As banks grow increasingly wary of concentrated exposure to the AI infrastructure sector, some institutions have begun actively divesting related loan assets. ING Bank and Societe Generale have recently put up for sale up to $800 million in data center debt combined, involving multiple loans tied to projects under Oracle, signaling that risks accumulated by the banking sector during the AI boom are accelerating their release into the market.
According to a November 17 report by the UK's Financial Times, people familiar with the matter said the loan sales have been underway for several months, but after Democrats scored a sweeping victory in the US midterm elections, heightened concerns over tighter US data center regulation prompted banks to accelerate the pace of cleaning up their balance sheets before year-end. Meanwhile, the Bank of England and the European Central Bank have also successively launched reviews of banks' data center loan exposures, with regulatory pressure tightening from both sides simultaneously.
From a market pricing perspective, some loans are still quoted above 90 cents on the dollar, indicating overall credit quality remains acceptable; however, for a New Mexico project facing power supply obstacles, the expected transaction price is projected to fall to around 85 cents on the dollar, reflecting substantive risks at individual projects.
ING and SocGen List Loans Tied to Multiple Oracle Projects
According to the Financial Times, people familiar with the matter said the auction being conducted by ING Bank involves approximately $500 million in loans, with borrowers including data center operators CyrusOne, EdgeConnex and Vantage Data Centres, as well as Oracle-backed Project Jupiter in New Mexico. Project Jupiter is currently facing delays in power permit approvals, and Oracle has issued a force majeure notice in response.
Societe Generale, meanwhile, is seeking to sell approximately $300 million in project finance debt, involving Oracle's Project Lighthouse in Wisconsin and Project Frontier in Texas. The three projects together raised more than $56 billion in construction debt from dozens of banks last year.
In terms of pricing, loans related to the Wisconsin and Texas projects are quoted above 90 cents on the dollar, and people familiar with the matter said the sale is more of a "portfolio management exercise" rather than an urgent concern over credit quality. However, loans for the New Mexico campus are expected to be priced at around 85 cents on the dollar, reflecting the additional uncertainty brought by delays in power permits for that project.
Concentrated Exposure Triggers Alarm, Banks Pursue Multiple Avenues
According to reports, people familiar with the matter said caution in the banking sector toward AI data center debt has been building for months, with the core issue being excessive concentrated exposure to a small number of large technology companies and frontier AI laboratories.
Don Dimitrievich, head of global energy infrastructure credit at Nuveen, said:
"The scale of capital we are talking about is historic by any measure and has pushed beyond the limits of traditional capital sources. Many financial institutions' chief risk officers have essentially made clear that these portfolios need to be compressed."
He also noted that recent debt sales have offered favorable financing terms and more attractive returns, driven by repricing of risk by some lenders and regulatory capital constraints.
David Lucking, a structured finance partner at law firm Linklaters in New York, said:
"Banks holding this debt is capital-inefficient. Whether from a capital standpoint or from the perspective of risk limits and internal balance sheets, banks need to find ways to continue lending."
Beyond directly selling loans, banks have also explored credit insurance and risk transfer transactions to expand lending capacity. Recently, a growing number of banks have structured deals in 144A format to sell debt to a broader pool of institutional investors.
144A bonds refer to private placement bonds issued under Rule 144A of the US Securities Act of 1933. The rule allows issuers to conduct targeted private placements to qualified institutional investors within the United States without the cumbersome public registration process with the Securities and Exchange Commission (SEC).
According to people familiar with the matter, Anthropic's $63 billion debt financing package and SpaceX's $40 billion financing for purchasing AI chips are both planned to enter the 144A bond market in the coming months.
Notably, some major lenders have already exited Anthropic's financing deal, precisely due to concerns about concentrated exposure to the AI laboratory. This development further confirms the overall tightening trend in the banking sector's risk management in the AI space.
Confidentiality Clauses Create Hedging Obstacles, Regulatory Scrutiny Escalates
On the risk management front, banks face a new operational challenge: some data center loan contracts contain confidentiality clauses that prohibit lenders from participating in third-party transactions requiring the sharing of loan information, including credit default swaps (CDS) and synthetic risk transfer instruments, making it difficult for banks to hedge related exposures through conventional derivative tools.
David Lucking said: "If these types of clauses become more common in the market, it will create considerable problems."
Meanwhile, European regulators are also stepping up scrutiny. According to reports, people familiar with the matter said the Bank of England is requesting detailed information from UK banks about their data center project loans as part of routine regulatory work; the European Central Bank is conducting a supervisory review to assess eurozone banks' exposure to and management of data center financing, with related plans discussed at a meeting with bank executives as early as late October.
Comments