The Pulse
Fed's Barr: AI firms' off-balance-sheet obligations could top $3 trillion
Federal Reserve Governor Michael Barr says potential off-balance-sheet obligations tied to the AI buildout now exceed $3 trillion in nominal value. He warned that complex financing and leverage could amplify a sharp repricing of AI investments.

The potential maximum nominal value of the largest AI firms’ off-balance-sheet obligations now exceeds $3 trillion, according to the Federal Reserve’s Financial Stability Report, Governor Michael S. Barr said in a speech on financial stability. Debt carried on the balance sheets of the seven largest AI firms is also expected to exceed $800 billion before the end of 2026, he said.
Barr discussed the figures on November 16 at the Seventh Annual Association for Computing Machinery International Conference on Artificial Intelligence in Finance in Milan, Italy. If AI investments were significantly repriced, he warned, debt growth, circular financing and complex obligations could amplify the adjustment.
Investment and expectations
In one possible scenario among several, Barr said, AI turns out to be a transformative innovation that vastly improves productivity, yet investors may not see returns on the buildout on the timeline and to the extent they expect. That could lead to a significant repricing.
A large repricing could reduce output growth through a direct drop in investment and through lost wealth that lowers consumer and potentially business spending, he said. The Fed's report also found that the equity premium—a rough measure of the extra return investors require for holding stocks rather than risk-free bonds—was low by historical standards, though not nearly as low as during the dot-com stock market boom.
Obligations beyond debt
Barr said firms' obligations outside their balance sheets, including long-term contracts, leases and residual value guarantees, had grown faster than their on-balance-sheet debt. The estimated potential maximum nominal value includes energy supply contracts, long-term data center and computing leases, and a variety of financing structures. Firms generally present these liabilities before discounting, Barr noted, which complicates comparisons with on-balance-sheet debt, which is discounted.
Some arrangements involve large firms with strong credit ratings indirectly supporting projects involving firms with weaker credit, or supporting public utility investments. Barr said such complexity can be opaque to unsophisticated investors and can worsen selloffs when investor sentiment shifts, because it is hard to know where risk resides; the structures, he said, warrant close monitoring.
Money flowing back to suppliers
Hyperscalers and major graphics processing unit producers have more than $100 billion in cumulative equity investments in large language model labs and cloud-computing start-ups, Barr said. Those investments form part of a circular financing arrangement: smaller firms can use the funding to rent or buy computing capacity and chips from the companies that invested in them.
The investments are relatively small for the suppliers but critical for the smaller recipients, according to Barr. For the buildout to become economically viable, he said, the smaller firms will eventually need to replace supplier funding with revenue from paying customers.
Another form of support comes through guarantees on the resale value of data centers or chips used as collateral for asset-based financing. Such guarantees are reported as off-balance-sheet liabilities because they represent losses the larger firms might absorb if borrowers default and collateral is liquidated.
Leverage and spillovers
Barr said he was less concerned about direct losses given the still-limited amount of debt backing the boom. With hedge fund leverage and risk appetite apparently high, however, he cautioned that a sharp reversal could trigger rapid deleveraging and potentially self-reinforcing fire sales.
Although high capital levels at the largest banks provide protection against spillovers, counterparty exposures could rise during equity market volatility, he said. The governor also noted that specialized AI chips could depreciate quickly as newer technologies overtake them, while lease rates for older chips currently appear to be holding up.
Source
Related stories
- Fed's November survey ranks AI first among financial-stability risks
- IMF Scenario Sees AI Market Correction Cutting Global Output 1.1% by 2028
- BIS: AI-to-AI Deals Accounted for 55.2% of Disclosed Incoming Investment Value
- Fed researchers say AI's effect on productivity appears small so far
- Fed Governor Lisa Cook Says Elected Officials Should Set AI Policy