Carlyle Group has warned that private credit firms racing to finance artificial intelligence infrastructure could repeat the credit concentration mistakes seen in the software industry, according to a white paper released on Thursday.

The white paper said the industry may need to provide roughly $1 trillion to finance AI computing infrastructure, a sum equal to more than half of total private credit assets under management.

It stated that failing to set clear limits on concentration in AI computing could prove to be "the biggest mistake."

Mark Jenkins, co-president of Carlyle and head of global credit and insurance, said in an interview: "We are in a period where the revenue model remains uncertain so far. In such an environment, as a credit investor, it is hard for us to say 'okay, we're going all in.'"

Private credit managers are increasingly being asked to finance the massive expansion of AI infrastructure. Capital expenditure related to this is estimated to exceed $5 trillion by 2030. Financing takes many forms, including data center construction and power financing, loans secured by chips that support the technology, and loans to special purpose vehicles.

The white paper noted that, unlike software, the credit risk of data centers and other AI-related assets is more speculative and more likely to be tied to the overall economy, while many of the financing structures now in use remain largely untested.

For Carlyle, this does not mean avoiding AI investment. Jenkins said: "We want to take risk, but we want to take it in a balanced way." He said one of the biggest challenges lenders face is that it remains unclear where AI's eventual profits will accumulate—whether with chipmakers, data centers, or application development companies.

The white paper showed that the software industry went through a similar boom between 2020 and 2022, accounting for about half of private equity deals in that period. Lenders poured into software companies, partly because their recurring subscription revenue was seen as stable and relatively insulated from recessions.

But the rise of generative AI challenged that assumption, exposing software companies to a shared threat of technological obsolescence. Since then, software loans have struggled in the syndicated loan market, borrowers have found refinancing difficult, and some private credit funds have faced increased redemption requests.

Jenkins believes AI infrastructure financing is repeating a similar lesson: what appears to be a diversified set of financing projects may ultimately channel underlying money into a handful of leading companies. He observed that the vast majority of underlying financing in the market is concentrated in seven or eight high-quality targets.

He said it is especially important to understand the ultimate counterparty, the contracts supporting the financing, and the value of the underlying assets. Jenkins said: "As an investor, people need to think very, very carefully about what your counterparty risk exposure is, how the contract terms are written, and what the final asset value is. In a crisis scenario, all of these will matter. And when everything is going well, they seem irrelevant."

RIVERNORTH CAPITAL AND INCOME was also referenced in the market discussion.