AI Layoffs Fell 75% — Goldman Says They're Just Parked

US employers blamed AI for 18,393 job cuts in the third quarter, down 75% from 74,098 in the second. In an October 22 note, Goldman Sachs economist Joseph Briggs argued the cuts didn't stop. They are being stored in hiring freezes and offset by a data-center construction boom with a built-in expiry date.

The headline numbers look like relief. In May, American employers cited AI in 38,579 cuts — 40% of all announcements that month and the highest monthly total since Challenger, Gray & Christmas began tracking AI as a reason in 2023. By September, the figure was 3,961. AI dropped from the top layoff reason to fifth.

Goldman's read is less comfortable. Its economists expect AI to eventually displace about 9% of US workers, roughly 15 million people over a decade, plus a further 3% drag from slower hiring. The team's August 2025 estimate was 6% to 7%. The number went up, and Briggs's history of past automation waves says this kind of displacement mostly waits for a recession to appear.

What the Goldman note found

Briggs, who co-leads Goldman's global economics team, published three findings on October 22:

• Employment in the most AI-exposed industries has slipped about 10,000 jobs below long-term trend over the past three months.
• Hiring in AI-related construction has largely offset that shortfall, keeping national payroll numbers calm.
• Across service industries, higher AI adoption lines up with higher unemployment — a statistically significant relationship — alongside a weaker link to slower employment growth.

If AI lifts productivity by 15%, Goldman's baseline, up to 9% of workers could be displaced during the transition. The historical section is sharp: the share of Americans in routine jobs held steady through expansions and dropped in the recessions of the early 1980s, 2001, 2008 and 2020. Across 25 occupations that earlier technology hollowed out — file clerks, proofreaders, bank tellers — employment fell an average of 25% in the five years after decline began. Briggs wrote that AI disruption "could well amplify the broader business cycle."

Where the missing layoffs are hiding

Run the arithmetic. Goldman's 15 million displaced workers over a decade averages 125,000 a month. Third-quarter cuts citing AI ran about 6,131 a month. That is 4.9% of the forecast pace. Either the forecast is wrong, or roughly 95% hasn't happened yet.

Tank one is the job that never gets posted. Goldman's extra 3% from slower hiring works out to about 5 million jobs — a resignation not backfilled, a requisition that stays closed. The unemployment gap between entry-level and experienced workers has widened by 0.6 percentage point since pre-pandemic, and the wage gap by 1.3%. Challenger counted 90,787 planned hires in September, the weakest September since 2011.

Tank two is construction. The same capex that lets a bank skip backfilling a call-center seat hires the electrician wiring the server hall. But construction hiring tracks the growth rate of spending, not its size. Consensus forecasts have hyperscaler capex growth slowing from 84% in 2026 to 22% in 2027; even Goldman's bullish $1.1 trillion 2027 call implies growth halving to about 45%. Alphabet alone has guided to $195 billion to $205 billion this year.

What could break the storage thesis

The skeptics have real material. MIT economist Daron Acemoglu puts net AI job losses at roughly 2% to 4% over five years. Challenger itself notes AI-cited doesn't mean AI-caused — in August, restructuring led with 16,173 cuts, and a relabel looks like a pause in this data. The US economy destroys about 29 million jobs and creates 30 million in a typical year, per Briggs, so 125,000 a month is one in twenty of normal churn. Only about 23% of US establishments use AI in regular workflows, and Goldman's central case is gradual disruption, not a sudden break.

Still, the watch items are clear: the growth rate of hyperscaler capex and the backfill rate at white-collar employers. If those two turn down together, the other 95% of the forecast stops waiting.