AI Could Displace Up to 9% of U.S. Workers Over a Decade

Artificial intelligence could displace up to 9% of U.S. workers during a 10-year adoption cycle, while broader productivity scenarios lift the estimate above that level to about 15 million workers.

The baseline assumes AI eventually increases economy-wide productivity and gross domestic product by 15% after full adoption. The measure counts workers pushed out of existing jobs, including those who find new roles or leave the labor force. It does not mean 15 million people would remain unemployed.

The higher figure comes from scenarios using larger productivity shocks and comparisons with the late 1990s. Those cases put displacement above the baseline’s 9% ceiling, with the total reaching about 15 million workers.

The labor-market effect depends on pace. If job losses are spread across the decade and most displaced workers find new jobs within a year, the unemployment-rate increase would likely remain below 1 percentage point in any given year. That threshold describes the annual increase under those conditions, not the number of people displaced over the decade.

Historical shifts show why the measures differ. Each 1% increase in technology-driven productivity has been followed by a cumulative 0.5 to 0.6 percentage-point rise in the job-destruction rate over the next two years. About 85% of U.S. job growth over more than 80 years has come from occupations that did not exist in 1940. In a typical year, gross job gains and losses each range from 25 million to 35 million.

In technology, management consulting and graphic design, the estimated AI-related drag on job growth was 10,000 to 15,000 jobs a month. Joseph Briggs, Goldman Sachs’ senior global economist, described the effect as narrow at the time. “We’re not expecting that displaced workers will be displaced over the long run,” Briggs said.

A recession could make the transition more painful if AI-related job losses cluster when businesses are less able to create new positions. The projection expects new roles to absorb displaced workers over time, though the timing and scale of the shift remain uncertain.

A separate estimate that AI could displace 18% of current U.S. jobs over five years used a task-based approach, making it a distinct measure from this productivity-linked scenario.