The Decibel - If AI is a bubble, how will it pop?

Episode Date: November 26, 2025

Some of the world’s largest tech companies, like OpenAI, Google and Meta, have invested hundreds of billions of dollars into artificial intelligence as they try to build the data centres they need. And right now, a lot of the stock market’s growth is based on AI companies. But what if it’s all a big financial bubble? And if it is, what are the signs it’s about to pop?Globe business reporter Joe Castaldo, who covers AI, explains why markets are twitchy about AI right now and what’s behind investors’ concerns.Questions? Comments? Ideas? Email us at [e-mail] Hosted by Simplecast, an AdsWizz company. See pcm.adswizz.com for information about our collection and use of personal data for advertising.

It's hard to escape all the chatter about AI potentially being one big financial bubble. And there's good reason to worry about that. So much of the stock market's growth is currently tied to AI companies. And the ripple effect of a potential bubble could be wide-ranging. So today, I'm going to try to burst through the bubble talk with my guest Joe Costaldo. Joe writes for the globe's report on business, and he's been reporting about what is actually actually going on with the AI industry. I'm Cheryl Sutherland, and this is the decibel from the Globe and Mail.

Hi, Joe. Thanks for coming back on the show. Thanks for having me. So I feel like everywhere I turn right now, I hear talk about this AI bubble. Why are we even talking about this right now? I think at one level, people just love talking about bubbles. It's just kind of exciting in a weird way. But more to the point, like, there is. There's been talk of an AI bubble pretty much since chat GPT came out at the end of 2022.

Like very shortly after that, you had people who work in AI or in the like financial community just pointing out like, hey, there's a ton of hype around how this technology is going to change the world. It's very expensive to build out. And we don't really know what the financial return from this will be. So those are like kind of all the elements of a financial bubble right there from the get-go. But since that time, like the money pouring into the AI sector to build out data centers specifically is just huge, right? It just keeps getting bigger and bigger. And those questions around, well, what's the return from all this are still there? Like, we still don't know.

And then even more recently, like some of the big AI companies are entering into financial arrangements that are not. novel or unusual. And there are a lot of questions about that as well. So I think just the questions around like, hey, is AI going to pay off or just getting louder and louder and louder. All right. And you mentioned a couple of things that we're going to get into in this episode. Before we dive into what has been happening in the industry that is causing people to toss around the B word, I want to establish the scope of this issue. How much of the economy is riding on AI's growth at this moment? A not insignificant amount. So, if you look at the stock market, there's one study showing that, like, 30 AI-related stocks make up something like 44% of the market cap of the S&P 500.

That sounds like a lot. It is a lot. Yeah. Yes. So there's the stock market aspect. And then because, you know, AI requires data centers, there's a lot of data center construction happening in the U.S., that's impacting the economy there. Again, like up to half of GDP growth in the U.S. in the first six months of the year is tied to AI. Wow.

So it's having a real impact like in the economy and in the stock market as well. And as you mentioned, the industry has seen quite a bit of hype and also a lot of investment. How much money is being poured into AI investments right now? So this year around the world, an estimated 590 billion. dollars, which is a lot. Like, that's more than global oil investment by a few billion. If you adjust for inflation, it's well over the cost of the Apollo project to put a man on the moon more than the, you know, the Marshall project to rebuild Europe after World War II. There's estimates that, you know, the amount of money required to build out data centers and, like, energy infrastructure.

to power those data centers over the next few years is like in the trillions, like $5 trillion. Astronomical. Yes. Yeah. Yeah. And you said you mentioned $590 billion here. That's a lot. But are these companies that are involved in this?

Are they making money? Not really. So, you know, there's different companies in different parts of the AI ecosystem. So, I mean, you know, companies like Google, Microsoft meta, yeah, they're making money. They have very established businesses. Open AI, however, around which so much of the AI ecosystem and hype revolves, no, it's not making money. The company has said, you know, they're on track for $20 billion in revenue this year, which is substantial. But, you know, they've also said their sort of cumulative cash burn out to $209 is in the order of $115 billion.

dollars. And they have made spending commitments to build data center infrastructure totaling like $1.4 trillion in the next few years. So there's a huge gap there in terms of what they're bringing in, what they're committed to spending, when, you know, when they can turn a profit, just a lot of questions around that. Yeah, I think the important point here is that revenue does not equal profit, right? So even if these companies, some of them, are making money, that does not equal a profit in the end. No. And, you know, tech companies can be unprofitable for long periods of time, for sure.

