Banks Reduce AI Data Center Debt Exposure as Regulatory Scrutiny Intensifies
nashnova research
ING Bank and Société Générale are marketing up to $800 million in data-center loans for sale, some tied to Oracle's Stargate-linked projects. This means → the banks that funded the AI-infrastructure boom are now actively shedding risk, and the credit-tightening signal is spreading.
Who is selling, and what exactly is on the block?
ING is auctioning roughly $500 million in loan stakes across data-center operators CyrusOne, EdgeConnex, Vantage, and Oracle's New Mexico "Project Jupiter."
Société Générale is marketing about $300 million in project-finance debt tied to Oracle's Wisconsin "Project Lighthouse" and Texas "Project Frontier."
Those three Oracle projects alone raised over $56 billion in construction financing from dozens of banks last year. This means → the current sales are a sliver of a far larger exposure — and a signal of system-wide anxiety over concentrated AI lending.
Why do the discount levels matter?
Wisconsin and Texas project loans are quoted above 90 cents on the dollar; people familiar with the sales call them routine "portfolio management," not acute credit deterioration.
The New Mexico campus loan is expected to price around mid-85 cents — noticeably lower.
In plain terms = same borrower, same bank syndicate, sharply different discounts. The gap maps directly to project-specific risk: Oracle has issued a force-majeure notice on the New Mexico site because power permits remain stuck, making construction delays real.
Why are banks offloading now?
The core driver is growing caution over concentrated exposure to the AI sector and a handful of big-tech and frontier-AI borrowers — senior risk officers at many institutions have explicitly called for portfolio reduction.
After the Democratic win in the U.S. midterms, fears of tighter data-center regulation intensified, pushing banks to clean up balance sheets before year-end.
Nuveen's global head of energy-infrastructure credit, Don Dimitrievich, said: "The scale of capital we are discussing is historic by any measure — it has burst through the limits of traditional funding sources."
What is 144A, and why is it becoming the go-to structure?
144A — a format that lets debt be sold to a broader pool of institutional investors — is emerging as the new channel for data-center financing.
In plain terms = banks' risk limits on any single borrower are close to full. 144A repackages the loans and sells them outside the banking system, freeing up capacity to keep lending.
Anthropic's $63 billion debt package and SpaceX's $40 billion financing for AI-chip purchases both plan to tap the 144A bond market in coming months — yet some lead lenders have already walked away from the Anthropic deal over concentration concerns.
What structural obstacles do banks still face?
Linklaters partner David Lucking noted that holding this debt is "capital-inefficient" for banks — both on regulatory-capital and internal risk-limit grounds — and that they need a path to keep lending.
He also flagged that some data-center loans carry confidentiality clauses restricting lenders from hedging via credit-default swaps (CDS) or synthetic risk transfers.
This means → banks want to offload risk, but contractual terms may block the exit. Lucking warned that if such restrictions become more widespread, they would pose "quite a significant problem."
What are regulators doing?
The Bank of England has requested detailed information from UK banks on their data-center project lending, as part of routine supervisory work.
The European Central Bank is conducting a supervisory review of eurozone banks' data-center financing exposure and risk management, first discussed with banking executives at a late-October meeting.
This reflects a synchronized tightening of regulatory attention on both sides of the Atlantic. Whether banks can effectively disperse AI-infrastructure debt through 144A and risk-transfer mechanisms will be the key test of whether this credit contraction stays orderly — or becomes systemic pressure.
市场有风险,内容仅供研究参考,不构成投资建议。
