AI Boom Quietly Reshapes Wages and Wealth in Opposite Directions

A growing body of empirical research from Wall Street and academic institutions indicates that the financial benefits of the AI boom are concentrating heavily among affluent Americans who own substantial equity portfolios, while the costs fall disproportionately on the rest of the population. This diverging landscape is challenging early projections that artificial intelligence would act as an economic leveler by boosting the productivity and earning potential of junior or non-credentialed workers. Instead, economists find that the technology is reinforcing existing capital-labor imbalances.

Between the first quarter of 2024 and the first quarter of 2026, U.S. household wealth expanded by approximately $21 trillion. Direct equity holdings accounted for half of that increase, despite representing only about 30% of total household assets. This wealth surge has been heavily propelled by AI-related equities, which Morgan Stanley equity strategists project will drive nearly 40% of S&P 500 earnings growth through 2027.

The ownership of these financial assets remains highly concentrated in a pattern that aligns with historical Federal Reserve Survey of Consumer Finances data. The top 20% of U.S. earners hold 87% of direct equity and mutual fund exposure, college-educated households hold 85%, and households over the age of 55 hold 79%.

According to an analysis by Morgan Stanley’s economics team led by Heather Berger, the households with the highest exposure to AI-driven workplace disruption are also the ones best insulated from its financial consequences. The bank’s research identifies a highly exposed demographic it terms “CHIC” households: college-educated, high-income, and city-dwelling. Nearly 70% of workers in occupations with high AI exposure hold at least a bachelor’s degree, compared to just 10% in low-exposure roles. The median salary in these high-exposure positions stands at $97,000, more than double the $46,000 median for low-exposure work.

A study by Apollo Global Management bypassed theoretical modeling to analyze real-world usage logs from the Anthropic Economic Index, which tracks interactions with the Claude AI model. Researchers Sania Edlich and Torsten Slok cross-referenced these logs with Bureau of Labor Statistics (BLS) wage data across 321 occupations from 2015 to 2025, using 2023—the year generative AI adoption accelerated—as a dividing line.

While wealthy professionals rely on stock portfolios to absorb potential disruptions, workers lower down the income scale are experiencing a quiet erosion of their earning power without losing their jobs. The Apollo study identified 11 highly exposed occupations. While employment levels in these fields remained stable, real wage growth after 2023 ran 6.7 percentage points slower than in low-exposure occupations.

For the bottom 40% of earners, total equity wealth roughly equals a single year of wages: $1.7 trillion in equities against $1.8 trillion in labor income. In contrast, for the top 20% of households, equity wealth is six times larger than annual labor income, standing at $49.8 trillion compared to $8.3 trillion in wages. According to Morgan Stanley’s financial modeling, this asset concentration creates a structural hedge: a top-earning household needs its investment portfolio to rise by just 4% to completely offset a 1% decline in labor income caused by AI-related workplace changes.

The wage deceleration was concentrated at the bottom of the pay scale. The lowest wage quartile saw real wage growth decline by 10.7 percentage points. The second wage quartile experienced a decline of 5.4 percentage points. The third wage quartile saw a decline of 4.0 percentage points. The top wage quartile showed no statistically significant wage effect.

Service-sector roles saw the sharpest estimated impact, with wage growth dropping by 24.3%, though the authors noted this figure relied on a limited sample size. Management and professional roles experienced a more modest 4.1% wage growth slowdown, while physically intensive blue-collar work showed no wage effects. Apollo estimates that roughly 5.8 million workers—representing 3.7% of the total U.S. labor force—are collectively losing about $28 billion annually in foregone wage growth, a figure the authors described as a conservative floor.

This pattern is corroborated by research from the IESE Business School. Analyzing a separate dataset covering 138 million workers, IESE researchers found that at firms adopting AI, starting salaries fell most sharply for entry-level and junior roles, decreased moderately for mid-level staff, and remained flat or increased for senior executives. Furthermore, AI-exposed companies reduced their hiring of junior personnel relative to mid-level employees, raising the barrier to entry for younger workers trying to establish a career.

This phenomenon aligns with the economic concept of “skill-biased technical change,” historically documented by labor economists like MIT’s David Autor. During the computerization waves of the 1980s and 1990s, technology hollowed out middle-skill administrative and manufacturing roles. In the AI era, however, the technology is acting as a substitute for the tasks typically assigned to junior, white-collar employees, reducing their market leverage and slowing their wage progression.

The stagnation in wages is occurring against a backdrop of historic corporate profitability, suggesting that companies are utilizing the narrative of AI-driven efficiency to expand margins rather than pass savings along to workers or consumers. A Morgan Stanley analysis of second-quarter 2026 corporate earnings by economist Michael Gapen showed that non-financial corporate profits rose by $400.9 billion. This surge pushed profit margins to 15.2% of gross value added, approaching post-World War II highs.