But the scale of this one is really just eye-popping, I think. And if it's not making money internally, it has to, you know, raise external capital. And so far it hasn't had problems doing that. But maybe it does become more problematic. Maybe they can't get the funding they need. Maybe they can't get the funding they need. on the terms that they want. Like Open AI is reportedly considering an IPO next year, which could raise more capital,

but it all remains to be seen. Yeah. So an IPO mean that they would go public? Yes, you could buy shares in OpenAI. So we're talking about revenue and profit, and there's also this thing value, right? How is the value of these companies? How is that different from their revenue? So the end customer.

So like your company that has bought. this AI tool to, you know, automate contracts or write legal contracts or handle customer service inquiries? Like, what is the return from that? Is it making things more efficient? Is it increasing productivity? Is it saving costs? Does it mean they can lay people off? Like, part of the value of AI is like replacing human labor, unfortunately. Right? So like, can companies replace people with this technology? That's the value that we're talking about. And And it's unclear what that is right now because generative AI, it's still new. Companies need time to figure all of this out.

And there's basically conflicting studies right now. Some showing a very small percentage of companies are seeing value. Others showing, you know, a larger percentage. But it's still an unknown. I'm curious, like, how much debt are these companies carrying? Well, it depends. So the big tech companies like meta, Amazon. Microsoft, Oracle, they've been funding this data center expansion primarily with cash from their balance sheets.

That's starting to change because the scale is so large, there's more acknowledgement of risk that they're starting to take on debt, which is a change. Then you have another layer of companies, like there's one called CoreWeave. It buys chips from Nvidia, sets them up in data centers, and like sells access to these chips to companies that want to build. build an AI model, run an AI application. It's a multi-billion dollar company, but it's very expensive to do. Chips are expensive to buy and operate. And so Corrieve has to borrow money to do this. It's taken on a lot of debt.

I think it has like $14 billion in debt. Some of that debt carries really high interest rates. So far this year, it's paid over $800 million in interest. So by the end of the year, it could be a billion dollars in interest. Wow. It's losing money. It's just an example of, hey, this is a really high-cost business, and we're kind of like waiting for the return, right? It has a lot of debt, spending a lot of money, no profit yet to speak of.

Yeah. How are companies able to carry all this debt if they don't have a ton of revenue right now? Because investors are willing to give it to them, because they think AI is going to pay off, right? If you believe that, then maybe Corweave is a safe bet because doing what it does is complicated. You need expertise. Corrieve has it. It has good relationships with all of the companies that need access to these chips like OpenAI. So if you believe we need AI, AI is going to pay off. AI needs lots of data centers.

Then you might want to bet on Corrieve. So how are these companies structuring their debt? Some companies just issue bonds in sort of the ordinary way that companies do. META has chosen a different way recently. So it has a partnership with a private equity company. And together, they formed a separate company that raised $27 billion in debt for a massive data center that META is building in Louisiana. And it's interesting because that debt doesn't show up on META's balance sheet at all. It's like META doesn't have the debt because it's held.

by this third entity that they formed in which META owns a stake. So it's great for META because it doesn't have to carry that debt on its balance sheet. It doesn't affect its credit rating or its ability to raise capital elsewhere. But it's raised a lot of eyebrows because, you know, why are they doing this? Is it so risky that META doesn't want this debt on a balance sheet? It's sort of like an acknowledgement that like, yeah, this is risky. And the company doesn't want to be on the hook necessarily if something goes wrong. And also, when you have these off-balance sheet arrangement, there's just like less transparency around it.

It's harder to figure out what's going on and where the risks are. And given the scale of infrastructure that is believed to be required for AI and the amount of money it will cost, you know, analysts are saying other companies are going to do the same thing, right? They're going to start these, you know, special purpose vehicles to hold debt off their balance sheets. We'll be right back. So, Joe, we've laid out kind of like the hype around AI. And we hear a lot of talk about what's happening in AI right now as being similar to the dot-com bubble of the late 90s and early 2000s. I want to actually sort out how accurate these comparisons are. So let's start with the similarities.

