Carlyle warns: Private credit rushes to finance AI infrastructure worth 1 trillion, concentration risk could repeat the software loan crisis

According to information from Zhiyin Finance APP, Carlyle Group has indicated that private credit institutions are competing to provide financing for artificial intelligence (AI) infrastructure construction, which may repeat the concentrated exposure risks previously seen in the software industry’s lending market.

In a white paper released by Carlyle on Thursday, it is noted that the sector may require about 1 trillion US dollars in funding to finance AI computational power infrastructure. This scale amounts to more than half of the current total assets under management in private credit.

The white paper states that failing to set clear limits on concentration in the AI computational power sector could become “the biggest mistake.”

Mark Jenkins, Co-President and Global Head of Credit & Insurance at Carlyle, said in an interview: “We are in a period when revenue models are still uncertain. In such an environment, as credit investors, it’s hard for us to say, ‘Okay, we’re all in.’”

Private credit managers are increasingly being asked to provide financing for the large-scale expansion of AI infrastructure. It is estimated that by 2030, related capital expenditures are expected to exceed 5 trillion US dollars. Financing comes in various forms, including data center construction and power financing, loans collateralized by chips supporting the technology, and loans to special purpose vehicles.

The white paper points out that, unlike software, data centers and other AI-related assets carry more speculative credit risk and are more likely to be correlated with overall economic trends, and many of the financing structures currently in use have not yet been extensively tested.

For Carlyle, this does not mean avoiding AI investments. Jenkins stated: “We want to take on risk, but we want to do it in a balanced way.”

He said one of the biggest challenges facing lenders is that it remains unclear where AI’s ultimate profits will accumulate—whether with chip manufacturers, data centers, or application development companies.

The white paper shows that the software industry experienced a similar boom between 2020 and 2022, accounting for about half of private equity deals during that period. Lenders flocked to software companies, partly because their recurring subscription revenues were considered stable and relatively resistant to economic downturns.

However, the rise of generative AI has challenged this assumption, exposing software companies to the shared threat of technological obsolescence. Since then, software loans have struggled in the syndicated loan market, borrowers have had difficulties refinancing, and some private credit funds have faced increased redemption requests.

Jenkins believes that AI infrastructure financing is repeating similar lessons: what appears to be a diversified set of financing projects on the surface may ultimately funnel funds into just a handful of leading firms. He observed that the vast majority of underlying financing in the market is concentrated on seven or eight premium targets.

He stated that understanding the ultimate counterparties, the contracts supporting the financing, and the underlying asset values is especially important.

Jenkins said: “As investors, people need to think very, very carefully about what your counterparty risk exposures are, how the contract terms are structured, and what the ultimate asset value is. In a crisis scenario, all these things will be critical. When everything is going well, they may seem irrelevant.”