AI in action · Markets
Fed: AI's Off-Balance-Sheet Ties Top $3 Trillion
Debt raised for AI computing passed $680 billion by October 2026, the Fed says, while leases and guarantees kept off balance sheets exceed $3 trillion.
How much has the AI build-out borrowed? The Federal Reserve's answer, in a box titled "Leverage in the Data Center Buildout" inside its November 13 Financial Stability Report, is that the visible debt is only part of it.
Start with the visible part. Debt issued to finance advanced computing infrastructure rose to more than $680 billion by October 2026, against $130 billion in all of 2023, according to Bank of America and Bloomberg data cited by the Fed. Investment-grade technology companies issued most of it, and that borrowing went from less than 6% of non-financial investment-grade bond issuance in 2023 to about 31% in 2026. Outstanding public debt tied to advanced computing is estimated at more than $1.1 trillion, and the Fed notes that private credit deals make the true total harder to measure.
Then the less visible part. Across Amazon, Alphabet, Broadcom, Meta, Microsoft, Nvidia and Oracle, leases that have been signed but have not yet started carry undiscounted values above $1 trillion, and other commitments such as take-or-pay contracts and residual value guarantees exceed $2 trillion. These obligations have grown faster than the companies' debt. The Fed cautions that they are undiscounted figures and not directly comparable with balance-sheet debt.
The mechanism is familiar to anyone who has followed the sector. A hyperscaler signs a long lease with a data centre developer, and the developer uses the lease to borrow for construction. A chip designer guarantees what its GPUs will be worth later, which helps a smaller AI firm finance buying them. When the larger company also takes an equity stake, the Fed writes, the result is a "circular financing" dynamic.
Who would feel a downturn? The Fed sees two sides. The strongest firms may absorb losses that would otherwise hit the wider system, but their real exposure "may be substantially larger" than their balance sheets suggest. The riskier tranches of the special purpose vehicles behind many of these deals sit with high-yield funds, life insurers and business development companies, and the largest US bank holding companies have close to $1 trillion of exposure to big tech firms, data centre developers, these vehicles, private credit funds and utilities, which the Fed says it examines in its annual stress test.
The report also compares the current cycle with the telecom boom that began in 1995. Hyperscaler capital spending has tracked that cycle closely as a share of GDP and has accelerated lately, which the Fed says suggests it may soon surpass it. The difference so far is that strong cash flows kept these firms from needing much debt, at least until this year.
Sources: Federal Reserve — Financial Stability Report, November 2026, Box 2.2 (PDF); American Banker — Fed warns of financial risks from AI-driven market optimism (Nov 13, 2026)