What parallels are people seeing? So a big one has to do with circular financing. Okay. So it's going to take some explanation. I'm ready. Okay. So in September, Invidia, which is, you know, the dominant maker of GPUs, which are the chips necessary for AI, said it would invest up to $100 billion in OpenAI.

as Open AI builds out data center infrastructure over the next few years. And that's interesting because OpenAI is obviously a huge user of Nvidia chips. So it looks like you have a supplier investing in their customer. So the customer has money to buy the product. Now, Invidia has downplayed that and said, this allows us to work more closely together, to develop the technology that Open AI really needs. And plus, like, it's a financial investment, opening eyes of once-in-a-generation company

that's going to generate, like, extraordinary returns. So, like, now we have a piece of it. But it's raised a lot of eyebrows because that's what happened in the dot-com era as well. So in the late 90s, early 2000s, like, the Internet was the hot new thing. And to connect the world, you need infrastructure. So you need, like, fiber optic cable and routers and stuff like that, which, is expensive to build. So what happened is you had companies like Nortel in Canada, which makes the equipment, providing financing to the companies that buy the equipment and

install it, right? So. Hence the circular. Yes. The money just goes around and around. And that's not necessarily economically sound. And what happened in the dot-com era is they just built too much. There was too much fiber optic cable, like way ahead of internet demand. And so the whole thing collapsed, right? Like these carriers that weren't making money relying on vendor financing, you know, they couldn't pay their debt. Some of them went bankrupt. Companies like Nortel saw lower sales because they didn't have carriers to provide vendor financing to to buy their stuff anymore. So the whole structure unraveled.

It's not dissimilar to what's happening now between like NVIDIA and OpenEI. It has another similar agreement with Anthropics. which is a competitor to open AI. Oracle has its own kind of circular deals involving open AI. So there's more than one. But is this just happening with Nvidia and the AI companies that have these kind of large language models like OpenAI? Or are they also doing this with smaller AI companies?

Yeah. So there's a report from Veritas Investment Research in Toronto where they found some 80 investments that Nvidia has made in the past couple years. into a customer or an AI startup that would need compute. And from NVIDIA's perspective, these are just like venture capital investments, right? There's a promising company, let's invest, we'll get a return someday. Veritas has a different view of that.

Like, their argument is like, Nvidia is already such a large company. It doesn't have to be doing this. Its focus should be on revenue growth. So in their view, these investments are more akin to vendor financing. Helping a cash-strap startup obtain funds so they can spend it on Nvidia compute. And what Nvidia has said is like, well, we don't require companies we invest in to use our technology. But the funny thing is they don't have to because there just aren't many alternatives out there. NVIDIA is the only game in town.

Yeah, yeah. So if NVIDIA is having to finance its customers to buy its GPUs, what should we make of the record revenue of $57 billion U.S. dollars it reported last week to investors. Like, how real is that if some of that demand is partially funded by NVIDIA? It's not really known. I think analysts have tried to parse that, like, how much of NVIDIA sales are tied to some of these circular financing deals. But the earnings, you're right, the earnings were great.

The CEO of NVIDIA was saying, you know, oh, there's lots of talk of a bubble. Like, we're not seeing that here. Demand is great. And that's true, it is. But it's interesting that, you know, Invidia, idea of stock is still down from those blockbuster earnings, I think, because investors are asking a lot more questions about AI. Interesting.

Interesting. So we talked about how the AI hype is similar to the dot-com bubble. But what makes the AI situation not like the dot-com bubble? Yeah. There are differences. And for some people, those differences are more important than the similarities. So one is, you know, the biggest buyers of NVIDIA GPUs are companies like Microsoft and meta and Amazon, and they're doing it either with cash or in some cases some debt, right?

But they are existing profitable, hugely profitable companies. They don't need vendor financing for this. So that's one difference. Secondly, the dot-com era had a lot of internet startups that went public that didn't have business plans, didn't have revenue, but had big valuations. Today, AI startups tend to be privately held, so there's arguably less risk in the stock market. And thirdly, is demand to go back to Nvidia's earnings. In the dot-com era, there was too much fiber optic cable. Today, it's like we can't get enough data centers.

