Fed’s Barr Warns AI Financing Could Amplify Risks From a Market Selloff

Barr highlighted growing debt, circular financing and more than $3 trillion in potential contractual obligations as concerns if AI valuations fall sharply.

Eric Baker
Written by Eric Baker
Published
Share

Federal Reserve Governor Michael Barr warned Monday that the increasingly complex financing behind the artificial intelligence investment boom could magnify a market selloff, particularly if investors sharply reassess the returns they expect from data centers and related technology. The danger, he said, lies not simply in falling technology share prices but in the borrowing and financial ties that could transmit pressure to lenders and investors.

Speaking at an artificial intelligence finance conference in Milan on November 16, Barr described a buildout that has drawn substantial debt funding, long-term commercial commitments and investments among companies that depend on one another for revenue. He did not predict an imminent crash or argue that AI would fail as a technology. His concern was that financing arrangements could make any adjustment more difficult to contain.

The debt figures tell only part of the story

In his prepared remarks published by the Federal Reserve, Barr said the on-balance-sheet debt of seven major AI-related companies was expected to exceed $800 billion before the end of 2026. That is a projection for a defined group of large companies, not an estimate of all borrowing tied to the AI industry. The distinction matters because data centers also draw financing from developers, private lenders and specialized investment vehicles.

The Fed’s November Financial Stability Report estimated that public debt outstanding in the wider advanced-computing sector exceeded $1.1 trillion as of October. That measure covers a broader set of obligations than the seven-company figure Barr highlighted. The report also noted that borrowing has spread beyond conventional corporate bonds into structured lending and project finance, making the full amount of privately arranged debt harder to measure.

Commitments that do not appear as ordinary debt on a company’s balance sheet form another part of the exposure. They include future data-center leases, energy-supply agreements and promises to support the resale value of equipment used as loan collateral. Barr cited the Fed’s estimate that the maximum nominal value of such obligations now exceeds $3 trillion. Those potential commitments are not the same as debt already due, and their values generally are not discounted in the same way as reported borrowing. Adding the two numbers together would therefore give a misleading picture of liabilities.

The structures can help finance projects that would otherwise struggle to borrow cheaply. A well-capitalized technology company may promise to lease computing capacity from a data-center developer, giving that developer a dependable customer agreement that supports its borrowing. Equipment suppliers can offer guarantees that make graphics processors more acceptable as collateral. If demand falls short, the obligations and the risks attached to them may become much more consequential than they appeared when projects were being financed.

Barr also pointed to more than $100 billion in cumulative equity investments by large cloud-computing and chip companies in AI laboratories and other smaller businesses. Some recipients use the money to buy chips or computing services from their own investors. Such circular financing can sustain the buildout while making the health of suppliers and customers more closely intertwined. Barr emphasized that smaller firms ultimately need revenue from paying end users rather than continued support from companies selling them computing capacity.

A repricing could reach far beyond AI stocks

A central question is whether the spending underway today produces enough future revenue to justify the prices investors are paying. Barr acknowledged that AI might eventually bring major productivity improvements. Even that outcome, however, would not guarantee that investors earn the returns they anticipate on the timetable they expect. A gap between commercial results and market expectations could force a large reset in valuations.

The immediate economic consequence would be lower investment as projects are delayed or canceled. Losses in financial portfolios could also weigh on household wealth and business spending. Those effects become more severe when investors must sell assets to meet financing requirements, repay borrowed money or reduce exposures after prices decline. Barr highlighted the possibility that such selling could feed on itself when leverage is high.

The Fed’s stability report found hedge-fund leverage at its highest level since the series began in 2013. It also described significant links between advanced-computing projects and pension funds, insurers, mutual funds, private-credit providers and other investors. The report estimated that large U.S. bank holding companies had close to $1 trillion in exposure across a broad group encompassing technology firms, data-center developers, utilities and related financing vehicles. That figure describes exposure, not expected losses from an AI downturn.

There are important limits to the comparison with past technology booms. Barr said he was less worried about direct credit losses from the present buildout than about the wider effects of a reversal in risk appetite. He noted that the largest banks have substantial capital buffers that may help contain spillovers. Still, lenders’ relationships with leveraged clients could become more difficult to assess during a sudden selloff, especially where guarantees, special-purpose vehicles and private lending obscure where the ultimate risk sits.

Physical assets present another challenge. Graphics processors and specialized chips account for a large share of data-center costs, but newer equipment may displace them quickly. Barr noted that rental prices for older chips had so far held up, evidence that older computing capacity continued to find users. Power connections and other infrastructure could retain value even if AI demand weakens, although some facilities might take years to find sufficient alternative use.

Trading algorithms add a separate channel of risk

Barr’s warning extended beyond the companies financing AI projects. He also examined what could happen if financial-market participants come to rely on similar AI trading systems. Models built with related data and methods may respond to market news in similar ways, increasing the chance that buying or selling becomes synchronized. A shared weakness in those systems could then turn a localized problem into a broader disruption.

He stressed that large-scale autonomous trading with generative AI was not yet, to his knowledge, widespread. The risks he described were therefore prospective rather than evidence that AI models had already triggered a market-wide failure. Alongside the potential for improved market surveillance and more efficient information processing, he identified manipulation, misleading market information and opaque automated decisions as areas needing attention.

The Fed’s market-contact survey illustrates why the issue has gained prominence. In responses gathered during September and October, 87% of 23 contacts cited AI as a salient financial-stability risk, up from 50% in the spring survey. Those are the views of a small group of financial-market contacts, not a representative poll of U.S. investors. Concerns also included possible corrections in risky assets and cyberattacks.

Barr did not announce a new AI-financing rule or a monetary-policy decision in the speech. His remarks instead pointed toward closer scrutiny of borrowing structures, guarantees and counterparty exposures as the buildout continues. The November Financial Stability Report provides the Fed’s latest assessment of those vulnerabilities, including the ways financing could connect a downturn in technology investment to the broader financial system.

Eric Baker

About the author

Eric Baker

Trading and Quantitative Markets Contributor

Eric Baker writes about trading, probability and risk. Drawing on more than two decades of experience in personal and proprietary trading, he explains position sizing, expected return, downside exposure and the difference between a sound decision and a favourable outcome.

View author profile