Paul Krugman Argues AI Data Centers Could Crowd Out Jobs

Summary

Paul Krugman presents a more pessimistic assessment of AI’s labor-market effects, focusing on the scale and job intensity of data-center investment. Drawing on a paper by Stijn Van Nieuwerburgh, he says a representative 200 MW AI training campus costs about $8.2 billion but typically employs only 50 to 150 workers once operational. By comparison, deploying a similar amount of capital in a non-AI sector would, using average U.S. fixed assets per worker of roughly $700,000, support more than 10,000 jobs. Krugman therefore estimates that data-center investment creates about one percent as many jobs as the same capital invested elsewhere. His argument depends on the claim that AI investment is crowding out other uses of capital rather than simply adding to total investment. Van Nieuwerburgh estimates that hyperscalers could spend more than $10 trillion on data centers through 2032; a back-of-the-envelope calculation suggests this capital could employ roughly 10 million fewer workers than it would elsewhere. Krugman stresses that this does not mean 10 million people will become unemployed. Instead, lower labor demand could put downward pressure on wages and shift economic power further toward capital, reinforcing an existing decline in labor’s share of national income relative to profits. He says the scale of AI capital spending and the small number of jobs created have changed his view since June, when he was more skeptical of large negative effects on labor. The article frames the argument as a modern version of concerns about capital-biased technological change and links it to 19th-century economic analysis associated with David Ricardo.