Banks are now trying to get rid of the AI ​​data center debt they just took on

JPMorgan, Morgan Stanley, SMBC and MUFG are quietly buying pieces of the AI ​​debt they helped build, a sign that the lenders closest to the money have started hedging their own bets.

Six months ago, banks were competing to underwrite data center loans related to Oracle, Meta and the rest of AI development. Today, some of these same banks are trying to remove this debt from their books. JPMorgan, Morgan Stanley, SMBC and MUFG have explored large risk transfers, according to a Financial Times report. These are agreements that allow a bank to keep a loan on paper while transferring its riskiest tranche to outside investors. Banks are doing this because they face internal limits on the amount of AI-related exposure a lender can hold.

One case shows how blocked this situation is. A group of banks, including JPMorgan and MUFG, spent more than six months trying to distribute $38 billion in construction debt tied to a data center project in Texas and Wisconsin leased to Oracle, Bloomberg reported. Some lenders are now willing to sell pieces of them to non-bank buyers at a discount, just to move them. Six months is a long time to be in debt forever.

Matthew Moniot, co-head of credit risk sharing at Man Group, told the Financial Times that the size of the transactions involved is “out of scale compared to anything we’ve ever thought about”. This isn’t about a hedge fund manager badmouthing a trade he doesn’t want to be a part of. This is someone whose job it is to assess exactly this type of risk, claiming that the numbers no longer fit any model he’s used before.

Oracle is the clearest indication. Its five-year credit default swaps, the company’s cost of default insurance, climbed to about 151 basis points in early October, levels last seen in 2009, according to market data cited by Bloomberg. S&P had already lowered Oracle’s credit rating to BBB-, one notch above junk, in July. The company is reportedly targeting more than $40 billion in additional debt and equity this fiscal year. That’s on top of the $18 billion in bonds sold in a single day last September, and the money is needed to continue funding its locations in Texas, Wisconsin and New Mexico.

Demand for this paper has also declined. Bonds of Alphabet, Amazon, Meta, Microsoft and Oracle now trade at a premium to similarly rated non-AI company bonds. Order books for their offerings fell from about five times the amount sold in February to less than twice in July, Bloomberg reported. It’s the difference between a bond that sells itself and a bond that a bank has to work to place.

None of this happens in isolation. The 10-year Treasury yield rose to 5.304% this month, its highest level since 2002, after a $39 billion auction settled at 5.30%, the highest auction yield since 2000. Minutes from the Federal Reserve’s September meeting cited AI-related corporate borrowing as contributing to the move. Capital Economics says the main driver is inflation and government deficits, not tech companies’ bond sales. No matter who is right about the cause, the effect is the same for anyone currently trying to finance a data center: the risk-free rate at which they borrow has become much more expensive.

Frankly, the mechanics here matter more than the number in the title. Banks do not sell debt they trust. They sell debt when they worry about concentration, which is exactly what “approaching internal limits” means in plain English: too much of this bank’s balance sheet is now a bet on the resilience of a sector. Spreading $120 billion in AI infrastructure spending across special-purpose vehicles was supposed to make that risk easier to swallow. This is the structure that Oracle, Meta, xAI and CoreWeave have all used to keep debt off their own balance sheets. Instead, the risk lies with the banks that financed the SPVs in the first place.

Investors who still want to participate have become more picky about choosing a location. Capital is shifting toward data centers leased to investment-grade hyperscalers like Microsoft and Google, and away from weaker credits, Bloomberg reported. This is exactly why the price of paper leased by CoreWeave is between 7.00% and 9.875%, compared to 5.70% to 7.875% for bonds leased by hyperscalers, according to bond market data compiled by the CCIR. A smaller, less capitalized AI infrastructure company trying to raise debt next quarter is now competing for the same nerve pool of lenders as Oracle. Except without Oracle’s balance sheet or existing customer base to point to.

This is the part of this story that doesn’t show up in a single day’s stock movements. The AI ​​boom still relies on the assumption that credit remains open indefinitely on reasonable terms. Six months of a bank struggling to sell $38 billion in loans is the clearest evidence yet that this hypothesis is being tested in real time, not in a theoretical downturn that might still be years away.

Also read: Trump-linked USD1 stablecoin adds $427M in single week • IREN shares fall 20% in ten days as scathing infrastructure report collides with Wall Street’s AI bull case • Micron board authorizes record $35B stock buyback

This article is published in Financial Markets News, check it out for more related stories.