Michelle Bowman, the Federal Reserve’s vice chair for supervision, gave a speech at the Council on Foreign Relations on November 10 and devoted an entire paragraph to discussing the financing risks of AI construction. she isSpeech notes released by the Federal ReserveZhong said that AI-related debt is expected to exceed US$800 billion this year. If AI revenue cannot catch up with rapidly rising costs, AI company stock prices may correct significantly, and financing costs will also rise.

The speech came on the day the U.S. CPI was released in October. Bowman said that day's price data showed that inflationary pressures may last longer, and that the October PCE and November employment reports will be watched before the December meeting.

Bowman: AI revenue is not enough to cover model and data center costs

Bowman said that AI revenue is growing rapidly, but "it is not enough to cover the costs of maintaining existing and developing new AI models and data centers." As capital expenditures surge, companies such as super-large cloud providers are increasingly relying on bond and stock market financing; investors believe these investments are risky, and financing costs have risen significantly.

She also said that the growth rate of corporate fixed investment driven by data center investment has been close to that of the technology boom in the 1990s. Most of the growth in corporate investment since 2025 has come from the construction of high-tech equipment, software and data centers.

Off-balance sheet obligations are estimated to be more than three times AI debt

The AI-related debt pointed out by Bowman includes corporate bonds, private placement credit and project financing. In addition to the debt itself, obligations such as long-term leases, contractual commitments and residual value guarantees have also increased significantly, and she said potential payments are estimated to be more than three times the debt.

The structure she described is: Very large cloud companies and large technology companies with high credit ratings bear off-balance sheet liabilities, allowing smaller, higher-risk companies to use them as guarantees, and use asset-based financing to build data centers and purchase accelerated computing equipment. At the same time, very large cloud companies and chip design companies have also invested in small companies that purchase their own products, "forming what appears to be circular financing."

Bowman said that the largest bank holding companies in the United States have certain exposures to AI financing, which will be reviewed in annual supervisory stress tests; life insurance companies, pension funds, private equity credit and mutual funds also have exposures. Most of these institutions have sufficient capital and stable funding sources, but their financing structures are complex, making it difficult to correctly assess counterparty risks.

Risks are concentrated in a small number of AI laboratories and very large cloud players

Bowman believes that risks are concentrated in a small number of large companies that provide AI services, including AI laboratories and very large cloud players. She said that if AI revenue cannot keep up with the rapidly rising costs, and public opinion and local opposition may bring more controls and delays, AI company stock prices may correct sharply, financing costs will increase, and access to credit will shrink. This will slow down AI construction, and economic activity and inflation will also cool down.

She said some of the reassessment may have occurred in recent weeks: stock valuations fell, corporate bond spreads widened and credit default swap quotes rose for some companies involved in building AI.

In terms of prices, Bowman estimates that in the 12 months to September, the part where core PCE was higher than the target was due to tariffs, energy and AI-related price pressures each accounting for about a quarter; the prices of electronic products such as memory chips, semiconductors, and computers have risen sharply this year, with some items rising by double digits. She forecast core PCE growth could rise to 3.3% annually in October.

announced on the same day October CPI higher than expected, Governor Waller has stated that he may consider raising interest rates by 2% in December. Bowman said that the benefits of further tightening must be carefully weighed against the costs borne by interest-rate sensitive industries, and that a clearer understanding of the sustainability of AI financing and investment will increase her confidence in future policy decisions.

this article Fed Bowman: AI debt will exceed US$800 billion this year, AI stocks may correct sharply first appeared in Chain News ABMedia。