The data indicates this margin expansion was not driven by a sudden leap in productivity. Instead, it was a function of corporate pricing power: firms increased per-unit prices by an average of 2.30 cents, while their underlying labor and non-labor input costs remained virtually flat.

This shift comes at a time when the broader U.S. labor market is exhibiting unusual structural dynamics. David Kelly, chief global strategist at J.P. Morgan Asset Management, noted in a client memo that overall payroll growth has slowed to about half of its pre-pandemic rate, even as GDP growth remains steady. The U.S. unemployment rate sits at an 18-month low of 4.1%, but Kelly attributes this not to robust hiring, but to a shrinking labor force. An aging population, coupled with significant stock market gains that have enabled wealthier older workers to retire early, has quietly reduced the pool of active job seekers.

Concurrently, average wage growth has slowed to its lowest rate since May 2021. Kelly pointed out that despite low unemployment, actual hiring activity is subdued, which discourages workers from changing jobs to secure higher pay. This lack of labor mobility is compounded by a long-term decline in collective bargaining power; the private-sector unionization rate in the United States currently stands below 7%, down from historical peaks of over 30% in the mid-20th century. Without organized leverage, workers have been unable to claim a larger share of the corporate profits generated during the current tech cycle.

As these economic divisions widen, public discontent is increasingly focusing on the physical infrastructure required to power the AI boom: data centers. The vast electrical and water requirements of these facilities have put them at the center of localized environmental and economic battles. Organizations like the Electric Power Research Institute (EPRI) project that data centers could consume up to 9% of total U.S. electricity generation by 2030, roughly doubling current levels. This surging demand has already strained regional power grids, such as the PJM Interconnection in the Mid-Atlantic, prompting utility companies to delay the retirement of fossil-fuel plants and raise residential transmission rates to fund grid upgrades.

These rising costs have directly influenced public opinion. A Gallup poll conducted in May 2026 revealed that 70% of Americans oppose the construction of data centers in their local areas. A subsequent Politico poll in July found that only 16% of respondents believe data centers benefit their local communities, while approximately 60% believe these facilities cost more in resources than they return in economic value. The same poll showed that the proportion of Americans who believe data centers are driving up their household electricity bills rose from 43% in January to nearly 60% by July.

According to Public Citizen, 60% of Americans express distrust toward AI technology, and nearly 75% support government intervention to address AI-related job losses. A Pew Research Center survey similarly found that 57% of Americans feel the risks associated with AI outweigh its potential benefits. This public dissatisfaction has catalyzed organized, bipartisan political opposition. Data Center Watch reported that citizen opposition successfully blocked or delayed 75 data center projects valued at $130 billion during the first quarter of 2026 alone, matching the total number of disruptions recorded in all of 2025.

In July, New York became the first state to sign a statewide moratorium on certain large-scale data center operations. By August 2026, political candidates from both major parties in swing states like Michigan, Wisconsin, and Pennsylvania began running campaign advertisements targeting the expansion of data centers and the associated strain on local utility grids.

Despite these broad macroeconomic trends, researchers caution against assuming that an individual worker’s vulnerability can be predicted solely by their job title. A working paper published by the Federal Reserve Bank of St. Louis—which motivated the methodology of the Apollo study—highlights a significant gap between theoretical AI exposure and actual workplace adoption. Using a survey of nearly 14,000 workers, researchers Alexander Bick, Adam Blandin, David Deming, and Tyler Schumacher discovered that standard occupational exposure models only explain about half of the real-world variation in how workers use AI tools.

For instance, medical secretaries are classified as highly exposed by traditional models, which predict a 61% AI adoption rate. In practice, however, actual adoption stands at just 16.8% due to strict health privacy regulations and the high liability costs of automated errors. Conversely, professions such as computer repairers, special education teachers, and laundry workers are adopting generative AI tools at roughly twice the rates predicted by economists, finding niche applications that theoretical models failed to anticipate. The St. Louis Fed researchers concluded that individual AI adoption is driven less by demographic factors like age, education, or broad occupational categories, and more by personal experimentation and hands-on familiarity with the software.

While white-collar roles face significant task automation, Morgan Stanley researchers argue that these affluent families are “safer than advertised” due to their “elevated equity wealth.” Yet while individual usage remains fluid and unpredictable, the macro-level economic data suggests that the financial gains of the AI transition continue to flow along well-established lines of capital ownership and corporate pricing power, widening the gap between the nation’s asset owners and its wage earners.