Hanna Rosen (2:43)

In the years since, AI has come a long way. It's crept into Our emails, our term papers made its way into medical image analysis. Large data sets that companies use to make projections are late night musings. But one thing it has not done is make money on a scale that even remotely matches these mind boggling investments. So the obvious questions have started coming up more and more. Is this AI boom actually a bubble? A little over a week ago, an OpenAI investor named Brad Gerstner posed the question on everyone's mind about this gap between money going out and coming in.

Sam Altman (3:28)

You know, how can a company with 13 billion in revenues make 1.4 trillion of spend commitments? You know, and you've heard the criticism, Sam. First of all, we're doing well, more revenue than that. Second of all, Brad, if you want to sell your shares, I'll find you a buyer. I just enough. I think there's a lot of people who would love to buy OpenAI shares. I don't think you would. Including myself. Including myself who talk with a lot of like breathless concern about our compute stuff or whatever that would be thrilled.

Hanna Rosen (4:01)

To buy shares because of the amount of money we're talking about here, trillions of dollars in the US economy. Even some people who would not be thrilled to buy shares, who have zero interest in investing in this AI boom might actually have some skin in the game. Whether they want to or not.

Charlie Warzel (4:21)

These are stocks that are not only a big part of the s and P500, but they're also where a lot of retail investors are put in. They're in a lot of, you know, in basically every index fund. So there's this way in which the investment is extremely like woven into the American economy. And you know, even if you are not thinking that you are investing in the AI boom, if you are in index funds, if you are just sort of doing the passive investing that a lot of people are doing, you are in this boom in some way.

Hanna Rosen (4:56)

But what happens if it's true that the boom is a bubble? And what if that bubble is about to burst? I'm Hanna Rosen, this is Radio Atlantic. A huge amount of the US's economic growth is fueled by AI, but it's probably more accurate to say that it's fueled by the promise of AI. The stories people are telling about how AI can make work and workers more efficient, the promise of superintelligence, all the things that we've been seeing in TV and movies for decades.

Charlie Warzel (5:27)

Almost everyone is accustomed to some science fiction image of an ultra smart computer that is sentient, right? Whether it's the Terminator, I'll be back, whether it's Hal from 2001, I'm sorry.

Hanna Rosen (5:41)

Dave, I'm afraid I can't do that.

Charlie Warzel (5:45)

There's all kinds of preconceptions of this technology.

Hanna Rosen (5:49)

But as we'll hear from Charlie, so far, none of the grand promises have really come to pass. And the bubble just keeps getting bigger.

Charlie Warzel (5:59)

There's a recent McKinsey report that's been sort of passed around in these spheres where people are talking about this that said 80% of the companies they surveyed that were using AI discovered that the technology had no real, they said, significant impact on their bottom line, right? So there's this notion these tools are not yet, at least as they exist now, as transformative, as people are saying, and especially as transformative for productivity and efficiency and the stuff that leads to higher revenues. But there's also these other reasons. The AI boom, in a lot of ways, is a data center boom. For this technology to grow, for it to get more powerful, for it to serve people better, it needs to have these data centers which help the large language models process faster, which help them train better. And these data centers are these big warehouses that have to be built, right? There's tons of square footage. They run on these. They take a lot of electricity to run. One of the problems is with this is it's incredibly money intensive to build these, right? They're spending tons of money to build out these data centers. So there's this notion that there's never enough, right? We're going to need to keep building data centers. We're going to need to increase the amount of power, right? And so what you have basically, is this really interesting infrastructure problem on top of what we're thinking of as a technological problem. And that's a bit of the reason why people are concerned about the bubble, because it's not just like we need a bunch of smart people in a room to push the boundaries of this technology or we need to put a lot of money into software development. This is almost like reverse terraforming the Earth. We need to blanket the earth in these data centers in order to, you know, make this go.

Hanna Rosen (7:50)

Right? It's a. It's real estate, okay? So there's this huge amount of investment required, but everybody's kind of known this calculus for a while. So why are there suddenly headlines about it being a bubble? Like, why did that start. Why did that worry start to intensify in the last few weeks?

