Highly AI-exposed U.S. workers see real wages slide while headcount holds steady

Wage growth trails in AI-exposed occupations

The Apollo paper, co-authored with Sania Edlich, compared real wage growth in 321 of roughly 800 Bureau of Labor Statistics occupations between 2023 and 2026. Only 11 occupations met what the authors defined as a high-AI-exposure threshold, and the gap widened over time. Real wage growth in those 11 was nearly flat after 2023, while growth across the rest of the labor market held closer to its pre-pandemic trend.

The BLS Employment Cost Index showed that inflation-adjusted wages and salaries for private-industry workers decreased 0.4 percent year over year through June 2026, even as nominal wages rose 3.1 percent. Higher-pay sectors such as tech and professional services are shedding jobs while lower-pay sectors such as hospitality and health care have been leading the gains, which mechanically pulls the average pay package downward.

Apollo stressed that its findings count as "early evidence." The high-exposure sample is small, and the paper said the analysis is also significant because it shows that "AI research has entered a new phase, one in which labor market impacts can be measured from observed adoption rather than predicted from theoretical exposure."

Labor's share of output falls to a 79-year low

The wage data sit inside a broader shift in who is capturing U.S. economic output. The BLS productivity release for the second quarter of 2026 reported that the labor share, the percentage of nonfarm business output paid out as worker compensation, was 52.8 percent, the lowest reading in the series that starts in the first quarter of 1947.

In the same report, real hourly compensation fell 3.3 percent in the quarter and was down 0.1 percent over four quarters. BLS computes unit labor costs as hourly compensation divided by productivity; unit labor costs rose 1.4 percent over the last year. The combination of rising productivity and softer real pay is the structural pattern that AI, if it spreads, is expected to deepen.

Some researchers attribute the multi-decade slide in labor's share to automation, which AI could accelerate. The Apollo paper situates itself inside that longer arc: its authors say the post-2023 wage slowdown in AI-exposed occupations looks consistent with companies routing productivity gains into smaller pay packets rather than smaller workforces.

Entry-level workers carry the heaviest share of the pain

The Federal Reserve Bank of Dallas looked at the same pattern in February and found that wages in the top 10 percent of AI-exposed industries actually grew 8.5 percent since fall 2022, more than the 7.5 percent national average for average weekly wages. The tech design sector stood out, with wages up 16.7 percent.

But the same bank found that hiring has trailed. Employment in computer systems design and related services fell 5 percent since fall 2022, while the most AI-exposed sectors overall lagged the rest of the economy by roughly 3.5 percentage points. Stanford researchers Erik Brynjolfsson, Bharat Chandar and Ruya Chen found the gap falls almost entirely on workers under 25. An earlier Dallas Fed analysis put the employment decline for the 22-to-25 age band at roughly 13 percent in the most AI-exposed roles.

The bank's reading is that AI substitutes for codified, textbook-style knowledge while complementing the tacit, experiential knowledge that older workers bring. That means the routine tasks new graduates are paid to learn are exactly the ones AI now performs, collapsing the traditional on-ramp to white-collar careers.

Why economists disagree about what the data mean

MIT economist Daron Acemoglu, who has written extensively on automation and wages, said he expects the impact of AI on pay to be visible before any impact on jobs. "Ultimately, given that the U.S. labor market is relatively flexible and has a fairly weak social safety net, I expect the impact on wages to be bigger than those on employment," he told CNBC. He cautioned that current models are not yet widely adopted across many occupations, so some displacement effects "may be exaggerated."

MIT labor economist David Autor warned that the language of "AI exposure" can mislead. In a paper with Neil Thompson, he compared two occupations that looked identical decades ago. Both were "seemingly destined for obsolescence" in the computer era. Accounting clerks saw wages climb 39 percent while employment fell 32 percent; inventory clerks saw wages drop 13 percent while employment rose 175 percent. Same automation force, opposite outcomes, because AI can automate the routine slice of one occupation and leave the expert slice untouched in another.

Ben Zipperer, senior economist at the Economic Policy Institute, said the Apollo sample is too small to be convincing. He argued that if AI lowers the cost of building software by cutting demand for developers, the saved money flows elsewhere, lifting demand for other workers and making highly exposed roles look weaker "by comparison, even though some of that measured loss is just income increases for other workers."

Jennifer Huddleston, senior fellow in technology policy at the Cato Institute, said the debate overlooks the jobs AI is generating. "One often underappreciated element is the way AI is leading to potentially new categories of jobs and opportunities for entrepreneurship," she said. The Department of Labor is funding AI-literacy programs aimed at workers, not industry protections, and the question now is whether that upskilling can keep pace with the entry-level squeeze.