How growing credit risk could pop the AI bubble
Add Axios as your preferred source to
see more of our stories on Google.
Illustration: Lindsey Bailey/Axios
Big Tech firms are taking on more off-balance-sheet debt — loans which do not appear in standard financial statements — to finance their AI ambitions, says venture capitalist and MIT research fellow Paul Kedrosky.
Why it matters: It remains unclear how that debt will be paid back: AI isn't making money (yet), and with each new chip cycle comes costly upgrades. This financing binge could pop the AI bubble, Kedrosky tells Axios.
What they're saying: Bulls may argue that Big Tech firms have enough free cash flow to finance their AI goals without taking on debt, but "it turns out that's increasingly not the case," Kedrosky says.
- "The Metas, the Apples of the world are all funding" AI, according to Ruth Yang, the global head of private markets analytics at S&P Global Ratings. "Sometimes they put it on their balance sheets, and sometimes they don't."
- That means their debt loads may not always be crystal clear to investors.
- Yang says investors are likely to see more private debt moving forward.
Between the lines: Private credit is difficult to track. Since much of this debt sits in special purpose vehicles, it can be hard to know who ultimately holds the risk: Big Tech, the lenders or the investors in funds that own the loans.
- Data center financing is growing, and Meta is looking to raise $29 billion from private credit giants to finance its AI buildouts.
- Carlyle says $1.8 trillion of capital will be deployed by 2030 to meet AI demand, which private credit will need to facilitate.
Catch up quick: How did we get here?
- After the financial crisis, banks pulled back on financing, allowing private credit to come in and fill lending gaps. Private credit has deep pockets, ones that prefer capital intensive projects. Data centers fit the bill.
Zoom in: AI infrastructure like data centers is very expensive. Over half the costs of a data center buildout is from the chips inside them, Kedrosky says.
- The chips are not a one-time investment. Each time Nvidia releases new, better ones, there's an upgrade cycle, which means continued expenses with no clear returns, given the lack of monetization around AI so far.
Yes, but: "If done correctly, this is part of the innovation of capital markets and how you fund these things," Yang says.
- When she speaks with private credit firms, she finds their funding is not done "willy-nilly, they've put a lot of thought into this. It's a very specific, almost, project finance."
- But even if the AI projects can repay their debt, that does not mean they will generate returns for the Big Tech companies that invested in them.
What we're watching: The AI buildout cycle.
- "I think that there is a risk of overbuilding, and if the economy really is going to weaken, it's going to be a problem," according to Yang.
- A falloff in consumer demand could pressure margins at Big Tech firms on the hook for some of this debt, and snuff out their stock market rally.
