Banks explore offloading AI data center risk as $38 billion Oracle-linked debt stalls

The banks that raced to underwrite AI data center loans six months ago are now working to cut their exposure to that same debt. JPMorgan, Morgan Stanley, SMBC and MUFG have been exploring significant risk transfers, arrangements that let a lender keep a loan on its balance sheet while passing the riskiest slice to outside investors, according to the Financial Times. The reported reason is straightforward: the banks are running into internal limits on how much AI-linked exposure any single lender can hold.

A $38 billion deal that will not clear

The clearest measure of how congested this market has become is a single transaction. According to Bloomberg, a bank group including JPMorgan and MUFG has spent more than six months trying to distribute $38 billion in construction debt tied to data centers in Texas and Wisconsin that are leased to Oracle. Some lenders are reportedly willing to sell portions of it to non-bank buyers at a discount just to get the debt moving.

The scale is unusual even for people who price this risk professionally. Matthew Moniot, co-head of credit risk sharing at Man Group, told the Financial Times that the deal sizes involved are "out of scale to anything we've thought about, ever."

Oracle is the pressure point

Much of the stress traces back to Oracle's finances. Market data cited by Bloomberg shows Oracle's five-year credit default swaps climbed to roughly 151 basis points in early October, a level last seen in 2009. S&P cut Oracle's rating to BBB-, a single notch above junk, in July. The company is reportedly targeting more than $40 billion in additional debt and equity this fiscal year, on top of the $18 billion in bonds it sold in a single day last September, with the money funding sites in Texas, Wisconsin and New Mexico.

Investor demand is thinning

Appetite for AI-linked bonds has weakened alongside those credit signals. Bonds from Alphabet, Amazon, Meta, Microsoft and Oracle now trade at a premium to similarly rated corporate debt without AI exposure, and order books for their offerings have shrunk from roughly five times the amount on offer in February to under two times by July, Bloomberg has reported.

The rate environment is compounding the problem. The 10-year Treasury yield reached 5.304% this month, its highest since 2002, after a $39 billion auction cleared at 5.30%, the highest auction yield since 2000. The Federal Reserve's September meeting minutes flagged AI-related corporate borrowing as one contributor to that move, though Capital Economics argues that inflation and government deficits are the bigger drivers. Whichever explanation holds, the risk-free rate that data center financing is priced against has become markedly more expensive.

The SPV structure is pushing risk back onto banks

The financing behind this buildout was designed to spread risk around. Startup Fortune notes that roughly $120 billion of AI infrastructure spending has been spread across special purpose vehicles, a structure used by Oracle, Meta, xAI and CoreWeave to keep debt off their own balance sheets. In practice, the risk has landed on the banks that financed those vehicles, which is why internal exposure limits are now starting to bind.

Capital is rotating toward the safest credits

Investors still willing to buy this paper have grown selective about who the tenant is. Capital is moving toward data centers leased to investment-grade hyperscalers such as Microsoft and Google and away from weaker credits, Bloomberg has reported. Bond data compiled by CCIR shows the pricing gap: paper leased to CoreWeave yields 7.00% to 9.875%, against 5.70% to 7.875% for hyperscaler-leased bonds.

Why it matters

Banks do not usually sell down debt they are confident in. They do it when concentration becomes a concern, and approaching internal limits is the plain-English version of saying too much of one balance sheet now rides on one sector holding up. The AI buildout has rested on the assumption that credit stays available indefinitely on reasonable terms. A bank group failing for six months to move $38 billion in loans is direct evidence that the assumption is being tested in real time rather than in a distant downturn, and the tightening will bite hardest for smaller AI infrastructure companies raising debt next quarter, which now compete for the same wary lenders as Oracle without its balance sheet or customer base.