Business

AI-generated text

Investors and Banks Struggle to Sell AI Chip and Data-Center Loans

After a rapid surge in lending for data-center buildouts, some investors and banks are finding it difficult to resell those loan portfolios.

Investors and Banks Struggle to Sell AI Chip and Data-Center Loans

The market for loans tied to AI infrastructure has soured in a hurry. After a rapid burst of lending to fund data-center buildouts, some investors and banks are now finding it difficult to resell those loan portfolios. Multiple reports indicate whole portfolios are being shopped at steep discounts.

According to one source, GPU-backed financings are being particularly hard-hit — described as “getting smoked” — amid worries that the loans’ terms could outlast the useful life of the underlying hardware.

ING and the Stargate exposure

Bloomberg reported on Friday that Dutch bank ING was seeking to sell about $500 million of data-center loans. Included in that package is a $25 million tranche tied to the Stargate project, which involves OpenAI, SoftBank, and Oracle.

The SpaceX deal will be a major test

The next big test for Wall Street is whether Goldman Sachs and Apollo can persuade enough investors to participate in a $40 billion chip financing for SpaceX. That deal would be the largest of its kind, and the firms have begun marketing the package to potential buyers.

Some firms are opting out of GPU financing

A number of prominent players have publicly declined to pursue GPU financings. Michael Arougheti, chief executive officer of Ares Management, said on a recent earnings call that the firm would pass on such deals. “I, for one, personally do not know how to underwrite the decline and depreciation curves on that type of technology,” he said.

Why this matters

The situation highlights a core risk for hardware-heavy AI investments: rapidly advancing technology can erode the value of assets even as loans remain outstanding. If large financings — such as the $40 billion SpaceX package — fail to find buyers, it could tighten financing channels for future AI infrastructure projects.

(Original report by Ellen DiMauro)