Artificial intelligence has become the most frequently cited potential threat to U.S. financial stability among the market contacts surveyed by the Federal Reserve Bank of New York. In the Federal Reserve's Financial Stability Report released on November 13, 2026, concerns ranged from cyber incidents to elevated technology valuations and increasingly complex financing for AI infrastructure.

According to the report’s overview and survey box, 87% of surveyed contacts identified AI-related risks, compared with 50% in the spring. That is a significant change in sentiment, but the denominator matters: the latest survey drew responses from just 23 market participants, including financial institutions and advisers. It is not a poll of U.S. households, a prediction of an imminent crisis or a claim that 87% of banks are unsafe.

Why borrowing has moved into the spotlight

The Fed's report contains a separate examination of funding for the data-center buildout. It estimates that advanced-computing-related debt issuance exceeded $680 billion by October 2026, up from roughly $130 billion in 2023. The report says the sector's share of nonfinancial investment-grade debt issuance rose from under 6% in 2023 to about 31% in 2026.

These are issuance figures, not a measure of debt currently outstanding or a tally of defaults. The Fed estimates outstanding public debt connected to advanced computing at more than $1.1 trillion, while noting that private deals are harder to measure. The report also flags leases, guarantees, project-financing structures and reciprocal business relationships as ways exposure can travel beyond a company's most obvious balance-sheet liabilities.

That distinction changes how investors and policymakers should read AI spending announcements. A long-term contract for data-center capacity can help a developer secure financing; it does not necessarily demonstrate that future model demand will generate the revenue required to justify the investment. A financing structure can spread losses, but links among lenders, infrastructure developers and technology suppliers can also spread stress.

Three risks, not one AI bubble prediction

The survey respondents described three broad mechanisms. First, expensive AI-related equities could fall if earnings undershoot optimistic projections. Second, malicious cyberattacks or failures associated with frontier systems could disrupt financial activity. Third, large and intricate borrowing arrangements could become difficult to assess during a downturn.

These concerns overlap, yet none establishes that a financial crash is inevitable. The report says the banking sector remained sound and resilient, while identifying areas where leverage, valuations and funding practices deserve monitoring. Its survey box explicitly says participants' responses should not be read as the Federal Reserve Board's own judgments.

The Fed also compares parts of the current capital-spending cycle with the late-1990s telecommunications buildout. That analogy is about the pace and financing of infrastructure investment, not proof that AI demand will follow the dot-com bust. Hyperscalers generate substantial cash flows, and the productivity benefits of AI remain an important upside scenario.

What to watch next

For enterprise technology buyers, the immediate question is less dramatic than the word “bubble”: can contracts, utilization and unit economics support promised capacity? For lenders and policymakers, it is whether guarantees and special-purpose vehicles make risk harder to see. The cost pressures facing enterprise AI adopters are a related but different problem: customers weigh model value against ongoing operating expense, while infrastructure providers must finance capacity years in advance.

For now, the Fed has identified a growing cluster of vulnerabilities and a sharply higher level of concern among respondents. It has not declared AI investment unsustainable. That distinction is the most important line in the story.

Sources: Federal Reserve, Financial Stability Report (November 2026), especially Box 2.2 and Box 5.1.