Large banks that have financed the rapid expansion of artificial intelligence infrastructure are actively seeking to reduce their exposure to data centre debt amid heightened regulatory scrutiny and concerns over concentrated risk. This shift comes in the wake of the recent US midterm elections, which have intensified apprehensions about potential regulatory clampdowns on data centre operations.

Financial institutions including ING Bank and Société Générale have launched efforts to sell off substantial portions of their data centre loan portfolios. Sources familiar with the transactions indicate that ING is offering up to $500 million in loans related to major data centre operators such as CyrusOne, EdgeConneX, and Vantage Data Centres, as well as Oracle-backed Project Jupiter in New Mexico, which has experienced delays in securing necessary electricity permits. Société Générale has simultaneously sought buyers for approximately $300 million in project finance debt tied to Oracle’s Project Lighthouse in Wisconsin and Project Frontier in Texas. These three projects collectively raised over $56 billion in construction debt from numerous lenders last year.

The sales, ongoing for several months, reflect banks’ growing caution over their concentrated lending exposure to a small number of Big Tech companies and AI frontier labs. Market participants suggest that pricing for loans connected to the Wisconsin and Texas projects remains relatively stable—quoted above 90 cents on the dollar—indicating the sales are largely a strategic portfolio adjustment rather than indicative of pressing credit deteriorations. By contrast, loans related to the New Mexico project, impacted by power permit delays and a recent force majeure declaration by Oracle, are likely to trade at discounts in the mid-80s cents, reflecting higher risk.

In addition to outright loan sales, banks are exploring alternative risk management strategies including credit insurance and risk-transfer agreements to extend their lending capacity while maintaining capital efficiency. There is also a notable trend toward structuring data centre and AI-related debt within the 144A bond format, facilitating wider institutional investor participation. For instance, financing packages related to Anthropic’s AI operations and SpaceX’s planned AI chip acquisitions, valued at $63 billion and $40 billion respectively, are slated to enter the 144A market soon. However, some lenders have reportedly exited the Anthropic financing due to apprehensions about overconcentration of risk.

Industry experts highlight the unprecedented scale of capital deployed in AI infrastructure financing, which has exceeded the risk appetite of many traditional lenders. Don Dimitrievic, global head of energy infrastructure credit at Nuveen, noted that senior credit officers are urging portfolio reductions to manage concentrated exposures. Legal and financial advisers underscore the challenges banks face in maintaining regulatory capital efficiency and managing internal risk limits, driving the push to offload or restructure such debt.

Complicating these efforts, confidentiality clauses embedded in many data centre loan agreements restrict the ability of lenders to engage in third-party risk mitigation activities, such as credit default swaps and synthetic risk transfers, which typically require sharing detailed loan information.

Regulatory bodies in Europe have also intensified oversight of bank exposures to data centre financing. The Bank of England recently requested detailed loan data from UK banks as part of its standard supervisory reviews. Meanwhile, the European Central Bank is conducting a supervisory assessment focused on Eurozone lenders’ data centre exposure and risk management practices, following discussions with banking executives in late October.

ING Bank, Société Générale, the Bank of England, and the European Central Bank declined to comment on the ongoing transactions and regulatory reviews.