The Indicator from Planet Money - OpenAI's deals are looking a little frothy

Episode Date: October 16, 2025

There have been many headline-grabbing AI deals recently: Nvidia investing up to $100 billion in OpenAI. OpenAI promising to buy $300 billion worth of computing power from Oracle. Oracle buying tons of chips from Nvidia. But … where’s the money coming from? Is all this AI overhype … a bubble? On today's show, how money flows in the AI hyperscaling flood. Related episodes: Is AI overrated? Is AI underrated? The messy human drama behind OpenAI For sponsor-free episodes of The Indicator from Planet Money, subscribe to Planet Money+ via Apple Podcasts or at plus.npr.org. Fact-checking by Sierra Juarez. Music by Drop Electric. Find us: TikTok, Instagram, Facebook, Newsletter.  See pcm.adswizz.com for information about our collection and use of personal data for sponsorship and to manage your podcast sponsorship preferences.NPR Privacy Policy

NPR. There has been a whirlwind of gargantuan AI deals recently, often involving OpenAI. Oh, yes, the softly spoken wizard of chat GPT, Sam Altman, has been busy. There was this $300 billion deal with the database and cloud company Oracle. Basically, Open AI said it would buy $300 billion worth of computing power from Oracle. But how it OpenAI pay for that, it's unclear. Yeah, we're going to get into that. We'll also talk about Nvidia.

It came in soon after, saying it would invest up to a hundred billion dollars in OpenAI. It's all starting to get a little heady. $100 billion here, $300 billion there. Invita paying Open AI, who pays Oracle, who pays Nvidia. We're going to have to break it down. This is the indicator from Planet Money. I'm Terriam Woods. And I'm Waylon Wong.

Today on the show, OpenAI's deals. We look at the AI data center boom and how the world's most valuable startup is trying to win the AI race. To untangle all the open AI deals, we're going to start from the ground floor. And what's happening on the ground floor is acres and acres and acres of data center construction, from Las Vegas to northern Virginia. According to the Financial Times, around 1.2% of America's economic output is going to Amazon, Microsoft, Mehta, and Google building data centers. That's roughly $1,000 for every American. The data center buildout is happening at an incredible rate right now. Gilluria is the head of technology research at D.A. Davidson, an investment firm.

Microsoft, Amazon, Google, the largest companies in the land with almost infinite resources can't build data centers fast enough. You need land. You need access to power. You need chips. You need electricians and HVAC engineers. But they have the resources, they have the capital, they have the money. And so they're doing that to catch up to how good all these AI tools are getting.

So we can all use SORA to create silly little 10-second clips of cats giving us advice or raccoons fighting with each other and driving away. I can just see that in my alley, IRL late at night. But more seriously, AI can be used. for reviewing legal files, doing a lot of computer coding grunt work, or speeding up medical research. Every few weeks, the many AI models keep hitting new benchmarks. It feels kind of like the early days of search engines when we had Altavista, Ask Jeeves. Do you remember any more? Oh, yeah, like Yahoo, MSN, search.

Yeah, yeah, yeah, yeah. So eventually a search engine came out that was better than all the others, and it could stay better, at least partially because it had so many users giving it data. That was Google, of course. Now, this winner-takes-all dynamic may or may not apply to AI, but companies are acting as if it does, trying to become the best AI company as if the best AI system is going to get most of the cash. And that puts Open AI in a tough situation. They are a small, scrappy upstart competing in a space. Sport of Kings, in order to participate in the AI market, you need tens of billions, if not

hundreds of billions of dollars. Gill says that Microsoft has that money. So does Amazon, Google, Apple, meta, and Elon Musk. And so Open AI needs to raise a lot of capital very fast in order to have access to the compute that will keep it competitive. And that brings us to all those weird seeming deals that Open AI has been making, the $300 billion promise to Oracle, the $100 billion invidia is investing in open AI. And to think about whether these kinds of deals make sense,

Giloria says we have to make a distinction with all the AI data center investment. There's a lot of stuff that's real. The AI tools are fantastic. There's a lot of companies willing to pay for access to those AI tools. There is also a part that is an inflated demand. environment. That sounds like a euphemism to me, an inflated demand environment. I'm trying to avoid using the word bubble. Why use one word when you can use three? He's not getting paid by the word, to be clear. But seriously, the word bubble is kind of a loaded term in economics. It's a situation

where the price of an asset is way above what it's really worth, and it's followed by a panic and a crash. But there's a debate over whether you can truly identify a bubble while it's happening, especially when the future is so unclear, like in technology. Galeria isn't using the B word, but he does think the companies might be pumping up more excitement than is warranted. Nvidia, Oracle, OpenAI, AMD are engaged in this exercise of funding each other that creates the impression of demand that's even greater than it really is. When Open AI made a $300 billion commitment to Oracle, it didn't have that capital. It won't have that capital.

That was artificial. It's not real. It's inflated. That's not to say Open AI is inflated. Open AI has phenomenal success and great products and great product design. it's just that's writing a lot of checks that it really can't cover any time soon. Yeah, where would it get that $300 billion from? Well, good point.

They're going to have to raise all this capital mostly in the debt markets. So they're going to have to get hundreds of billions of dollars of debt to finance a startup. That's never happened and it's unlikely to happen at that scale. Again, Open AI will be successful. They have great products. They'll continue to grow. But right now, they're losing $10 billion a year. I've heard Sam Altman say that making a profit is not even one of his top 10 concerns.

Yes. And that's understandable because startup companies rarely focus on profits first. They focus on getting as many users as they can. And that's the mode open AI's in. So yes, they'll continue to grow. but they just need to continue to fund their losses. They can't make the type of commitments they're making and then live up to them.

So that part of the story is inflated. If Gill is right and the recent open AI deals are inflated... I mean, even Sam Oatman told the Virgin August that he thought investors were getting over-excited. Right. And it does raise the question about what could happen to the rest of the economy. These kinds of deals are not just limited to Open AI, Meta, X-AI, Anthropic, these are other companies raising money in atypical ways right now.

And like it or not, if you have a retirement account, a decent chunk of your money is invested in companies like Nvidia, Microsoft, Amazon, Google, and Oracle. And this web of deals ties their face together to some extent. So the question is, would Open AI reneging on its commitments, bring down the rest of the economy? The glass half-empty view is that AI won't be as transatlantic. formative as promised, and that this could bring down companies and investors in some kind of chain reaction. We'll just be left with these unused data centers humming all around the country to the chagrin

of locals who are already pushing back at all this construction. It would vanquish the one clear bright spot in an economy suffering from slow jobs growth and tariff uncertainty. It would be the inflated demand bursting. We're not using the word bubble. Yes. And that burst demand could trigger a recession. No, don't say recession.

Well, there's good news. Gil Luria doesn't actually share those concerns. The AI buildout will continue. It's just a matter of what parts of it are healthy and what parts of it are unhealthy. The healthy part will continue. The chats and the video generation and the voice interaction will make us all more productive. We just need to get through the unhealthy parts in order to focus on the great. benefits that are going to accrue to us from artificial intelligence. Lots of railroad companies went under after the railroad booms and busts in the 1800s. Yet we can still ride those tracks today. Exciting new technology can be a bumpy ride.

And if you want to hear more about those fights between locals and AI data center construction, we have an episode coming out soon on that very topic. Stay tuned. This episode was produced by Angel Carreras with engineering by Jimmy Keely. was fact-checked by Sierra Juarez. Kate Cancannon edits the show and The Indicator is a production of NPR.