AI Project Debt Mounts, Making Investors Nervous
Jakarta, CNBC Indonesia — The growing pile of debt used to build artificial intelligence (AI) infrastructure is starting to worry investors. The risk could weigh on shares in several AI computing providers, including CoreWeave, Iren and Nebius, amid substantial funding needs to build data centres.
Concerns have intensified after the cost of protecting against a CoreWeave debt default surged. Yields on the company’s bonds have also risen, prompting investors to question the scale of borrowing used to finance data-centre construction.
However, Nebius Chief Revenue Officer Marc Boroditsky said his company was in a different position from CoreWeave.
“Because we do not have the type of debt they have,” Boroditsky said, as quoted by CNBC International on Thursday (22 October 2026).
Despite the concerns, Boroditsky said demand for AI computing remained very strong. He expects all of Nebius’s computing capacity scheduled to come online in 2027 to be taken up by customers. The same could happen in 2028.
Nebius has also recorded rising prices for its AI computing services. Results from an auction of computing capacity held by the company on Tuesday showed that prices per megawatt continued to climb. Asked whether prices would keep rising beyond expectations, Boroditsky replied: “Yes.”
Crusoe chief executive Chase Lochmiller expressed similar optimism. He said rental prices for older-generation Nvidia GPUs, including the A100, were now higher than in 2020 or 2021. H100 prices have also exceeded the level seen when the chips first came into use in 2023.
According to Lochmiller, rising prices reflect strong demand for AI computing as AI models become more capable.
This is helping AI infrastructure companies maximise revenue from GPUs they have already purchased and speed up returns on their investments.
Positive sentiment has also followed SpaceX’s report that revenue from its AI business had increased ninefold to US$7.56 billion.
SpaceX Chief Financial Officer Bret Johnsen said current computing-service prices were slightly above the upper end of the US$30–US$50 per watt range.
Boroditsky said the results confirmed that demand for AI computing still far exceeded available supply.
Nevertheless, strong demand and rising prices have not dispelled investor concerns. The substantial capital needed to buy GPUs and build data centres remains a challenge, particularly for companies relying on debt to fund expansion.