Banks struggle to offload $38 billion in AI data center debt
Major banks are quietly trying to dump billions in AI infrastructure debt as exposure limits hit and bond demand shrinks.
Business · Source: Hacker News
What happened
Major banks are quietly trying to shed the AI data center debt they just underwrote. Lenders like JPMorgan, Morgan Stanley, SMBC, and MUFG are exploring significant risk transfers. They want to shift the riskiest slices of this debt to outside investors while keeping the loans on paper. This is happening because banks are hitting strict internal limits on AI exposure. No single lender wants to hold this much concentrated risk in one sector.
One major deal shows exactly how stuck the market has become. A group of banks including JPMorgan and MUFG has spent over six months trying to distribute $38 billion in construction debt. This massive debt pile is tied to Texas and Wisconsin data centers leased to Oracle. Banks are now reportedly willing to sell pieces of it to non-bank buyers at a steep discount just to clear their books. Six months is an eternity to hold debt that was never meant to stay on a balance sheet.
The broader market for AI debt is cooling fast. Order books for tech bonds shrank from five times the available amount in February to under two times by July. Oracle is feeling the squeeze directly as it reportedly targets another $40 billion in debt and equity this year. Standard and Poors downgraded Oracle to one notch above junk status in July. The cost to insure Oracle debt against default just hit levels last seen in 2009.
Key facts
- $38 billion — Construction debt tied to a Texas and Wisconsin data center project leased to Oracle
- 151 basis points — Cost of insuring against Oracle defaulting on five-year credit default swaps in early October
- $120 billion — Total AI infrastructure spending spread across special purpose vehicles
- 5.304% — The 10-year Treasury yield peak this month, its highest level since 2002
- BBB- — Oracle credit rating after Standard and Poors cut it to one notch above junk
Why it matters
The era of blank checks for AI infrastructure is ending right now. If you are building a smaller AI infrastructure company, your cost of capital just went up significantly. You are now competing for the exact same hesitant pool of lenders as giants like Oracle. Lenders are getting incredibly picky about where they park their money. Capital is rotating heavily toward data centers leased to investment-grade hyperscalers like Microsoft and Google. If you lack a massive balance sheet or a proven customer base, raising debt next quarter will be a brutal wake up call.
This debt bottleneck threatens the entire physical AI buildout. Tech companies used special purpose vehicles to keep $120 billion in infrastructure spending off their own balance sheets. That massive risk is now piling up at the banks that financed the vehicles. At the same time, the macroeconomic environment is turning hostile. The 10-year Treasury yield hit a two-decade high of over 5.3 percent this month. Borrowing costs are surging across the board. If banks cannot sell these existing loans, they will simply stop underwriting new ones.
For builders
Cost of capital spikes for smaller firms
Lenders are demanding much higher yields from companies without hyperscaler backing. CoreWeave debt is already pricing up to 9.875 percent compared to cheaper rates for hyperscaler projects. Smaller builders will pay a massive premium for infrastructure debt while established players maintain their advantage.
Special purpose vehicles face intense scrutiny
Founders can no longer easily hide infrastructure costs in special purpose vehicles. Banks are choking on the risk they absorbed from these exact structures. Startups relying on these vehicles to finance compute clusters will likely face outright rejection or punishing terms from nervous lenders.
Collateral demands will drain startup cash
Regulators are forcing infrastructure projects to post massive collateral. Oracle is currently fighting a rule in Wisconsin that requires $7 billion in collateral for a single campus. Startups building physical infrastructure must prepare for sudden regulatory demands that lock up their operating capital.
My take
I have been warning about this invisible ceiling on the AI boom for months. We all assumed credit would stay open forever to fund these massive compute clusters, but now banks are choking on $38 billion of Oracle debt. If Wall Street is begging hedge funds to buy Oracle debt at a discount, they definitely will not finance your unproven GPU cloud startup.
Original reporting: Hacker News. This is my rewrite and opinion.