Carlyle Warns Private Credit Could Get Too Deep Into AI as $1 Trillion Financing Wave Looms

Jason Thomas, Carlyle Group Inc‘s (NASDAQ:CG) head of global research and investment strategy, told CNBC’s Squawk Box that investors should limit exposure to the AI ecosystem to roughly 15%.

The risks extend beyond data centers to semiconductor manufacturers, energy companies, and other businesses tied to the industry, he explained.

"I think that’s much too much given what’s just been experienced," Thomas said, referring to the prospect of private credit accounting for roughly 40% of its assets under management in AI-related financing.

Private Credit Faces a $1 Trillion AI Bet

The AI infrastructure buildout could require $5 trillion over the next several years, including as much as $1 trillion in private credit financing. While companies’ internal cash flows are expected to cover more than half the investment, the growing reliance on debt raises concerns that lenders could be exposed to businesses disrupted by AI.

Thomas pointed to the recent software stock selloff following new AI tool releases as evidence of how quickly market sentiment can shift. Unlike equity investors, who can offset losses with gains elsewhere, private credit lenders face potentially permanent losses if borrowers default and their collateral fails to cover outstanding debt.

AI Infrastructure Deals Face Questions About Collateral

Thomas also questioned the assumptions underpinning AI infrastructure financing arrangements, including the useful life of GPUs, depreciation schedules, and the ability of unprofitable AI companies to meet long-term contractual commitments.

Some deals rely on residual-value guarantees or other support from semiconductor manufacturers and hyperscalers, intended to give lenders additional protection if projects run into trouble.

But Thomas compared the arguments surrounding these arrangements with those heard before the 2008 financial crisis, when investors were reassured about the safety of complex financial structures.

He said the documentation differs from deal to deal, with lenders potentially holding claims against GPUs, data centers and power infrastructure. Those overlapping claims could complicate restructurings if borrowers default.

Carlyle Sees Risk, Not A Systemic Crisis

Thomas said the concentrated exposure and uncertain collateral values could create problems for individual lenders.

He advises investors to establish exposure limits before committing additional capital, rather than relying on market conditions to signal when an industry has become too crowded.

Thomas said having no AI exposure would also be unreasonable, given the range of potential outcomes and the returns some investments could generate.

The challenge, he said, is balancing those opportunities against the risk that a rapidly expanding industry could expose lenders to risks not yet fully tested.

Photo: Golden Dayz / Shutterstock