Nobel Memorial Prize-winning economist Paul Krugman says the United States would almost certainly enter a recession if the artificial intelligence investment boom proved to be a bubble and collapsed soon. The depth of the downturn is uncertain, but he considers it likely to be worse than the 2001 recession that followed the dot-com boom. Even a mild contraction could be followed by a recovery lasting years.
In an analysis published on Substack on November 8, Krugman does not decide whether AI is a bubble. Instead, he asks a conditional question: what would happen if confidence underpinning the large investment boom were to falter in the near future?
A useful technology with uncertain returns
Krugman regards artificial intelligence as a remarkable technology that could generate substantial economic benefits, even if those benefits are not yet clearly visible in the data. A separate question is whether companies spending vast sums on data centers and other capacity can capture enough of those gains to justify their investments.
If they cannot, investors and lenders could simultaneously realize that their expectations were excessive. Krugman compares this to the cartoon coyote who continues running over a cliff until he looks down. Similar abrupt reversals of confidence preceded the collapse of the dot-com bubble and, later, the housing bubble.
He already sees signs of nervousness: shares in hyperscalers, the large cloud and data-center operators, have fallen 27% from their peak, while insurance against default on AI-company bonds has become much more expensive. The change in sentiment alone does not establish that the market has reached a turning point.
The speed of recovery matters
In the United States, the independent business cycle dating committee of the National Bureau of Economic Research officially determines recession dates. It does not simply look for two consecutive quarters of falling GDP: it assesses whether economic activity has declined significantly, broadly and for more than a few months.
There have been eight official recessions over the past six decades. The 1980 and 1981–82 downturns are often treated together, while the two-month pandemic collapse in 2020 is considered exceptional. Krugman therefore compares recoveries from six modern recession episodes.
His measure is the employment rate among people aged 25 to 54. Education and retirement distort this age group's rate less than they do the figure for the entire adult population. Employment rebounded relatively quickly after the 1969–70, 1973–75 and 1980–82 downturns. Recoveries after the 1990–91 and 2001 recessions were slower: after the dot-com bust, GDP was growing again while the employment rate remained weak for years.
Krugman partly attributes the difference to the causes of the downturns. The earlier recessions were largely brought on by interest-rate increases by the Federal Reserve System, or Fed, to combat inflation. Expensive credit particularly constrained homebuilding, which fell by roughly half. When the Fed eased policy, pent-up housing demand was released and helped pull the rest of the economy along. In early 1984, real GDP grew at an annualized rate of 8.6%.
Later recessions, in Krugman's account, were driven more by private-sector excess: too much money flowed into commercial property before the 1990–91 downturn, technology companies before 2001 and housing before the 2007–2009 crisis. When the investment and lending that powered a boom disappear, interest-rate cuts alone have a harder time replacing the lost demand.
An AI downturn could exceed the dot-com shock
In a recent opinion article, Jared Bernstein and Ryan Cummings estimated that the bursting of an AI bubble would deliver a negative economic shock three times as large as the end of the dot-com boom. Krugman says the estimate should be taken seriously despite the uncertainties and considers it likely that an AI collapse would cause at least as much damage.
One argument concerns the scale of technology investment. A calculation reported by Portfolio and based on Bureau of Economic Analysis data through the third quarter of 2026 puts private investment in information-processing equipment and software at 5.35% of GDP, compared with roughly 4.5% around the dot-com peak. The measure covers broader technology investment rather than AI spending directly.
The other factor is stock-market value relative to GDP, which Krugman says has also risen more than it did in the dot-com era. A loss of confidence in AI companies could destroy substantial wealth and consequently depress consumer spending.
There are, however, factors that could soften the US shock. Much AI investment has gone into imported equipment, particularly Asian semiconductors, so part of the burden would fall on South Korea and Taiwan. Investors might also move money into other stocks. During the dot-com bust, however, gains elsewhere did not offset the fall in technology shares: total stock-market capitalization relative to GDP fell by 50%.
Debt accumulated by AI-related companies presents another risk. Krugman says bond investors are concerned about the repayment capacity of companies including CoreWeave and Oracle. He has not yet formed a firm view on how serious those financial risks might be.
More limited policy tools
During the dot-com downturn, the Fed ultimately cut its benchmark rate from 6.5% to 1%, an easing of 5.5 percentage points. Krugman puts the current rate at about 4%; the latest Federal Reserve Economic Data figure puts the effective rate at 3.88% in October 2026. The central bank would therefore have less room to maneuver, and even a rate near zero might not restore full employment after a large shock.
Fiscal stimulus could in principle replace lost demand, but it is politically difficult to enact. Krugman says the Barack Obama administration's 2009 package helped, but political resistance made it smaller than necessary. After Republicans took control of the House of Representatives in 2010, there was no further stimulus and fiscal policy tightened. Unemployment did not return to its pre-crisis level until 2015.
For the next two years, Krugman considers an agreement on stimulus between President Donald Trump and a Democratic-controlled Congress unlikely. He believes political conditions could be more favorable after January 2029, but high government debt and a large deficit could complicate the response by then. He attributes the fiscal position partly to pandemic spending and more heavily to Trump's One Big Beautiful Bill law. He worries that another large deficit-financed package could unsettle markets, although he still considers support for the economy necessary.
Krugman's conclusion is conditional: if the AI boom is a bubble and ends soon, a recession is to be expected, potentially followed by a prolonged labor-market recovery. It would therefore be important to watch not only for renewed GDP growth but also for a recovery in the employment rate of 25- to 54-year-olds.