Charlie Warzel (8:07)

Well, I think a lot of it is seeing the numbers just continue to go up, right? Like when you start to See things like you have AI expenditures accounting for 92% of GDP growth during the first half of 2025. Right. That is a pretty wild number. But I think also you have this real cultural sentiment right now that a lot of people lived through the dot com boom and they see something that, that maybe feels a little bit similar.

Hanna Rosen (8:44)

Right?

Charlie Warzel (8:44)

The Internet was this paradigm shift of a technological innovation. It was very clear to people that the Internet was going to affect commerce, it was going to change a lot of things. And you had all these companies that came out, the famous pets.com example, as sort of like the height of the bubble, which was like, people can buy pet food online. The dynamic of the dot com crash wasn't that the Internet was a fad. It was that all this investment came in and all these companies were created and the ecosystem wasn't yet developed enough to support it. Right. Like people buy dog food online all the time now. Pets.com was an idea ahead of its time. And I think that a lot of people are looking at all this funding, all this money, seeing it kind of rhyme a little bit with the dot com boom, and also thinking maybe some of these companies are a little ahead of their time. Is it going to pay out soon enough or is there going to be this idea that this was kind of a hype cycle? There was this hysteria. And that's where you see this feeling of, oh no, we are betting so much. We are putting so many eggs in this basket. What if it doesn't pay off soon enough? There's an investor named Harris Kupperman and he's bearish on all of this stuff. And he has said in order for these companies to break even on what they are spending in 2025 on capital expenditure, they would need to generate $160 billion of revenue. That is to break even on what they're investing this year in these data centers. The revenues are nowhere near $160 billion in terms of what the generative AI tools are, are generating. He went out with this blog post and then a bunch of people who are in the industry who run data centers, who are other investors in this and know the nuts and bolts of it. They told him his estimate was way off and that the revenue that these tech companies would need to generate to break even on this year's capital expenditure alone is closer to 320 or $480 billion. That is what these companies need to generate this year in order to break even on what they've invested. We're dealing with a lot of Back of the envelope math on all of this, but that gives you a sense of kind of the dizzying numbers that we're talking about here.

Hanna Rosen (11:22)

Okay, Charlie, the more you talk, the less this makes sense. So, so what's the plan? Like When Sam Altman, CEO of OpenAI, was asked about this enormous amount of spending, he got somewhat defensive. So do other CEOs own investors outline the plan? Like, what's the good news scenario by which this all works out?

Charlie Warzel (11:43)

Well, I'm really glad that you said the more that I talk, the less sense it makes. Because I just want to say that is entirely indicative of my experience reporting on this. I come to this, you know, I don't have an economics and finance background. I write about technology. I, I have looked into these things and spoken with people and I come away sort of with this thousand yard stare. And I think that is a bit of what's going on here when people look into this stuff. It's not that there isn't maybe some magical, you know, save here, right? Like it's entirely possible that there are these leaps in this technology, right. The Bloomberg columnist Matt Levine has this great line that says we are going to create God and then ask it for money, right? That is what these AI companies are sort of saying, or what Sam Altman.

Hanna Rosen (12:34)

Said, which was we'll just build a super intelligence and then ask it to solve this math problem for us.

Charlie Warzel (12:39)

That's essentially the. Yeah, that's the same thing. And that's why it's a little staggering, right, that there is just no one is blinking when it comes to the money that's being invested. Like the lights are all green here in terms of all of these tech companies. People are plowing ahead this narrative of the AI bubble right now. It definitely exists in media, it definitely exists in like the tech analyst space. But like Wall Street's not stopping on this at all. If anything, there's like a FOMO that is happening right now. There is a notion of we need to get in on this, right? This is a race. Like not only is it a race from company to company, but it's also got these geopolitical components to it, right? Like there is a desire to, on the American end to want to beat China. There is this feeling like there is no time like the present. There's the money is too good right now, we're not going to stop. But when you drill down into it, there's this notion that essentially what's happening is people are lighting money on fire in the hopes that this Thing just continues to turn into the promise that everyone's hoping.

