The National Bureau of Economic Research surveyed 750 chief financial officers at U.S. companies and asked them, privately, what they planned to do with their workforces because of AI. Less than half said they planned any AI-related job cuts at all. But the ones who did? Their numbers add up to 502,000 roles eliminated by the end of 2026.
That is nine times the rate of 2025. And according to the CFOs themselves, it is still not the worst-case scenario.
What the Numbers Actually Say
Employers reported roughly 55,000 layoffs attributed to AI in 2025. The NBER paper projects that number reaching 502,000 in 2026. About half of those roles are in white-collar positions — not factory floors, not call centres. Desks. Offices. People with degrees doing knowledge work that companies now believe AI can replicate at a fraction of the cost.
John Graham, the study’s co-author and director of the Duke CFO Survey, was careful with his framing. “It’s not the doomsday job scenario that you might sometimes see in the headlines,” he told Fortune. He is technically correct. 502,000 is approximately 0.4% of the total U.S. workforce. In an economy of 160 million workers, that reads as a rounding error on a spreadsheet.
It does not read that way to the 502,000 people losing their jobs.
The Productivity Paradox Nobody Wants to Talk About
Here is the finding that got buried in most of the coverage: the same NBER paper found that more than 80% of firms reported no measurable productivity gains from AI, despite billions invested. Companies are cutting workers because of AI’s potential, not because of what it has actually delivered so far.
That is not a subtle distinction. It means hundreds of thousands of people are losing their jobs today based on a bet that has not yet paid off. The CFOs who answered this survey are pricing in a future where AI works as advertised, eliminating labor now to fund the infrastructure that is supposed to deliver returns later.
80,000 tech workers lost their jobs in the first 90 days of 2026, with companies openly citing AI as the reason while simultaneously announcing hundreds of billions in AI infrastructure spending. The math is not complicated. Payroll savings fund the data centers. Human workers are the line item being optimized away to pay for the machines that might, eventually, replace them.
Why Less Than Half of CFOs Plan Cuts — and Why That Understates the Damage
The survey found that 44% of CFOs plan some AI-related job cuts. That means 56% say they do not. But those numbers obscure a critical variable: the CFOs who said no cuts planned are not necessarily safe employers. Many are in industries where AI displacement is slower, more structural, and harder to attribute directly. Their workers are not being fired for AI. They are not being hired because of it.
The entry-level positions that companies are choosing not to backfill — the junior analyst, the support coordinator, the content writer — those disappearances never show up in layoff statistics. They are invisible cuts. AI is erasing 16,000 jobs a month, but a significant share of that erasure happens quietly, through hiring freezes and role eliminations that never generate a press release.
Block Cut 40% of Its Workforce. Others Are Watching.
Jack Dorsey’s Block eliminated more than 4,000 positions — approximately 40% of its entire workforce — attributing the cuts directly to AI. That is not a company trimming edges. That is a company rebuilding its operating model from scratch around the assumption that it needs far fewer humans to function.
Oracle sent 30,000 termination notices via 6am email while posting a $6 billion quarterly profit. Meta announced plans to cut 20% of its workforce — approximately 15,000 people. These are not struggling companies making painful tradeoffs. They are profitable companies choosing AI infrastructure over human labor as a strategic allocation of capital.
The CFO survey says 502,000 this year. What it does not say is what happens in 2027, when the models are more capable, the infrastructure is more mature, and the productivity gains that 80% of companies are currently not seeing start to arrive.
“Not a doomsday scenario” is a low bar. It means something is still happening. It just means the people setting the bar have decided the scale is acceptable.