AI could boost U.S. GDP by up to 32% by 2030 but disrupt jobs, study says

The Anthropic Institute working paper puts U.S. gross domestic product between 1.6% and 32.4% above a baseline without AI by 2030. In its most extreme scenario, overall unemployment reaches 11.9%, with joblessness approaching 18% among workers initially employed in cognitive occupations.

Released in September, the paper examines modest, substantial and extreme scenarios for AI’s effects on growth, wages and employment between 2026 and 2030. Its authors, Anton Korinek, Charles I. Jones, Szymon Sacher, Tess Cotter and Peter McCrory, emphasize that these are not forecasts and assign no probabilities to them. They also state that their views do not necessarily reflect those of Anthropic, the U.S. AI developer affiliated with the institute.

The model divides workers into two groups. Cognitive occupations include management, professional, sales and office jobs, which accounted for 62.4% of U.S. employment in 2025. The other group includes construction, repair, transportation and personal care, which the model assumes AI does not directly affect. Workers displaced from cognitive roles can move into the second group, but finding new employment takes time.

The sharpest changes occur in the extreme scenario, where AI performs almost half of today’s cognitive work by 2030 and more than doubles productivity on the tasks it handles. GDP reaches 32.4% above the no-AI baseline, while annual growth accelerates to 15.4% in the year to 2030, compared with 2% without AI.

Those gains are distributed unevenly. Labor’s share of national income falls from 60% to about 45%, while capital’s share rises to nearly 55%. Total labor income remains close to its no-AI baseline, meaning the additional output accrues to capital owners, whose income is 81% higher.

Employment in cognitive occupations falls 21.5% from its mid-2026 level. Unemployment among workers who began in those occupations reaches 17.9%, pushing the overall rate to 11.9%.

Pay also diverges. Wages in cognitive occupations fall 11.5% below the no-AI baseline, leaving them slightly lower in 2030 than in mid-2026. Wages in other occupations, where labor becomes relatively scarce within the model, rise 33.6% above the baseline.

The authors calculate that transfers equivalent to about 9% of GDP, roughly the combined size of Social Security and Medicare, could keep cognitive workers’ income at its no-AI level while leaving the rest of the economy more than 20% better off.

However, they caution that no technological shift has been accompanied by compensation on that scale. Drawing on research into the regional effects of Chinese import competition, they argue that economic gains do not automatically translate into compensation for those who lose out.

The substantial scenario produces smaller but still significant changes. GDP stands 8.3% above the no-AI baseline by 2030, and annual growth reaches 5.4%. Labor’s share of income falls to about 56%, a four-percentage-point decline that the authors compare with the reduction over the four decades after 1980.

Cognitive employment falls 3.9% in this scenario, while unemployment among workers initially in those occupations rises from 2.9% to 4.5%. Their wages remain close to the no-AI baseline, while pay in other occupations is 5.9% higher.

The modest scenario leaves the economy near its baseline trajectory. GDP is 1.6% higher by 2030, annual growth reaches 2.4%, and overall unemployment stands at 3.9%, compared with a baseline of 3.8%.

To examine public expectations, the researchers surveyed a representative sample of 10,980 U.S. adults in August 2026. Feeding the median responses into the model produced results close to the substantial scenario: GDP 8.6% above the no-AI baseline and overall unemployment of around 4.6%.

Respondents differed sharply on AI’s potential. About 30% expected it to save no time on tasks suited to the technology, while 49% expected it to cut the time required by at least half. The median respondent estimated that an AI-displaced worker would need about eight months to find a job in a new occupation, compared with roughly three months in normal times.

The findings depend heavily on the model’s assumptions. If new capital, much of it computing power, becomes available more slowly, average wages in the extreme scenario fall below the no-AI baseline instead of rising. Allowing wages to adjust more freely reduces unemployment but produces much steeper pay cuts in cognitive occupations; more rigid wages protect pay but increase joblessness.

Faster scientific discovery contributes relatively little within the period studied, adding well under 1% to labor productivity before 2030 even in the extreme scenario.

The framework excludes advances in robotics, catastrophic risks, political dynamics, business cycles and possible financial market disruptions. It also does not examine outcomes beyond 2030. All three scenarios begin with the same mid-2026 measures of AI use, with most of their divergence occurring after 2027.

The authors say the economy’s direction may become clearer within a year or two. They identify retraining, income support, universal basic capital and universal basic income as possible responses if substantial disruption emerges, highlighting that higher output alone does not ensure workers share in the gains.

Earlier, Qazinform reported that a senior AI researcher who spent the past three years building foundational models at tech giants OpenAI and Anthropic resigned, issuing a stark warning that leading industry labs are rushing recklessly toward uncontrollable superintelligent systems.