Inequality and the Distribution of Gains

AI could make the total economic pie substantially larger. But a larger pie does not automatically mean a more equitably shared one. Who captures AI's economic gains — and who bears its costs — depends on ownership structures, market dynamics, skill premiums, and policy choices. This lesson examines the mechanisms through which AI may concentrate or distribute its benefits, drawing on economic theory and early empirical evidence.

Capital vs. Labor: The Factor Share Question

Economists divide national income between two broad factors of production: labor (wages paid to workers) and capital (returns to owners of machines, software, real estate, and financial assets). In most advanced economies over the past four decades, the labor share of income has declined slightly while the capital share has risen — a trend partly attributed to prior waves of automation that substituted capital for labor. AI intensifies this dynamic. When a firm deploys AI that does the cognitive work previously done by employees, the gains flow to the firm's owners (capital) rather than to the displaced workers (labor). If AI is owned by a small number of firms — as is currently the case with frontier AI models — those firms' shareholders capture disproportionate gains. The workers whose tasks AI performs either lose their jobs or see their bargaining power reduced: they are competing with AI rather than with other workers. This is not hypothetical. The market capitalizations of leading AI firms have surged into the trillions while labor productivity gains in the broader economy have been modest. The gap between the fastest-adopting, AI-native firms and the rest of the economy is widening, suggesting an 'AI premium' for firms that can leverage frontier models — and a potential 'AI penalty' for competitors who cannot.

Within the labor market, AI does not affect all workers equally. The skill premium — the wage gap between high-skill and low-skill workers — has been a central driver of rising income inequality since the 1980s. The standard economic story is skill-biased technological change (SBTC): technology that complements high-skill workers while substituting for low-skill ones widens the premium and thus inequality. AI's relationship to the skill premium is more complicated than prior SBTC. Because AI is particularly capable at tasks that require broad knowledge but not genuine originality or interpersonal trust, it may actually compress the returns to certain types of high-skill work — the paralegal, the junior financial analyst, the entry-level coder — while leaving the returns to top-tier expertise intact or enhanced. If AI tools allow a senior expert to produce what previously took a team of junior workers, the premium goes to that expert, but the junior rungs of the career ladder are eliminated. Workers who cannot jump directly to senior expert roles face a missing middle. This produces a potentially counterintuitive result: AI could simultaneously reduce the value of some middle-skill professional work and increase the premium for the very top of each field — creating what economists call a 'superstar effect.'

Geographic inequality is an equally important dimension. AI development is concentrated in a small number of cities and countries — the San Francisco Bay Area, a few corridors of New York and London, Singapore, Beijing, Seoul. The workers and firms in these clusters capture most of the direct gains from AI development. Regions with heavy manufacturing, call centers, or routine clerical work — many of which are already economically fragile — face displacement with fewer locally available alternatives. At the international level, the gap may be even more stark. Advanced economies with large tech sectors and high educational attainment are better positioned to develop and deploy AI than lower-income countries whose labor cost advantage in routine tasks could be eroded by automation without a comparable opportunity to capture AI gains. The World Bank and IMF have both flagged that AI could worsen the income gap between high- and low-income countries if not actively managed.

Can AI Reduce Inequality?

The case for AI reducing inequality is not empty — it deserves honest engagement. Several mechanisms could operate in the other direction. First, the augmentation equalizer: research by Brynjolfsson, Li, and Raymond (2023) on customer-service AI found the largest productivity gains among the lowest-skilled workers, not the highest-skilled. If AI is a great equalizer within occupations — bringing low-skill workers up toward high-skill performance — it could compress wage inequality rather than expand it. Second, access to professional services: AI could make expert-level legal, medical, financial, and educational advice accessible to lower-income individuals who currently cannot afford it. A high-quality AI tutor that any student can access for free is an equalizing force if it genuinely replaces expensive private tutoring. Third, reduced barriers to entrepreneurship: AI tools for coding, design, marketing, and customer service lower the capital and skill requirements to start a business, potentially enabling more people to capture gains as owners rather than workers. Whether these equalizing forces outweigh the concentrating forces is an empirical question still being contested among researchers. The honest answer is: AI's effect on inequality depends enormously on policy choices — especially around taxation, access, and education — that have not yet been made.