Notes / US job market

Goldman says AI could displace 9 percent of workers. The second half matters more.

Home/News/The 9 percent number

23 October 20265 min readNumbers checked by hand

Quick answer

Goldman Sachs estimates AI could displace up to 9 percent of workers. The same forecast says the effect probably arrives with the next recession, not on a schedule. A capability number is not a timetable, and companies tend to cut in downturns, using AI as the reason.

9%Displacement estimate
16,000Monthly jobs AI removed
9,000Monthly jobs AI added
Chart of AI job displacement estimates
Capability and timing are different claims. Photo: VOXA.

What the number says

Fortune reported on 22 October that Goldman Sachs puts AI-driven displacement at up to 9 percent of workers, with the heaviest impact probably waiting for the next recession.

Read as a headline, 9 percent is alarming. Read as a forecast, it is conditional: the capability exists, the job losses are gated on a business cycle.

Why the timing clause matters

Firms do not restructure in a good quarter. They restructure when revenue slips and a board wants a plan. AI provides that plan now, in a way it did not in 2023, which is why the next downturn is likely to look different from the last one.

That also explains an oddity in the 2026 data: total announced US job cuts fell about 40 percent year on year in the first half, while the AI-attributed share of those cuts roughly doubled. Fewer cuts overall, with AI named more often.

The scale that gets lost

Goldman's earlier work found AI removing about 16,000 monthly payroll jobs while adding about 9,000, a net loss of roughly 7,000 a month. PwC's 2026 AI Jobs Barometer found headcount growing faster at the most AI-exposed firms than the least exposed.

Set that against Challenger's 112,713 AI-attributed announcements through July and the gap is obvious: announcements are events, payroll data is a stock. One moves markets, the other describes reality.

What to do with a number like this

Treat it as a risk map, not a countdown. The roles Goldman-style estimates put at risk are the ones with high overlap to text work: support, data entry, junior analysis. The roles with protection are physical, regulated, or responsible for the AI itself.

If you are in the first group, the useful move is not to wait for the recession. It is to be the person who owns the tool that replaces the task.

Sources: Fortune on the Goldman Sachs forecast (22 October 2026); Goldman Sachs workforce research; Challenger, Gray and Christmas via HRD America; PwC 2026 AI Jobs Barometer.No affiliate links

Read next: Barclays cuts 5,000 roles