According to Sina Finance, banks that lent billions of dollars to the AI infrastructure boom are seeking to manage their exposure as scrutiny increases following a sweeping Democratic victory in the U.S. midterm elections. In recent weeks, ING Group and Societe Generale have offered for sale up to $800 million in data-center debt, including loans tied to Oracle and OpenAI’s Stargate project.
ING Group is marketing $500 million in loans linked to data-center operators CyrusOne, EdgeConneX and Vantage Data Centers, as well as Oracle-backed Project Jupiter in New Mexico, where securing power remains difficult. Societe Generale plans to sell $300 million in project-finance claims tied to Oracle’s Lighthouse project in Wisconsin and Frontier project in Texas. The three projects obtained more than $56 billion in construction loans from dozens of banks last year.
Banks have been shedding data-center debt for months and are increasingly cautious about exposure concentrated in AI, a small number of large technology companies and frontier AI labs. They are accelerating efforts to clear assets from their balance sheets by year-end amid concerns that U.S. regulators could tighten rules for domestic data centers. Bids for loans tied to the Wisconsin and Texas projects are above 90 cents on the dollar, suggesting portfolio rebalancing rather than serious concern about credit quality. Bids for the New Mexico project are expected in the mid-80-cent range. Oracle recently issued a force majeure notice for the project after delays in power permits.
Banks are also exploring credit insurance and risk-transfer transactions to free up lending capacity. In recent months, they have increasingly used Rule 144A structures, which allow debt to be resold to a broader range of institutional investors. A $63 billion Anthropic debt-financing package and $40 billion in SpaceX financing for AI chip purchases are expected to enter the 144A bond market in the coming months. Some leading lenders have withdrawn from Anthropic’s financing over concerns about concentrated exposure.
Don Dimitrievich, head of global energy infrastructure credit at Nuveen, said the sums involved were unprecedented and had exceeded the capacity of traditional funding channels. He said many financial institutions’ chief risk officers had called for reducing these portfolios. Risk repricing and regulatory capital constraints at some lenders have made the terms and returns on debt for sale more attractive, he added.
David Leung, a New York-based structured finance partner at Freshfields, said banks needed to find ways to keep lending because holding this debt used capital inefficiently. Some data-center loans are becoming harder to hedge because confidentiality provisions restrict third-party transactions requiring loan information, including credit default swaps and synthetic risk transfers. Leung said wider use of such provisions could cause significant problems.
European regulators are also increasing scrutiny of banks’ exposure to data centers and other AI infrastructure. The Bank of England is seeking detailed information from British banks on data-center project loans in routine supervisory reviews. The European Central Bank is preparing an assessment of euro-zone banks’ data-center financing exposure and loan management; the plan was discussed as early as a meeting with bank executives in late October.