot com before this is not these are not the pets dot com of uh you know 2025 We're talking about very large companies that have stable, gigantic revenue bases, you know, Amazon, Microsoft, Oracle, Google. Now, those revenue bases come from doing things other than AI, but AI is starting to add revenue to those revenue bases, whether that's better advertising or maybe more targeted advertising, depending on how you look at it, whether it's selling AI services to businesses and to consumers.

So the biggest flashes in the pan of the dot-com era were trading really on their names, whereas here there is actual revenue happening. There's also questions about how this infrastructure gets used, and Paul mentioned the fiber experience of the dot-com era. It is true that once you build a data center, you can use that data center today, and you could probably use it in two years, and you'd probably use it in three years. so they may be white elephants for a little while, but they might get used. But that's part of the thing, is one of the reasons why people are concerned is this is, you know, evaluation based on promise.

It's not based on something that's happening right now, but on this idea that at some point in time, perhaps in the not so distant future, this will be the future. And I just wonder whether that's dangerous to make those assumptions based on potential that has yet to be fully realized. I think potential that has yet to be realized is an interesting way to look at it. There's the very long term, which is, I mean, AI God, right? It's artificial general intelligence, some idea of AI systems that are smarter than humans and can do all kinds of things, including cure disease and fixed climate change and whatever, you know, what have you. It doesn't actually have to get to that point for AI to become a technology that's very widely used. I mean, why do we have data centers today?

because the internet sort of permeates everything we do. It's conceivable that AI could become a technology that changes economies and societies without ever reaching the furthest goals that these AI companies are pushing for. And while it's certainly true that the rollout has not been at all perfect or uniform, it is happening. If we think of this as a bubble, I mean, Paul talked about the fact that this is. going to end, as it always ends, with a crash, that there will be carnage at the end of it. Let's talk about what that will mean here in this country. The logic reported in July that, I mean, we're speaking about the magnificent seven. In Canada, it's called the Maple 8, the largest pension funds in this country. CPP, the Ontario Teachers Pension Plan, have something

like 6.6 billion U.S. dollars invested in data centers and digital infrastructure firms listed on the stock market in the United States. What does that mean in terms of risk posed to Canadians. And they probably have several billion more invested in actual data centers because several of those Maple 8 have private equity arms that own data center companies or data center assets or the debt of data centers. I mean, not to trivialize this, but 6.6 billion is not actually that much in the scope of the Canadian pension fund. Fair enough. These are gigantic, gigantic institutions. And they are diversified. it's undoubtedly true that they would take a hit if this happened.

And not just them, but, you know, if you own an exchange traded fund or some kind of passive investing instrument that tracks the stock market, well, to Paul's point, these AI companies are an increasing part of the stock market. So you probably have a lot of personal exposure to it, too, in your own trading accounts. So those, you know, that is a significant hit that people will take. That's undoubtedly true. I have to let you go, but just very briefly, what are you going to be watching for. It feels like the conversation has taken a turn in the last couple of weeks, certainly,

around this. What are you going to be watching for as this continues? I think I'm going to be watching to see how quickly companies like Open AI and Google start to see that revenue grow. It's certainly going quite quickly. But if AI revenue continues to go from businesses and consumers, there will be the need for the data centers, at least some of the data center capacity that they're building. Marad, we'll leave it there. Thank you very much. Thank you so much. is a reporter who covers artificial intelligence for the logic. This has been the current podcast.

You can hear our show Monday to Friday on CBC Radio 1 at 8.30 a.m. at all time zones. You can also listen online at cbc.ca.ca slash the current or on the CBC Listen app or wherever you get your podcasts. My name is Matt Galloway. Thanks for listening. For more CBC podcasts, go to cbc.ca.ca slash podcasts.