Hyperscale my debt!

Unhedged

The episode examines how AI hyperscalers like Amazon, Meta, and Alphabet are executing a historic borrowing spree to fund data center construction, fundamentally reshaping global corporate bond markets. Hosts Katie Martin

Key takeaways

• Hyperscalers are expected to borrow approximately $1 trillion by 2030 to finance AI data center buildouts, shifting from cash-rich firms to massive debt issuers.
• Goldman Sachs reported ~$500 billion lent to AI-related groups this year, with hyperscalers accounting for roughly $200 billion of that total.

Main topics

• AI hyperscaler bond issuance
• Corporate vs sovereign bond spreads

Notable quotes

"The death robots. Yeah, you know, that one."

Conclusion

The episode provides a clear framework for understanding how AI infrastructure spending is

Transcript preview

Speaker 3 (0:00) Heaven knows there is a lot going on in markets at the moment. The AI companies are saying they might accidentally kill us all. Whoopsie. But they want our money anyway. Meanwhile, the mighty global government bond market has been thrown into a mincing machine and there's no end in sight. But the cool kids know there's more to markets even than that. And the biggest thing that people are not talking about enough is all going down in the bit of finance where companies get their hands on money to play with. The corporate bond market is one for the hipsters, honestly, and it's not usually the sexy stuff. But right now it's getting a huge revamp because US big tech is borrowing big bucks and rewiring the market in the process. So today on the show, hype, hyperscalers, and how this is all changing the game for companies everywhere. This is Unhedged from the Financial Times. I'm Katie Martin, a markets columnist in the FT dungeon in a weirdly warm and sunny London. Joining me down the line from New York City is my usual partner in crime, the big man, Rob Armstrong, who has not yet had his morning bagel. Rob, say hello. Speaker 1 (1:16) Hello. My stomach is rumbly, but otherwise I feel OK. If Speaker 3 (1:20) you hear a rumbling noise, listeners, that's what it is. But also with me in the studio, making her Unhedged debut, she's overworked and over here, it's Emily Herbert from our markets reporting team. Emily, how are you? I'm well, thanks, Katie. Yeah, I'm delighted you've carved out time to come and speak to us because you do the hard work, you and Ian Smith. So my top tip to you, if you're new to the Unhedged podcast, Emily, is just don't take any nonsense from Rob. Just remember that and you'll be fine. Perfect. Yeah. So... Listeners, if you listen to this humble podcast, you'll be at least vaguely familiar with the government bond market. If you're not, then clearly Rob and I are doing something wrong. But it's kind of the same deal for companies. If they want money... to buy another company or build a factory or whatever it is, then they can get a loan from a bank or they can go cap in hand and borrow the dosh from investors, generally big professional investment firms rather than normie retail investors. This is a really important part of the financial system and it's getting a big glow up at the hands of hyperscalers. So, Emily, you've been writing about this recently. First off, what is a hyperscaler? What are we talking about here? Speaker 2 (2:34) Yeah, so the hyperscalers are the kind of big companies that we talk about a lot when it comes to AI. And they're the ones which are investing in the kind of physical infrastructure of this. They're kind of building the data centers to house all of this cloud computing capacity. So it's companies like Amazon, it's Alphabet, it's Meta, it's Oracle, it's those ones. What it is not at this point is the... AI labs like OpenAI and Anthropic, which are building the models themselves. So they're already like Speaker 3 (3:06) companies that we're very familiar with from other things. And Speaker 2 (3:08) they're public companies. They're listed on the stock market and they can borrow money in the corporate bond market. So Speaker 3 (3:14) the labs are like these huge money furnaces where money goes to die. They're just like burning it on like creating AI widgets, whereas these are companies that are generally like... dripping with money already, aren't they? Speaker 2 (3:26) Yeah, they have so much money. And I guess the labs, we'll know more about the labs probably soon when they are public companies. Speaker 3 (3:34) And I mean, we should take a slight detour here. Like, Rob, are we allowed to call this AI now? Or does your glorious leader, he's told us we have to call it SI, right? Speaker 1 (3:45) Once again, Katie, I will not take responsibility for the US president. We are allowed to call it AI on this show, if not elsewhere in America. We're safe here in the studio. Speaker 3 (3:56) Donald Trump says there's nothing artificial about it. It's actually super intelligence. Anyway, we're going to call it AI. So for ages, like these massive tech companies, when they wanted to do AI build-ups, these companies got money coming out of their ears, right, just from their daily course of business. So they didn't need to borrow any money to do it. They could just find money down the back of the sofa and build a data center for a few million dollars. So why are they borrowing money to do it now? Speaker 2 (4:22) What's that about? Basically, they need more money. I mean, you know, we've all promised to do things and then realized we don't actually have enough money. Right. And so they have gone to the bond market and they're expected to borrow a very, very, very large amount of money over the next few years to do this. Do we have Speaker 1 (4:43) even a ballpark estimate of what that number is? Speaker 2 (4:46) Yeah. So, so far this year. Goldman Sachs said to us that investors have lent AI-related groups about $500 billion. The hyperscalers, they account for about $200 billion of that. But for the hyperscalers, over the next few years to about 2030, we think it's going to be about a trillion dollars of debt. Ooh. So, you know, we've all been there. Speaker 3 (5:11) A