The Indicator from Planet Money - How the bond market is handling AI risks

Episode Date: September 16, 2026

Big tech companies are borrowing billions of dollars to finance the build-out of AI data centers, and your 401(k) might be helping to fuel that. Today on the show, what AI hyperscalers are doing to the bond market.Fact checking by Sierra Juarez.Your Next Listen — Who's financing Meta's massive AI data center?Connect with The Indicator — Sign up for The Indicator’s weekly newsletter! — Buy the Planet Money book — Find our socials, YouTube and more! — For sponsor-free episodes, subscribe to NPR+ Support public media with NPR+ and enjoy perks for over 25 podcasts like this one. This show’s perks include sponsor-free listening. Learn more at plus.npr.org. 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. This is the indicator from Planet Money. I'm Ricky Mulvey. And I'm Waylon Wong. If you own stocks in the U.S., it's very likely that your money is tied up with mega-tech companies. That's because the 10 biggest stocks in the S&P 500 are names like Microsoft, Nvidia, and Amazon. And when these stocks dip, like they did this week on worries about AI spending, they take the market with them. These same companies are starting to reshape the bond market.

Tech giants are borrowing hundreds of billions of dollars to finance AI data centers. If your 401K has money invested in a U.S. bond fund, your savings could be fueling the buildout. A buildout, by the way, that is a dominant force in the economy right now and affecting all of us. This boom in data centers could lead to both productivity and financial gains, but it also exposes new risks, and it all comes back to bonds. Today, we look at how the bond markets are digesting this glut of AI-related business. bonds, and we pose two questions that help us assess whether all of this borrowing is sustainable. Data centers are all over the news these days. They are flashpoints and local debates about what should be built in communities. Now, if you're one of those people who don't want a data center in their

backyard, and that's the majority of Americans, you might want this bond-fueled buildout stopped in its tracks. Then, of course, if you're rooting for the stock market to keep going up, you need those data centers to power the hopes and dreams and profits of the tech giants. But enough about stocks. I want to talk about bonds. Again, why? I always want to talk about bonds because the bond market is like this thing that people never think about.

And yet, bonds have financed so much of human history. Wars, bridges, everything in between. That's it? Just wars and bridges? Yeah, just two things. Wars and bridges. But so many bridges, so many beautiful bridges, Ricky. Airports?

Airports are good, too. But they're not as volatile. They don't go upy-douty as much as stocks. but all right, I'll let you have it. And right now, bonds are funding data centers. Bonds are interesting. That's, I mean, that's what we're always saying to various kinds of receptivity. This is fellow bond geek Zachary Griffiths.

He works at an independent research firm called Credit Sites. His specialty is U.S. investment grade debt that refers to safe and highly rated bonds. U.S. treasuries are investment grade, and so are bonds from megatech companies like Alphabet Amazon, Meta, and Microsoft. These companies, along with Oracle, are considered the main hyperscalers. This is a term we didn't really use that much until a few years ago, and now it's getting thrown around all the time. Yeah, it's come about because it refers to corporations that run huge data centers.

These hulking structures take a lot of money to build, and the hyperscalers have a lot of cash. Still, it's not enough for the kind of massive infrastructure projects they're planning. So they're getting ahead of their future spending needs by borrowing money now. And they're not just waiting gently into the bond market. According to Vanguard, the five hyperscalers sold more than $90 billion in bonds last year. And they've already sold more than $130 billion so far in 2026. Zachary says these hyperscalers have made a splashy entrance.

For the corporate bond market for a buildout, these numbers are, are truly staggering. He says the hyperscalers are doing deals where they sell $25 billion of new bonds in one go. That's a big shift from a few years ago when these companies weren't very active in the bond markets. If you go back just a few years, some of these names that are now at the very top of the issuance or debt outstanding list wouldn't have showed up in the top 10 or 20. And remember, most of the hyperscalers are considered very safe and creditworthy. The bonds are selling are now competing with U.S. treasuries for investor dollars.

