For the first time, artificial intelligence is at the top of Wall Street’s concerns. In his half-yearly report on financial stabilitypublished on November 13, 2026there US Federal Reserve (Fed) reports that almost all of the market professionals it surveyed cite AI as a major risk. Above all, the central bank devotes an entire box to the debt which finances the construction of data centers, with figures which give the extent of the phenomenon.
AI, the primary source of market concern
Twice a year, the New York Fed questions market participants (banks, fund managers, investors) about the risks that could weaken the American financial system. During the investigation carried out in September and October 2026 near 23 contacts, 87% cited AI as a salient risk. The progress is spectacular: they were 50% in spring 2026 and 30% in the fall of 2025.
| Risk cited by market players | Share of responses (fall 2026) |
|---|---|
| Artificial intelligence | 87% |
| Correction of risky assets | 52% |
| Rising long-term interest rates | 48% |
| Cyberattacks | 43% |
| Public debt sustainability | 43% |
AI covers three distinct concerns here, according to the report: a sudden correction of stock market valuations companies linked to AI, the risk of cyberattacks carried out by state actors or made possible by the most advanced AI models, and the complexity of financial arrangements which make it possible to build infrastructure. The Fed also notes that stock valuations remain significantly above their historical medians, “partly driven by market enthusiasm” for advanced AI.
A debt that has increased fivefold in three years
The most commented on passage in the report is a box entitled “Leverage in the Data Center Buildout”. The Fed brings together figures that have been scattered until now.
Debt issues linked to advanced calculation (data centers, chips, AI infrastructures) have exceeded 680 billion dollars in 2026 at the end of October, i.e. approximately five times the 130 billion issued over the entire year 2023. They now represent approximately 31% of bond issues well-rated non-financial companies (“investment” category), compared to less than 6% in 2023. Public debt in circulation linked to this sector would exceed $1.1 trillionnot counting transactions concluded in private credit, which are more difficult to measure.
Off-balance sheet commitments much heavier than debt
The Fed goes further by looking at what does not appear directly in the accounts. For a group of seven large technology companies, the leases not yet started exceed 1 trillion dollarsand outright purchase agreements, residual value guarantees and similar commitments exceed 2 trillion. These amounts, not discounted, are greater than their debt. Their exposure to a downturn could therefore be, according to the Fed, “substantially greater” than what their balance sheets show.
The mechanism described is as follows:
- major cloud providers sign long-term leases for data centers not yet built;
- real estate developers rely on these leases to borrow and finance construction;
- graphics processor manufacturers guarantee resale value chips used as collateral for loans, often also taking equity stakes.
The Fed speaks of a dynamic of “circular financing”: the same actors find themselves at the same time customers, suppliers, guarantors and shareholders of each other. An analysis close to that of the IMF, which had already quantified the debt and circular financing of AI in its own report of October 2026.
Who bears the risk in the event of a rollover?
The riskiest tranches of these arrangements, the subordinated debt (repaid last in the event of difficulty) and the equity, are held by high yield bond fundsof the life insurersbusiness development companies and other vehicles of private credit. They are the ones who would suffer the first losses if AI revenues disappoint.
Banks are not left out. The largest US banks have an exposure of approximately 1 trillion dollars to tech giants, data center developers, ad hoc structures carrying out projects, private credit funds and electricity suppliers. The Fed nevertheless believes that the banking system remains strong and judges the vulnerabilities linked to corporate and household debt to be “moderate”.
Finally, it highlights another point of fragility: the leverage effect of hedge funds has reached its highest level since data began in 2013, and it is concentrated in a small number of large funds.
The precedent of telecoms from the end of the 1990s
The Fed makes a connection that speaks to investors: compared to GDP, investment spending by large cloud providers could soon exceed the pace of construction of telecommunications networks at the end of the 1990s. At the time, operators had taken on massive debt to lay optical fiber in anticipation of a demand that took years to materialize, which led to resounding bankruptcies after 2000.
The comparison has its limits. Today’s tech giants are much more profitable than the operators of the time, and the computing demand for AI is already real. But she points out that an infrastructure that is useful in the long term can ruin those who financed it on credit if the revenue arrives too late.
Why it’s important, even seen from France
This report focuses on the American financial system, but its conclusions extend beyond the United States. A large part of European savings, via index funds, life insurance or stock savings plans, is invested in major American technology stocks, which weigh heavily in global indices. A correction in valuations linked to AI would therefore impact millions of savers.
The report also comes at a time when IPOs are increasing in the sector, like that of Anthropic, expected with a range of 70 to 78 dollars per share. The more money from the markets flows into AI, the more central the question of the solidity of its funding becomes.
What the Fed will watch
The report does not recommend immediate action, but it outlines points to follow in the coming months:
- transparency of commitments: the Fed emphasizes that leases, guarantees and private credit arrangements make total exposure difficult to measure;
- the quality of borrowers: as debt accumulates, the ability of projects to generate sufficient revenue will be scrutinized;
- interconnections: the concentration of risks among insurers, private credit funds and a few large hedge funds could amplify a shock.
For AI companies, the message is clear: the days when money flowed in without question may be coming to an end. Investors, like regulators, will now look closely at how the data center race is financed, not just its scale. The next report, expected in spring 2027, will say whether market concern has subsided or whether it has transformed into real tensions.

