“Our economy is right now being supported by AI investment—the AI bubble,” Stiglitz told Fortune in a recent interview. “Like, a third of the growth, or the non-growth, that we had last year was based on AI. So this AI bubble is having positive macroeconomic effects in the short run. I believe that it is a bubble in two ways.

“There’s both a short run and a long run,” he said. The problem, Stiglitz argued, is that almost everyone in the public debate is only hearing one of them.

If there’s a bubble, it will have an explosive burst

Stiglitz, who won the Nobel Prize in economics and wrote about the structural failures of modern capitalism in his 2024 book, The Road to Freedom: Economics and the Good Society, believes the current wave of AI investment rests on a foundation that cannot hold. 

“The market believes that there are going to be high returns to these investments. That is predicated on two assumptions: that AI will be technologically successful and that there will be limited competition,” he said. 

The problem is that global competition in AI is already fierce, from U.S. tech giants to Chinese firms: “Because if it’s technologically successful, but there’s a lot of competition, profits will be driven down to zero, and they will not get the returns that they expect.”

When that realization hits, the fallout will not be gentle. “If I’m right, and there is this bubble,” Stiglitz warned, “then the breaking of any bubble is really bad in the short term for the macroeconomy.”

That collapse, if it comes, would arrive while AI is simultaneously displacing workers across the economy—a worst-of-both-worlds scenario that Stiglitz does not think is far-fetched.

“We do not have the macro or micro framework for managing that kind of displacement,” he said. No active labor market policies. No large-scale retraining infrastructure. No industrial strategy to create the next round of good jobs in the places where old ones are disappearing. “It would require large retraining programs and so forth,” he noted, programs that do not currently exist at anything close to the scale needed.

The displacement gap

Stiglitz has seen what happens when societies lack those tools.

“When in the Great Depression, it was partly a success of agriculture,” he said. “We increased productivity enormously. We didn’t need as many farmers, but we had no ability to move people out of the rural sector—and we finally did it in World War II. But it was government intervention as a result of the war that resolved that problem. We don’t have the institutional framework for doing that.”

The parallel is not comforting. If AI succeeds at automating large portions of routine cognitive work—the research, the drafting, the analysis, the administrative processing that fills millions of office jobs—and the economy has no mechanism to redirect that labor, the result is not just a productivity story. It is a human one. “The foundations of a strong macroeconomy are almost inconsistent,” Stiglitz said. “I just can’t see how it can happen.”

He’s clear that AI will “make a particularly large difference in some routine white-collar jobs”—the very jobs filled by workers who tend to sit behind desks, carry college degrees, and feel farthest from the disruptions that hammered manufacturing workers a generation ago. This sense of safety many knowledge workers feel may soon be under threat.

Act two: The long game

But here is where Stiglitz’s argument pivots, and where it becomes more interesting than either that of the doomsayers or the boosters. Zoom out far enough, past the bubble and the displacement shock, and AI begins to look less like a replacement for human workers and more like a tool that makes them better at what they already do.

Take education. Stiglitz estimated it represents roughly 14% of the labor force, and he is unequivocal about what AI can and cannot do there. “It’s not going to replace teachers. It may help them do better lesson plans. It may help them tailor education better, but it’s not going to replace the teachers. We know enough about how students learn that the human interaction still seems to be very important.”

Health care tells a similar story, though with a sharper political edge. The U.S. health care sector represents nearly 20% of GDP and is, by global standards, spectacularly inefficient, spending far more than comparable countries for worse outcomes. While boosters argue AI will fix this, Stiglitz disagreed.

“Is AI going to solve that problem? No. We know precisely why our health care system is inefficient, and it has to do with rent seeking, with lack of competition, with the fact that we don’t have a public health system. It’s the politics. Is AI going to solve that political problem?”

AI can improve record-keeping, accelerate drug development, and sharpen diagnostic tools. What it cannot do is restructure the insurance industry, dismantle hospital monopolies, or make the political choices that a dysfunctional system requires. The problem was never a lack of computing power.

And then there is the plumber—perhaps the most vivid example Stiglitz offers of what the future actually looks like in practice. Stiglitz, who likens his own use of AI as “IA,” for intelligence assisting, says AI will supplement our labor in the future, and plumbing is a prime example. 

Far from being displaced, the plumber gets smarter. “It will help the plumbers maybe do their job better. They can feed in the symptoms of the problem, and it will give the diagnostics, and it’s probably a broken pipe in the wall, and may help them do their job better. That’s the intelligence assisting part.”

He paused, then added the line that captures his entire long-term argument: “But you still would need the plumber.”

The catch

The hopeful second act only materializes under the condition that societies survive the first act with their institutions intact. If the short-term bubble burst triggers mass displacement into an economy with no safety net, no retraining programs, and a government deliberately stripped of the capacity to intervene, the long-term IA vision becomes unreachable—not because the technology fails, but because the human infrastructure required to deploy it fairly was dismantled before it was needed.

Stiglitz’s warning is not that AI will destroy the future of work. It is that the transition between now and that future is the most dangerous part, and we are walking into it without a map.