NY
Closed
LDN
Closed
TKY
Closed
Calendar
Contact
Markets now

CreditMarket note

The Fed finds more than $3tn of AI leases and guarantees off big tech’s books

The Fed’s financial stability report finds AI borrowing of $680bn this year, and more than $3tn of leases and guarantees off big tech’s balance sheets.

By The Notebook Desk

The Federal Reserve has looked under the bonnet of the AI boom and found that much of what holds it up does not show on the balance sheets. In its Financial Stability Report, published on Friday, the Fed sets out how data centres are financed, and its central finding is that the promises investors cannot see have grown faster than the debt they can. Leases, contracts and guarantees signed by seven of the biggest tech firms have grown faster than their borrowing, and they tie those firms to smaller, weaker companies in ways the market cannot easily measure.

The visible debt is large enough. Borrowing to build data centres and other advanced computing reached more than $680bn this year by October, about five times the $130bn raised in all of 2023, the report says. Such borrowing now makes up about 31% of all investment-grade debt sold by non-financial companies, up from a small fraction in 2023. The Fed puts the public debt outstanding at more than $1.1tn, and warns that private deals make the full figure harder to pin down. As we reported on 4 November, lenders have started charging more for it.

The part that worries the Fed sits off the books. Amazon, Alphabet, Broadcom, Meta, Microsoft, Nvidia and Oracle have signed leases that have not yet started worth more than $1tn, and take-or-pay contracts and guarantees on the value of chips worth more than $2tn, both undiscounted. Michael Barr, a Fed governor, put the total plainly in a speech in Milan on Monday: the potential maximum value of these obligations “now exceeds $3 trillion”.

The mechanics are simple and the consequences are not. A cloud giant signs a long lease with a data-centre developer, and the developer takes the lease to a lender to borrow for the building. A chipmaker guarantees what its chips will be worth, so that an AI start-up can borrow against them, and often buys shares in the same start-up, which the report calls a “circular financing” dynamic. The strong firms lend their credit to weak ones without putting the debt on their own books.

That cuts both ways, the Fed says. In a downturn the giants may absorb losses that would otherwise hit lenders. But it also means their “economic exposure to such a downturn may be substantially larger than suggested by their on-balance-sheet liabilities”. The riskiest slices of the financing, the subordinated debt and equity of special-purpose companies, sit with high-yield funds, life insurers and private credit funds. And the largest US banks have close to $1tn of exposure to big tech, data-centre developers, the financing vehicles, private credit funds and utilities, the report estimates.

The Fed’s own leaders have turned this into a warning. Kevin Warsh, the chairman, told a conference in San Francisco on Monday that if what he calls super intelligence disappoints, “expected returns could turn down and equity valuations could fall, leading to an economic slowdown and a more notable rise in unemployment”, in his opening remarks. Michelle Bowman, the vice chair for supervision, had said a week earlier that AI revenues “are not yet sufficient to pay for the expense of maintaining existing and developing new AI models and data centers”, in a speech in Washington. Barr added that the chips themselves may lose value quickly as newer ones arrive, which matters when they are the collateral.

Markets agree that AI is the risk to watch. Among the market contacts the New York Fed surveyed in September and October, AI was the risk cited most often, ahead of a fall in asset prices and higher long-term interest rates, with AI-enabled cyberattacks on payment and trading systems among the concerns.

Our view: the report is less a warning of a crash than a map of who would pay for one. It reflects data to 28 October, before Jensen Huang left Nvidia, one of the seven firms. The banks look well capitalised, as Barr said, and the Fed’s stress tests already cover their exposure. The open question is the obligations nobody discounts: if demand for computing power falls short, the giants will be paying for capacity they do not need, and the smaller firms that borrowed against their promises will find out how firm those promises are.

Disclosure

This is analysis and opinion, not investment advice.

The story so farBig Tech’s AI bill

  1. Is the AI bubble bursting? The Nasdaq falls 2.6% as Meta’s spending outruns its cash−2.6%Nasdaq Composite
  2. Big Tech earnings this week: AI spending now uses 98% of their cash98%Capex as a share of operating cash
  3. Google earnings: Alphabet falls 10% on a SpaceX loss and a cash gap; Microsoft rises−10.1%Alphabet class A shares after hours
  4. Earnings rout: Alphabet, Amazon and Intel lose about $800bn as AI spending outruns cash−7.3%Amazon shares after hours
  5. Apple earnings: a record $114.5bn quarter, but memory costs squeeze its margins45%Apple gross margin forecast
  6. Big Tech borrowed $50bn for AI in six days, and lenders are charging more305bpOracle 2033 notes
  7. AMD earnings: data-centre sales nearly double and give the AI trade a breather$8.5bnAMD data-centre revenue
  8. The Fed finds more than $3tn of AI leases and guarantees off big tech’s books (you are here)> $3tnLeases
  9. Up next8 min read

    EquitiesMarket noteTuesday 17 November 2026, 22:50 UTC

    Nvidia’s best quarter yet was Huang’s. Its board has just turned him away

    Nvidia earnings results: revenue of $115.8bn beat its $108bn guide and it expects $130bn next. Its board does not intend to bring back Jensen Huang.

    Nvidia’s revenue more than doubled in a year, to $115.8bn in the three months to 25 October, well above the $108bn it had promised in August. It expects $130bn in the quarter to January, its results show. On Tuesday morning the man who built that business said he would come back if asked. On Tuesday evening the board said it does not intend to have him back.

    Continue reading

All 22 stories on big tech

ContactWrite to the Notebook

Something to add?

Spotted an error, or a source this piece should cite? Want your own research linked or featured? Tell us. We read every message.

Keep reading

Archive