There's not a lot of, I would say, objective data on that. We have the company's word for it that, like, you know, demand for GPUs is strong or when a data center comes online, like, all the capacity is used up. So there would seem to be demand today. The kind of flip side to that, though, is if you look at past financial bubbles, that demand can be a bit misleading. So in the dot-com era, companies were still saying things are great, we're building, you know, we have sales, but stock prices had started to decline like well before that because investor sentiment had changed. So today, yeah, demand looks really strong for all of this, but, you know, the past month, stocks tied to AI have been falling. Okay. I want to shift gears here a bit and talk about if there is an AI bubble right now, what would trigger it?

bursting. So I asked that question a lot, and it seems like the simple answer is anything or nothing at all. There's a ton of hype and excitement for AI right now. But if the returns take longer or it's just more challenging than people are assuming, then maybe investors get impatient and they start to sell their shares because like AI stocks have done fantastic. And, you know, everybody starts to cash out and that triggers a, maybe not panic, but just a long slow sell-off until AI starts to deliver. I mean, that's one possibility. And, you know, maybe if Open AI goes public and people get a look at its finances, maybe they're terrified and, like, start to sell. Maybe it has difficulty, you know, raising capital. I think markets are so, like, twitchy right now

that any statement could spook them. Open AI started talking. talking about government backstops for data centers recently, and they tried to walk it back. But I think that scared people because it's like, well, if this is so certain, VA is so great, like, what do you need a government backstop for? So it could be anything. When will we know when it's happening? Stock prices would decline first. And then you might see companies tied to data centers.

Maybe projects get delayed. Maybe they get scaled down. Maybe they're not hitting analyst expectations anymore. Maybe a company like Corweave has trouble making a debt payment, things like that. Aren't stock prices declining right now, though? They are, yes. What do you make of that? So the bubble talk has been relentless probably since late September, October, both in the media and in the financial community.

And even the leaders of big AI companies have acknowledged that, like, yeah, it's possible there could be a bubble. But, you know, they kind of gloss over it and look to the future. So I think because there's so much talk about it, there's just a lot more scrutiny on these financial arrangements and how sound are they? And like, how strong is the demand and what is the payoff of AI? Those questions just have led to doubt and skepticism. and that's being reflected in stock prices. Okay. So if the bubble were to burst, what would the follow-up be?

It could be bad, according to some people. So we mentioned the impact of AI on the stock market, you know, making up a large chunk of the S&P 500. So if there's a bubble that bursts and AI-related stocks decline, like that would hit the stock market in a big way, even if you're not, like, directly invested in meta or whatever, A lot of people own ETFs or index funds, so that would take a hit. And the stock market has an impact on consumer spending as well. Because of AI, the stock market in the U.S. has been doing well, and consumers feel wealthy because like, oh, stocks are doing well. And they're more inclined to spend.

That's the wealth effect. That reverses if there's a correction. So that would hit consumer spending. We mentioned the impact of data center construction on GDP as well. like that would wane, so that would hit the economy. So there are some analysts who say would trigger a recession in the U.S. And Canada, you know, we're somewhat lucky in a sense that a lot of the AI hype is concentrated in the U.S. But as much as we are trying to decouple from the U.S. economy these days, like we're still really tied together.

So there would be some impact here too. Yeah, usually what happens there definitely impacts what's happening here in Canada. Yeah. So just to end here, Joe. We've been talking a little bit in hypotheticals about if there is a bubble and if it pops. But I do have to ask you, the one question we've been kind of dancing around, is AI a bubble? I knew this was coming. I had to ask it.

Of course. My answer probably isn't very satisfying. But I will say if you look at history and what happens anytime there's a new world-changing technology on the scene. like railways, electricity, the internet, there was a bubble. Like, it's just human nature to get really excited, hype things up, look for opportunities to get rich quick, you know, pile in with the herd mentality and drive a bubble. And then it ends in tears. And then in the long run, like, things recover and the technology proves itself out.

I don't see why AI is any different from all of them. those technologies. It has all of the makings for a bubble today. You could argue, like, is there one now? Good one form later. But to think that, you know, markets are going to be perfectly rational about all this just seems unlikely to me. And maybe, if you look at the stock market, maybe some reality is already setting in. Joe, always great to have you on. Thanks again. Thanks. That was Joe Costaldo. A staff report. for the Globe's report on business, who writes about artificial intelligence.

That's it for today. I'm Cheryl Sutherland. Our producers are Madeline White, Michal Stein, and Ali Graham. David Crosby edits the show. Adrian Chung is our senior producer, and Angela Pichenza is our executive editor. Thanks so much for listening, and I'll talk to you soon.