Hanna Rosen (13:54)

After the break. The case for boom, not bubble. How the AI companies say they actually plan to make money. I find it hard to believe that it's this level of magical thinking like that. The real answer people have in their heads is superintelligence. Like, what does that mean and what does it have to do with making money? Like, even if they did build a superintelligence, how would that be a return on investment?

Charlie Warzel (14:34)

It's a good question. I mean, the idea of superintelligence is completely changing the paradigm of what humans are capable of doing. Right. It changes the ways that every single hedge fund will be able to do predictive modeling in terms of scientific discovery. Right. If you're a pharmaceutical company, a super intelligence that creates other super intelligences is sort of this exponential amount of research and brain power that hopefully can come up with a miracle cancer drug that then you can then go out and sell.

Hanna Rosen (15:09)

That's real. They are connected.

Charlie Warzel (15:10)

There are real ideas here with this. The problem is right now we have. You've played around with ChatGPT, right? Is that worth rewiring the entire world and global economy around that? I not sure. Right. And that's where we get this, I think, real unease and confusion with people.

Hanna Rosen (15:32)

Yeah. It does also feel to me that the difference between this and some of the other speculative technology moments in history is that everyone sees it and everyone's talking about it. That was not true of the housing bubble, for example. It was a fairly elite group of people who could see that one coming. But it feels like a lot of people are talking about this and no one really has an answer.

Charlie Warzel (15:54)

I think that the best cases for we are not in a bubble right now. From the AI industry that I've heard, the first is everyone says we're in a bubble and everyone's watching it. It's just what you said, right? Like there are so many eyes on this thing that how can it get so out of hand? Right. This is happening sort of in public, so all the exposure is there. People know what they're getting into. We're talking about this all the time. That's the first one.

Hanna Rosen (16:25)

Interesting.

Charlie Warzel (16:26)

The second one is the notion that there is this physical component, right? This infrastructural component, this data center component. There's only so many construction workers and construction companies. There's only so many chips. There is only so much land. You can only get these things up and operating quick enough and that all of that physical infrastructure will moderate Some of the investment and some of the spend, right? This notion that we can't scale up as fast as the companies want, and that will be a moderating influence. It will allow the technology to sort of meet the expectations a little bit better. The gap will dwindle. But the concern, though, is there are dynamics at play here that are starting to get a little bit weird and complicated in a 2008 financial crisis way. There's a couple complicated dynamics. There's just the notion of if the tech stocks fall, there's a lot of exposure there. There's a lot of contagion in the sense of highly leveraged hedge funds are invested in. They could be forced into fire sales, which could cause this vicious cycle with financial damage in pension funds, mutual funds, insurance companies, everyday investors, right? It's just. It's bad news for anyone who's trying to play it safe. But then there's this other component that we found in our reporting that really, it kind of made the hair on the back of my neck stand up a little bit. It's this notion of debt investing. So essentially, building these data centers is very expensive, and these tech firms don't necessarily want to take on that debt, right? They don't want to ask for loans to build these data centers because it looks bad on their balance sheets. And, you know, they're worried about shareholder returns. So to get around this, some of these tech companies are partnering with private equity and they're doing this financial engineering, right? The private equity firms will put up money or they'll raise money to build a data center, and then the company, the tech company will pay the private equity firm through the rent that they get renting out the data center. So what is happening is you're having a company repackaging their data center leases into a financial instrument, basically a bond that people can sell. And there are examples of this happening, and then they're combining that with others into these securities that are based off of tranches, that are based off of how risky they are in terms of default. And if any of those words sound familiar to anyone listening, it's because that's a little bit similar to some of the financial engineering of the 2008 crisis, right? With subprime mortgages. It's not the same thing. Data centers are not subprime mortgages at all. These tech companies are blue chip companies that make a lot of money. But there is this really interesting financial engineering situation that is going on that starting to just, you know, it's getting overly complicated, and it's getting harder to know where some of this money's going. What feels telling to me about this is less the idea of, you know, of tranches and repackaging and more this idea of that the money is not going in a way that is very easy to track.

