Productivity Without Prosperity? Who Captures the Economic Gains from Workplace Artificial Intelligence?
Year: 2026
Type: preprint
Abstract: Experiments show that generative AI raises productivity on specific tasks by 14 to 40 percent, but who benefits when it spreads through the economy is far less clear. This study builds a capture accounting from task to economy. It links published field and survey evidence on productivity gains, adoption and wage pass-through to an exact decomposition of productivity gains in the U.S. nonfarm business sector, using the Bureau of Labor Statistics (BLS) release of 5 November 2026 and the Census Bureau’s Business Trends and Outlook Survey. Between the first quarter of 2024 and the third quarter of 2026, output per hour rose by 6.35 log points; 3.28 points (51.7 percent; rounding interval 50.3 to 53.1) reached workers as higher compensation relative to output prices, and 3.07 points went to nonlabor income as the labor share fell to 53.0 percent, the lowest since 1947. Labor’s capture fell from 84 percent in 2024 to 70 percent in 2025 and turned negative (-16 percent) in the four quarters to 2026 Q3, while nonfinancial corporate unit profits rose 16.1 percent. At the worker level, Danish administrative data show only 3 to 7 percent of productivity gains reaching earnings. A consistency bound combining these pass-through rates with plausible shares of AI in productivity growth implies that AI explains a minority of the rise in nonlabor capture (median 23 percent; 5th to 95th percentile 8 to 38 percent) and cannot be the whole story. The evidence is descriptive and aggregate, and the AI shares are bounds, not estimates. Purpose - This study asks who captures the economic gains from workplace AI, and how much of the recent shift in the distribution of productivity gains could plausibly be attributed to it. Design/methodology/approach - The paper combines published task-level, worker-level and firm-level evidence with an exact capture decomposition of BLS quarterly data, rounding-error Monte Carlo intervals, and a consistency analysis that bounds the AI contribution using pass-through rates and adoption arithmetic. Findings - Roughly half of the recent productivity gain went to workers and half to nonlabor income, with workers’ share falling over time and turning negative in the latest quarters. AI use is concentrated in large firms and in information and finance. Pass-through of AI-specific gains to workers is small, but AI can account for only a minority of the shift. Research limitations/implications - Aggregate indexes are rounded and revised, AI shares of productivity growth are not observed, and field studies are not representative of the economy. The analysis cannot identify causal effects. Practical implications - Bargaining institutions, training and transparency over how time savings are used are likely to determine capture, and policy should track labor’s share of gains alongside productivity. Social implications - Productivity growth need not translate into prosperity for workers, and young entrants in exposed occupations face distinct employment risk. Originality/value - The paper provides a transparent accounting that connects micro and macro evidence on AI’s distributional effects and quantifies what current data can and cannot say.
Source: Research Square
Author: Sarfaraz Abu Abdullah
Institution: University of Rajshahi
Language: English
Subfield: Economics and Econometrics
Domain: Social Sciences
Open Access status: green
OpenAlex ID: W7235568016