That's helping push up yields for treasury. So we talked about that on a recent episode. But investors like hedge funds, pension funds, and insurance companies, they don't have unlimited dollars. So we wanted to know, is there enough demand out there for all of these new AI-related bonds? That's our first question. Is it just this like limitless appetite for this stuff?

Is it like mean girls? Like the limit does not exist? It's funny. The limit does not exist has definitely been thrown. around quite a bit in our office recently. We had a... Ricky, you remember this in the movie Mean Girls, right? When Lindsay Lohan's character clenches a victory for her team at a high school math competition?

Of course I do. And the underrated musical that came out a few years ago on big screens. Great musical. The limit never approaches anything. The limit does not exist. The limit does not exist. Our new state champions, the North Shore, Matt Leach. This is a reference that Zachary's firm understands.

So, is there a limit to how many AI-related bonds investors can buy? The limit may not exist, but what does need to shift is how much spread you're getting or how much yield you're getting. Zachary is talking about the returns that investors are asking for. The higher the risk, the more compensation bondholders want. Here's a recent example. In August, Alphabet sold $25 billion in bonds. That's Google's parent company. Zachary says Alphabet got orders worth more than four times that amount.

That indicates healthy demand. Still, investors wanted to be paid a little bit more in yield. They would lend Alphabet the money, but at a higher price. And the company agreed to that. The Alphabet deal is a great case study and what we think is likely to be the environment going forward. There's demand there, but the pricing is starting to get called into question. So is there investor demand? And yes, but with an asterisk.

The second question we want to ask is, how are bondholders going to be paid back? Some of these bond deals are tied to specific data center projects that are funded through a complex web of private credit firms and legal entities with cryptic names. We did an episode about this kind of financing last year. Jonathan Mondillo helps manage about $170 billion worth of fixed income at a firm called Aberdeen Investments. Each contract seems to be written a little bit different. There's a lack of standardization there, and that's a risk ultimately. One risk Jonathan identifies is construction risk. Will the project get done on time?

And looking further out, there's a risk of technological obsolescence. What if the hypers get it wrong on their forecasts, and they're stuck with data centers they don't actually need? These are long-term lease structures, 15, 20, in some instances, 30-year, lease structures or lease terms. And, you know, who's to say that there's a lot of, need for that amount of space in the data center in 20 years time or in 30 years time. That's why you set up a shell company, then it's not your problem.

Jonathan talks about lease terms because that is a common arrangement for these data centers. Companies like META are planning to rent these buildings rather than own them outright. And this adds another complication. The bonds for some of these projects won't fully pay out for more than 20 years. Jonathan says he's seen deals where the leases come up for renewal before the bonds matured. That's another risk that's been a concern to us. So lease renewal risk ultimately, you know, ties into that technological obsolescence risk. I mean, a lot could happen in 20 or 30 years in the AI world.

Maybe the expected productivity boom from AI doesn't materialize. Maybe some tech companies will quit the data center arms race. Maybe the current fears around AI safety prompt the hypers to pull back their plans, either willingly or because of government regulation. Maybe AI does bring about some kind of existential threat to humanity. All right, that's enough for Anxiety Corner. Just getting started, Ricky. I know. We'll talk to you when we're all paperclips.

A lot could happen in the economy, too. This afternoon, we'll learn whether the Federal Reserve is indeed hiking interest rates. If they do, that will raise the overall cost of borrowing across the economy, including for the tech companies that keep adding to their jobs. debt piles. These are scenarios that investors have to consider when they're deciding whether to lend money to the hyperscalers, money that they might not make back for decades. And again, if you have savings invested in a bond fund, that's your money at work. But for now, people are still showing up to buy these bonds. Tech companies are looking to tap markets outside the U.S.

as well. Alphabet recently sold bonds priced in Australian dollars for the first time. Each new mega deal, however, brings a fresh test of investor appetite. And by the way, Amazon, Microsoft, and Google are NPR sponsors. This episode is produced by Corey Bridges with engineering by Kwee Cuii Lee. It was fact-checked by Sierra Juarez. Kicking Cannon is our show's editor and The Indicator is a production of NPR.