Hanna Rosen (20:03)

Right. But I guess at the same time, like you said earlier, we're sort of smarter than we were in 2008. So we can see a lot of these dynamics, and maybe that makes some kind of a difference. One thing that strikes me as possible is a lose, lose situation where your one option is AI is a bubble and it bursts and we're in trouble for the reasons you described. And the other possibility is that it succeeds in some spectacular way and then unemployment skyrockets as a result. What do you think about that dilemma?

Charlie Warzel (20:34)

I see this a lot. This is the thing that I think about quite a bit.

Hanna Rosen (20:40)

Right.

Charlie Warzel (20:40)

It seems to me like you there aren't a lot of amazing options at the moment. Because the way that I look at this is that the amount of investment is so historic, it is so massive in this technology that the people who are pouring this money into it are expecting something huge.

Hanna Rosen (21:06)

Right.

Charlie Warzel (21:07)

Like maybe the money's not rational. Maybe there is this hype, maybe there is this, you know, it's just driven by FOMO and all kinds of, you know, whatever. But these companies also, they're not full of idiots, right? They have this notion of what the technology could do sometime down the line. And I think the only way that these investments make sense is that they are paradigm shifting for society in the way that the Internet wired us all, connected us all, and sort of weirdened the world. I think that's what they're hoping that artificial intelligence does. And we can already see that it has had a disruptive impact, even in its more rudimentary form. So it does frighten me to imagine what is enough to satisfy that investment. I think it would be a massive amount of job loss or just of reorganization and kind of chaos.

Hanna Rosen (22:13)

Mm. It sounds like you're saying it's chaos either way. Like there's a bubble that brings pain, like the kind of pain we had in the financial crisis, or a true breakthrough that brings disruption and possibly mass unemployment.

Charlie Warzel (22:28)

Yes, I think that absolutely we are in a position where if these companies create the thing that is worthy of this much investment, it is going to cause an extreme amount of societal rewiring and probably financial pain. And if they don't manage to pull it off, I think it's going to cause extreme amount of Financial pain, so.

Hanna Rosen (22:50)

Pretty much bad either way. But I guess there's a scenario where we could just live inside the bubble or this fear of the bubble for the next 10 years, but it never actually crashes.

Charlie Warzel (23:03)

There is a really interesting, I guess, debate or controversy, however you want to say it, around the notion of will these companies be put in a too big to fail bucket, right? There was a executive at OpenAI who mentioned in an interview something about the United States government backstopping the company. If there was a bubble and they had to walk that back, there was this big controversy. They said, no, no, we're not expecting a bailout for no reason. We want the free market do its thing, et cetera, et cetera. But there is this really interesting thought experiment here, right, with this amount of investment. And that's why I mentioned earlier there's these geopolitical stakes. Donald Trump, to the extent that he knows anything about artificial intelligence, knows that he doesn't want China to have anything resembling a super intelligence before the United States, right? There have been these data center partnerships through the White House, right, that have been facilitated by that Donald Trump, one of the first things he did was bring Sam Altman to the White House and announced this Stargate investment partnership, right, with OpenAI. So there is this potential notion of maybe these investments can't be made, allowed to fail. Maybe it is seen as this geopolitical imperative that it becomes very clear that the government's not going to allow these companies to fail or change the parameters of it or underwrite some of the losses in some way. And I think that is going to be a very interesting thing if it ever did come to pass. Because these companies have acted like they are building the Second Coming. And if all of a sudden they ended up in a situation where they had to get a bailout, I think there would be an intense cultural backlash to that.

Hanna Rosen (25:05)

Thanks again to Charlie Warzell. Starting Friday, November 14th, you can listen to Charlie host his own show. It's a weekly video podcast called Galaxy Brain where Charlie is going to talk to people who know a lot about our hectic information ecosystem. And Charlie is going to try and give us back our sanity. This episode of Radio Atlantic was produced by Janae West. It was edited by Kevin Townsend. Rob Smirciak engineered and provided original music. Genevieve Finn, fact checked. Claudina Baid is the executive producer of Atlantic Audio, and Andrea Valdez is our managing editor. Listeners, if you enjoy the show, you can support our work and the work of all Atlantic journalists when you subscribe to the Atlantic at theatlantic.com/listener. I'm Hanna Rosen. Thanks for listening.