AI named top financial risk by 87% in Fed survey

AI named top financial risk by 87% in Fed survey

The Federal Reserve released its twice-yearly Financial Stability Report on November 13. A survey of market contacts attached to the report shows that 87% of respondents named AI as a leading risk to the U.S. financial system, the top answer on the list. In the May edition, the figure was 50%.

The survey was conducted in September and October and covered 23 financial market participants, up from 20 in the previous round. As usual, the Fed noted that the responses reflect only the views of the individuals surveyed, not those of the Board of Governors or the New York Fed. Data in the report run through October 28.

The two risks that followed AI were a pullback in risk-asset prices, cited by 52%, and rising long-term interest rates, cited by 48%.

What respondents are worried about

Concerns about AI fall broadly into three groups.

The first is valuation. The report says equity valuation measures sit at elevated levels, partly because of market enthusiasm for “superintelligence” (the report uses the term and the abbreviation SI). The scenario respondents fear is that if AI-related revenue falls short of expectations, the broader market could see a sharp, wide-ranging correction.

The second is cybersecurity. The report splits the source of threats into two ends: sophisticated state-level attackers on one side, and the disruption that frontier AI models themselves could cause on the other. In other words, the models are listed as a risk source in their own right, not merely as tools in an attacker's hands.

The third is the structure of financing, which is also where the report spends the most space.

Where the $680 billion in debt comes from

The report's figure: as of October this year, debt issuance tied to advanced computing has exceeded $680 billion, roughly five times the total issued in all of 2023. By the measure cited in press coverage, that already accounts for about 31% of investment-grade corporate bond issuance by non-financial companies.

That covers only debt already sold. The report separately estimates that debt, guarantees and leases tied to AI infrastructure that have not yet begun to be executed could add up to more than $3 trillion.

The Fed compares this wave of investment with the broadband build-out of the late 1990s, and suggests that at the current pace its scale could soon surpass that earlier boom.

The report singles out one structure: large technology companies using long-term contracts, leases and payment guarantees to help data center operators raise funds. The operators borrow against those commitments, then turn around and supply computing capacity to the big companies. The report warns that because of this mutual entanglement, the exposure the large companies actually bear may be much greater than the liabilities shown on their balance sheets.

On the banking side, the report says the largest U.S. bank holding companies have combined exposure of close to $1 trillion to large technology companies, data center developers, special purpose entities tied to advanced computing, private credit vehicles and utilities. The report does not break down how much of this has actually been lent and how much is merely committed credit lines.

Three surveys side by side

Put the last three surveys together and the trend for AI is clear.

In the November 2025 edition, policy uncertainty ranked first, cited by 61% of respondents, while AI was at only about three in ten. In May 2026, geopolitics moved to first place and AI rose to 50%, ranking third. In this edition, AI has overtaken both geopolitics and policy to sit at the top.

What happened between a third and nearly nine in ten is visible to everyone: data center projects keep getting larger, and financing has extended from bond issuance to guarantees and leases. Our earlier coverage of Nvidia backing OpenAI's Ohio computing campus is one example of the “big company backstops the operator” structure described in the report; the size of that guarantee was later scaled back once.

The respondents' worries are also changing. In May, AI was mostly a general concern about stretched valuations. In this round, the answers point to specifics: revenue falling short of expectations, models being used in attacks, and off-balance-sheet commitments that are hard to see.

The report ends by linking these points into one chain: if returns on AI investment come in below expectations, falling asset prices, tighter credit and reduced investment could stack up and drag on the U.S. economy. High leverage in parts of the financial system would amplify the shock. The report specifically notes that leverage at large hedge funds is the highest since the Fed began collecting this data in 2013.

The Fed's wording is that the complex links among data center companies, public credit markets, private lenders and special purpose vehicles “warrant close monitoring.” The report proposes no new regulatory measures and does not show how the unexecuted $3 trillion in commitments is distributed among companies. The next report is expected in the spring of next year.

Sources: Federal Reserve Financial Stability Report and its survey of market contacts, CNBC, CocoLoop, SBS. Respondent shares follow the Fed survey's own basis; debt figures follow the advanced-computing-related issuance listed in the report.