Banks are now trying to offload the AI data center debt they just underwrote

The clearest example is a $38 billion construction loan package for a Texas and Wisconsin data center leased to Oracle. JPMorgan and MUFG are among the lenders that have spent more than six months trying to distribute it. Some are now willing to sell pieces to non-bank buyers at a discount. The banks are also exploring significant risk transfers, which move the riskiest slice of a loan to outside investors while the bank keeps a defined share. The driver is internal limits on how much exposure a single lender can hold to one borrower or sector. Demand has cooled elsewhere. Bloomberg reported that order books for Alphabet, Amazon, Meta, Microsoft and Oracle bonds fell from roughly five times the amount offered in February to under two times by July. Oracle's credit default swaps have widened, and S&P rates it BBB-, one notch above junk. Investors are rotating toward debt backed by investment-grade tenants such as Microsoft and Google. They are moving away from single-tenant structures like those used by CoreWeave and newer AI infrastructure borrowers. For AI founders, the banks are not exiting the sector. They are pricing risk by tenant quality. Companies that rely on leveraged neoclouds or weaker credits for compute may see higher financing costs, tighter terms or slower capacity.