Carlyle warns private credit is piling into the same AI data center bet

Carlyle, one of the biggest names in private credit, is telling its own investors that the industry has crowded into a single trade: AI data centers. The firm estimates the buildout could need roughly $1 trillion in financing, more than half of all the money private credit currently manages.

Jason Thomas doesn't work in PR. He's Carlyle's head of global research and investment strategy, and in mid-September he told Bloomberg Television something that should worry anyone who thinks private credit is the safe, boring corner of this AI boom. Lenders' approach to data centers, he said, has a troubling similarity to the model used for mortgages before the 2008 financial crisis. According to Bloomberg's September 15 report, Thomas pointed to a specific mechanism: AI project financing is often underwritten against the investment-grade credit of the sponsor, the hyperscaler or big tech backer, rather than against the actual economics of the data center itself. Nobody's really pricing the risk that the project fails on its own merits, because the loan is really a bet on Microsoft or Meta's balance sheet standing behind it.

That's the same shortcut that let subprime mortgages get bundled and rated as if the underlying house didn't matter, as long as someone upstream looked creditworthy.

Private credit isn't a sideshow anymore. The industry has exploded since the 2008 crisis, growing from a niche alternative into a multitrillion-dollar shadow-banking system that now funds the kind of lending banks used to do before post-crisis capital rules pushed them out of it. These are non-bank lenders, often tied to private equity shops like Carlyle, Blackstone, Apollo, and Ares, and they don't carry the same capital buffers or disclosure requirements banks do. They're also less liquid. If a public bond gets marked down, you see it instantly. A private loan to a data center project sits on a fund's books at whatever valuation the manager assigns, often quarterly, which means trouble can build for months before anyone outside the fund notices.

Carlyle itself isn't a bystander here. The firm was part of the lending syndicate, alongside Blackstone, BlackRock, and DigitalBridge, that put together CoreWeave's $7.5 billion private loan package. So this isn't a competitor sounding an alarm from a safe distance. It's a firm with real money in the trade, telling its own investors the trade has gotten concentrated and the underwriting has gotten sloppy.

The concentration problem compounds the credit problem. A small number of hyperscalers, Microsoft, Amazon, Google, Meta, Oracle, and a handful of AI-native firms like CoreWeave, are driving most of the data center buildout that private credit is financing. The collateral backing a lot of these loans is GPU hardware that Nvidia itself is racing to make obsolete with each new chip generation. Lend against a building and the building is still there in five years. Lend against a rack of Nvidia H100s and you're betting against Nvidia's own product roadmap.

The pattern is getting hard to ignore

Carlyle's warning didn't land in a vacuum. The Bank of England's Financial Policy Committee flagged AI and debt as a top financial stability risk in its report published September 30, warning that interconnected weaknesses in the system could crystallize and specifically citing the surge in AI-related debt issuance. Global AI-related debt issuance hit roughly $450 billion by early September 2026, more than double all of 2025's total, according to the Bank's analysis. JPMorgan has estimated that AI capital spending financed through debt could reach around $4.1 trillion between 2026 and 2030. That's not a rounding error. That's a sum larger than the entire private credit industry currently manages, all aimed at one bet.

Frankly, when the Bank of England, a private equity giant with skin in the game, and a chorus of hedge fund skeptics all start using the same word, concentration, in the same few weeks, it stops looking like noise. It starts looking like the system's own stress gauges flashing at once.

Here's the part that should actually keep founders and investors up at night. Private credit is the plumbing behind a huge share of AI infrastructure financing right now, precisely because banks have pulled back from this kind of lending since the last crisis. If concentration risk in that plumbing materializes, through a hyperscaler pulling back capex, a GPU generation turning last year's collateral into scrap value faster than expected, or simply investors asking for their money back from illiquid funds all at once, the financing spigot for data centers doesn't slow down gently. It seizes. And unlike a public market selloff, you may not see it coming until redemption requests start hitting fund gates.

Thomas isn't predicting a crash. He's pointing at a structure that resembles one that already failed once, and asking whether anyone underwriting these loans today has actually priced that in.

Also read: TSMC Is Weighing a Second US Chip Hub in Texas Worth Over $265 Billion • BMW Will Use AI to Cut a Fifth of Its Senior Managers by Mid-2027 • OpenAI accuses Moonshot AI of running a campaign to steal its model reasoning

This article is posted in AI News, check it out for more related stories.