Goldman Sachs Raises AI Job Displacement Forecast: Up to 15 Million U.S. Workers Could Be Affected Within a Decade
Taylor Wilson
Goldman Sachs now estimates generative AI could displace over 9% of the US workforce — roughly 15 million workers — within a ten-year adoption cycle, up sharply from its earlier 6%-7% range, signaling that the labor-market impact has been systematically underestimated.
Why did Goldman raise the number so sharply?
The key driver is a methodological overhaul: the old model measured how many people were unemployed at a given point in time ("stock"); the new model tracks the rate at which workers flow out of existing jobs ("flow").
This means → the old approach was a snapshot; the new one is a video — and the video captures people leaving continuously, so the count is naturally higher.
The new model finds that each 1% rise in productivity lifts the job-destruction rate by roughly 0.5–0.6 percentage points over the following two years — about 50% higher than the old estimate.
Is there a historical precedent?
Goldman's closest reference point is the late-1990s to early-2000s ICT boom.
During that wave, each 1% technology-driven productivity gain pushed the job-destruction rate up by 0.6–0.7 percentage points — higher than the current AI model's baseline.
This means → if AI's productivity shock exceeds the historical 75th percentile, displacement could rise further. The current 9% figure may not be the ceiling.
Could a recession make it worse?
Historical data show that the disappearance of routine jobs — positions with fixed processes easily replaced by automation — tends to concentrate during recessions.
In plain terms = during downturns, firms accelerate cuts to automatable roles while the economy's capacity to create new jobs is at its weakest — a double squeeze.
This reflects a structural risk: AI displacement will not arrive at a steady pace. If it coincides with a recession, the shock will be front-loaded.
Why is Goldman still optimistic?
In a normal year the US economy creates a net 25–35 million new positions, providing substantial absorptive capacity.
Goldman sees AI generating new employment through three channels: ① new job categories (the digital economy created roughly 15 million); ② deeper specialization (healthcare grew from 2 million to over 18 million workers in sixty years); ③ income growth lifting demand for discretionary services.
Briggs estimates that if job losses spread evenly over the ten-year cycle and most displaced workers find new roles within a year, the peak unemployment-rate impact stays below 1 percentage point.
What is the biggest hole in that optimistic case?
The critical assumption is that new-job creation keeps pace with displacement — and that is precisely the largest uncertainty in the forecast.
This means → Goldman's conclusion is essentially conditional: if new jobs keep up, the shock is manageable; if they don't, the unemployment-rate impact will far exceed 1 percentage point.
In plain terms = Goldman itself concedes that its optimistic scenario rests on an assumption no one can currently verify.
Content is for reference only, not financial advice.