Building AI data centers is becoming a stress test for banks
Key Points
- Major banks are reaching their internal risk limits with multi-billion dollar AI data center loans and are trying to pass on default risks to other investors.
- A $38 billion loan package for Oracle data centers in Texas and Wisconsin shows the scale of the problem.
- Political opposition to data centers, for example in Maine, adds to the uncertainty.
Constructing new AI data centers is consuming billions in borrowed capital. Major banks like JPMorgan and Morgan Stanley are now looking for ways to offload the growing credit risks to other investors.
The rapid buildout of AI infrastructure in the US is increasingly straining the banking system. According to the Financial Times, major banks including JPMorgan Chase, Morgan Stanley, and SMBC are searching for ways to pass on the risks from AI data center financing to other investors. The loan volumes for new data centers have grown so large that individual institutions are hitting their internal limits for risk concentration.
One deal illustrates the scale of the problem: a $38 billion loan package is financing data centers in Texas and Wisconsin tied to Oracle. According to the Financial Times, JPMorgan and MUFG have spent months trying to distribute portions of these loans more broadly across the market. Some banks reportedly even tried to sell the loans at a discount to non-bank buyers. Oracle itself had previously raised $18 billion through bond offerings.
Banks are "choking" on the numbers
To limit their exposure, banks are exploring loan sales and so-called significant risk transfers. With the latter, the loan technically stays on the bank's books, but a portion of the default risk gets passed on to credit funds, insurers, or other investors in exchange for a return.
Matthew Moniot of the Man Group summed up the problem in the Financial Times: the sums are so large that banks quickly start "choking." Frank Benhamou of Cheyne Capital sees these deals as riskier than typical risk transfers. There are only a handful of operators, the loans are heavily concentrated, and construction projects can fail or end up costing far more than planned.
Political pushback adds another layer of risk
Financial uncertainty isn't the only concern - there's political headwind too. In Maine, the state legislature passed LD 307, a moratorium on data centers with a capacity of 20 megawatts or more, set to run until November 1, 2027. The law would also have established a council to evaluate impacts on electricity customers, the grid, the environment, and the local economy.
Governor Janet Mills vetoed the bill on April 24, 2026, arguing that it would have blocked a $550 million project at the site of the former Androscoggin paper mill in Jay. Mills expects the project to create more than 800 construction jobs, at least 100 permanent positions, and increased tax revenue. The veto stood on April 29, 2026. Instead of the legislation, Mills signed an executive order creating a 15-member advisory council tasked with delivering recommendations for handling large-scale data centers in Maine by January 2027.
AI News Without the Hype – Curated by Humans
Subscribe to THE DECODER for ad-free reading, a weekly AI newsletter, our exclusive "AI Radar" frontier report six times a year, full archive access, and access to our comment section.
Subscribe now