Banks and investors are growing more cautious about debt tied to the artificial intelligence buildout following a market selloff, with financing backed by computing chips facing particular scrutiny. The shift is testing lenders’ willingness to fund equipment that could lose economic value before the loans used to buy it are repaid.
Some investors are seeking buyers for entire loan portfolios at steep discounts, Semafor reported, citing a person familiar with the market. That person described particular pressure on financing for graphics processing units, or GPUs, because of concerns that the debt could remain outstanding beyond the technology’s useful life.
The retreat is also reaching bank-held data-center debt. Bloomberg reported that Dutch lender ING was looking to sell $500 million of data-center loans, including a $25 million position linked to Stargate, the project involving OpenAI, SoftBank and Oracle. Those figures describe loans being offered for sale, rather than completed transactions.
A larger financing effort could provide another test of investor demand. Goldman Sachs and Apollo have begun marketing a $40 billion chip-financing deal for SpaceX, Semafor reported. Whether investors commit to that transaction will offer a measure of appetite for large-scale exposure to AI computing equipment as concerns about existing debt intensify.
The central challenge is how to value the equipment over the full term of a loan. GPUs are used extensively to train and run AI systems, but their financial value depends on more than whether they remain operational. Newer chips can offer better performance or lower operating costs, potentially making older equipment less competitive even while it continues to function.
That distinction matters for credit underwriting. Lenders financing equipment must assess both the borrower’s ability to make payments and the value of assets they could recover if payments stop. If collateral depreciates more quickly than expected, a lender may face a larger loss following a default. The timing of principal repayments, the loan’s maturity and the assumptions used to estimate resale value can therefore be important to the transaction’s risk.
Discounted loan sales do not, by themselves, establish that borrowers have defaulted. Loans can trade below their original value when investors demand higher returns, reassess credit risks or seek to reduce exposure. Nevertheless, weaker secondary-market pricing can complicate new financing by giving prospective buyers alternative investments at lower prices.
Some major investors are choosing not to take on the valuation challenge. On a recent earnings call, Ares Management Chief Executive Michael Arougheti said the firm would pass on GPU financing, explaining that he did not know how to underwrite the technology’s depreciation curve. His stated concern was the ability to assess how the equipment’s value would decline—a question that sits at the heart of lending against rapidly changing hardware.