Ex-NY Fed chief Dudley: AI boom may sow the next financial crisis

Felix Laurent 3 Nov 2026 4 min read

Big tech

Bill Dudley ran the Federal Reserve Bank of New York for nine years. On November 3, 2026, he warned that the AI boom may be sowing the seeds of another financial crisis. In a Bloomberg Opinion column, he put planned AI infrastructure spending at more than $10 trillion if all current plans go ahead.

His main worry is not the size of the bet but how it is funded. Much of the money comes through channels that are hard to see, from private credit to off-balance-sheet leases. Three weeks earlier, the IMF had named an AI-driven market correction as one of the top risks to global markets.

What Bill Dudley warns about the AI boom

Dudley lists where the money comes from: cash flows, stock and bond sales, private debt and off-balance-sheet leases. Hyperscalers and Nvidia also give guarantees to data center developers. He sees several reasons for concern, according to a summary of the column by Investing.com.

The first is competition. At least seven or eight companies are racing to become hyperscalers, but Dudley expects only two or three to dominate, as happened in cloud computing. The losers' spending would not pay off.

The second is the revenue bar. Dudley estimates that AI infrastructure could need more than $3 trillion in annual revenue to earn an adequate return. Data centers and chips also lose value fast, which raises that bar further.

If spending reverses, stock markets could take a hard hit. Lenders and investors who financed the build-out would be exposed too. That raises the question of who exactly they are.

Why AI debt is hard to see

The answer is not simple, Dudley argues. Much of the funding runs through private credit and leasing deals, so it is hard to establish who is exposed and on what terms. Some obligations stay off balance sheets until data center projects are finished.

The Bank for International Settlements described the same pattern in March 2026. Hyperscalers issued over $100 billion in bonds in 2025, mostly long-term. They also moved more borrowing into joint ventures and special-purpose vehicles with private credit firms, where the hyperscaler often holds only a minority stake.

The BIS called this "shadow borrowing": obligations that work like debt but sit mostly outside corporate balance sheets. Private credit funds and institutional investors hold the debt, and banks support the vehicles with funding lines.

Dudley also points to Nvidia. The chipmaker has guaranteed future values of its chips for some deals. If chip prices fall, those guarantees could hit Nvidia just as its margins come under pressure.

That opacity is the link to the IMF's warning.

What the IMF found in AI financing deals

The IMF's Global Financial Stability Report came out in October 2026. It named three main risks: an AI-driven market correction, a jump in government bond yields and cyberattacks. It singled out circular deals, where AI companies invest in each other and buy from each other.

The IMF reviewed 14 such transactions since 2023. The table shows what they add up to.

Circular AI deals since 2023 (IMF review of 14 transactions)	Amount
Equity investments	About $300 billion
Debt	$650 billion
Future revenue commitments	$1.4 trillion

The risk runs in a loop. Stress at a key company that receives investment would cut revenue forecasts and valuations, which would weaken the value of the debt in those deals.

Banks look solid overall, the IMF says. But in its stress test, banks holding 22% of global banking assets would end up below regulatory capital limits, up from 18% a year earlier. The IMF calls for limits on leverage and liquidity mismatches in AI, private credit and other nonbank finance.

Dudley's prescription is narrower, and it starts from a different reading of the risk.

Why Dudley does not expect a 2008 replay

Dudley does not see a repeat of 2008 as the most likely outcome. Before the 2008 crisis, mortgage risks were hidden from most investors, while AI risks are already widely discussed. AI financing is also mostly long-term, which makes a sudden wave of forced asset sales less likely.

What he asks for is better disclosure. Regulators should require clearer reporting of financing arrangements and contingent liabilities, he argues, so investors can see who carries the risk.

Is the AI boom heading for a financial crisis

Not by Dudley's own reading, at least not like 2008. The risks are public, the debt is long-term and the biggest spenders have strong cash flows. A correction would hurt stocks, but it would not necessarily freeze the financial system.

The danger is in what cannot be seen. More than $10 trillion in planned spending and $650 billion of debt in circular deals already blur the picture. With more borrowing moving off balance sheets, it is hard to know where losses would land. Dudley says AI needs roughly $3 trillion a year in revenue. If it falls short, private credit and lease deals are where the first losses would likely show up. Until regulators force those exposures into the open, nobody can say how big the hit would be.