AI Data Center Debt Has Passed $680 Billion. What the Fed’s Numbers Measure

Server racks in a data center with a glowing upward financial line

Federal Reserve staff estimated that borrowing for advanced computing infrastructure exceeded $680 billion by October 2026. The total was about five times the $130 billion issued during all of 2023. The report also puts outstanding public debt related to advanced computing above $1.1 trillion as of October.

Those figures describe different measures. The $680 billion is debt issued over time. The $1.1 trillion estimate is public debt still outstanding. Advanced-computing debt also grew from less than 6 percent of nonfinancial investment-grade corporate debt issuance in 2023 to about 31 percent in 2026.

What the debt estimates include

The Fed groups financing for data centers and related infrastructure under “advanced computing debt.” It includes corporate bonds and loans, structured credit, and project finance. The report says the mix has shifted toward public and private debt, while companies have also taken on leases, guarantees, and other contingent obligations.

The issuance estimate covers public deals and trackable private securities issued under Rule 144A. It excludes Section 4(a)(2) private placements, and the report says some private deals are not captured. The full amount of debt outstanding is therefore difficult to measure.

A November 16 speech by Federal Reserve Governor Michael Barr described a wider set of commitments surrounding the buildout. The on-balance-sheet debt of the seven largest AI firms was expected to exceed $800 billion before the end of 2026. Barr also cited potential off-balance-sheet obligations with a maximum nominal value above $3 trillion. Those include energy contracts, long-term data center and computing leases, and other financing arrangements.

The $3 trillion figure is a maximum nominal amount across contractual obligations, not a bill payable at once. Some commitments depend on future events, and leases and guarantees have different payment schedules. Their terms determine when a firm may have to pay.

How financing problems could spread

Data center loans can depend on expected project income or use buildings and computing equipment as collateral. A delay, higher costs, or weaker customer demand could reduce the cash available for debt payments. Lenders may also have to reassess equipment collateral if specialized chips lose value quickly or few buyers want older models.

Some financing connects companies across the supply chain. A large technology company may support a project through a guarantee or invest in a smaller company that then rents computing capacity from it. Barr described these arrangements as possible circular financing. They can make it harder to see which company ultimately carries a loss if a borrower runs short of cash.

If investors mark down AI-related assets, complex obligations and uncertainty about who bears them could amplify that repricing. A sharp change in demand could also leave lenders with collateral that is difficult to sell. These are risk channels discussed by Federal Reserve officials, not a forecast that defaults or forced sales are imminent.

What the figures say about the banking system

The borrowing totals alone do not show that a banking crisis is underway. Barr said capital at the largest banks could help absorb spillovers, while counterparty exposures still merit monitoring. The public debt estimates also do not reveal every lender’s exposure or include every private transaction.

Data centers, chips, and power connections may earn returns if customers generate enough revenue to cover their costs. The Federal Reserve’s figures show how quickly debt financing has grown and how much of the broader commitment sits outside standard on-balance-sheet borrowing. Future demand and the terms of those commitments will affect how that financing performs.

Sources

This article explains public financial-stability data for general information. It is not a forecast or a recommendation to lend, borrow, or invest.

Comments

Leave a Reply

Your email address will not be published. Required fields are marked *