Apollo Data: AI Exposure Linked to 6.7% Decline in Real Wage Growth for High-Risk Jobs Since 2023

For years, the dominant anxiety surrounding artificial intelligence has been the threat of mass unemployment. However, newly released empirical data suggests a different reality is unfolding across the U.S. labor market: AI is keeping workers at their desks, but it is quietly shrinking their raises. According to a comprehensive white paper published by Apollo Global Management, companies are capturing the productivity gains of generative AI through wage compression rather than workforce reduction.[1][3]

The Apollo study, authored by analyst Sania Edlich and Chief Economist Torsten Sløk, tracked wage and employment data across 321 matched occupations from 2015 to 2025. By utilizing the Anthropic Economic Index—which measures observed, real-world usage of AI tools like Claude rather than theoretical exposure—the researchers isolated the economic shifts that occurred after ChatGPT went viral in late 2022. They found that jobs with the highest exposure to AI experienced an average 6.7 percent decline in real wage growth post-2023.[1][2]

Crucially, the data revealed no statistically significant decline in overall employment levels for these highly exposed roles. This indicates a structural shift in corporate strategy. Instead of triggering the mass layoffs that many labor advocates feared, firms appear to be retaining their human headcount while using AI-driven efficiencies to justify capping labor costs and withholding standard pay increases.[1][3]

The financial burden of this wage compression is not distributed evenly. The Apollo data shows a highly regressive impact, with lower-income professionals bearing the brunt of the adjustment. Service workers—a category that includes concierges, childcare workers, and administrative support—faced a staggering 24.3 percent drop in real wage growth. Similarly, workers in the bottom 25 percent of earners saw their wage growth decline by 10.7 percent over the same period.[1][2]

In stark contrast, the highest-paid workers registered no significant negative impact on their wages. This divergence suggests that top earners are better positioned to absorb AI tools as productivity enhancers that augment their existing workflows, rather than as direct substitutes for their core tasks. The findings point to AI acting as a widening wedge for income inequality, insulating higher-paid professionals while squeezing the margins of those at the bottom of the wage scale.[1][3]

The distinction between who gets replaced and who gets augmented often comes down to the type of knowledge a job requires. Research from the Federal Reserve Bank of Dallas highlights the difference between 'codified knowledge'—established information found in textbooks or manuals—and 'tacit knowledge,' which is gained through hands-on experience and complex human interaction. AI excels at replicating codified tasks, substituting for entry-level or routine work, but struggles to replace the experiential judgment required in roles heavily reliant on tacit knowledge.[5]

This dynamic explains why some moderately AI-exposed professions have actually seen their pay increase. For example, personal finance advisors experienced an 8.4 percent increase in real wages, and administrative law judges saw wages surge 17.5 percent. In these roles, AI automates the routine data-gathering and preliminary analysis, freeing up the human professional to focus on high-value, complex decision-making that commands a premium in the market.

However, the headline 6.7 percent wage decline figure is not without its methodological skeptics. Independent analysis notes that Apollo's findings are highly sensitive to the statistical thresholds used to define 'high exposure.' When the exposure cutoff is broadened to include a wider array of jobs, the estimated wage hit shrinks to 1.89 percent. Under an even stricter cutoff, the negative effect becomes statistically insignificant, suggesting that the 6.7 percent metric should be viewed as a conditional warning rather than a universal law of the new economy.[4]

Furthermore, because the Apollo study relies on a difference-in-differences design comparing occupations rather than individual adopting firms, it cannot definitively prove that AI alone caused the relative wage gap. Broader post-pandemic macroeconomic forces, including inflation adjustments and shifting consumer demand, are also actively shaping relative pay scales across different sectors.[4]

Despite these methodological nuances, the broader trend remains clear. Apollo estimates that roughly 5.8 million American workers currently hold jobs with high AI exposure. As enterprise AI tools become more deeply embedded in corporate America, that number is expected to grow substantially. For the modern workforce, the immediate challenge is no longer proving that a human can do a job better than a machine, but proving that human expertise is still worth a premium raise.[1][2]