Anthropic’s economics team released three modeled scenarios describing how artificial intelligence might affect the U.S. economy. The company stresses these are models, not forecasts; results vary depending on how deeply AI is embedded into work processes, how quickly firms adopt it, and whether AI complements or replaces workers.
1. Modest change: internet-like impact
In the “modest change” scenario the economic impact of AI is roughly comparable to the internet’s. AI remains largely a useful technology but does not cause radical disruption. Under this model, U.S. GDP in 2030 would be $34.1 trillion, only 1.6% higher than the path without AI, and unemployment would change very little.
2. Significant change: higher growth, labor reallocation
The “significant change” scenario assumes AI can autonomously perform many more work processes. With notable productivity gains, GDP would reach $36.3 trillion in 2030 — 8.3% above the no-AI baseline — and annual economic growth could hit 5.4%, roughly double typical peacetime rates. At the same time, knowledge-based employment would contract by about 4% in this model.
3. Extreme change: huge GDP rise and severe job losses
The “extreme change” scenario assumes AI is used in 30% of all tasks — roughly half of today’s knowledge work. Productivity on those tasks could more than double, and 90% of AI-executed work would be fully automated. As a result, the model projects U.S. GDP of $44.4 trillion in 2030, 32.4% above the no-AI path. Annual growth could reach 15% — a pace only seen during the U.S. mobilization in the early 1940s.
However, this scenario comes with heavy social costs: employment would fall by 21.5%. Among entry-level workers in knowledge occupations unemployment would reach 17.9%, while economy-wide unemployment would rise to 11.9%.
Who gains: labor versus capital
The models also show that economic growth and worker well-being need not move together. Using a simplified comparison, currently about 60% of U.S. income is associated with labor and roughly 40% with capital. In the extreme scenario that balance shifts: labor’s share falls to about 45.2% while capital’s share rises to about 54.8%. That implies a growing portion of AI-created surplus could accrue to owners of technology, compute capacity, data centers and firms.
Wage effects would be mixed. The wages of workers whose jobs are not displaced by AI — for example plumbers, electricians, construction workers and other roles tied to physical skills and in-person presence — could be 33.6% higher than in a no-AI economy. By contrast, wages for knowledge workers could fall by 11.5%, a negative outcome for programmers, accountants, lawyers, analysts and many office employees whose valuable tasks are increasingly automatable.
How plausible are the extreme outcomes?
Anthropic notes the most severe results require several pessimistic assumptions to hold simultaneously: AI becoming extremely versatile, rapid and widespread corporate adoption, 90% displacement of knowledge work by AI use, lack of new human tasks emerging, and displaced workers failing to find other employment. At present, the U.S. labor market does not resemble the extreme world in the model.
Common denominator: AI raises GDP, but distribution matters
All three Anthropic scenarios agree that AI increases overall U.S. economic output. The difference is scale and distribution: the modest path looks like another technological advance, the significant path implies major economic and labor-market restructuring, and the extreme path would mark a wholly new economic era. The central policy question is not whether AI can make the U.S. richer — all three models answer “yes” — but whether people broadly will become wealthier along with the economy or whether gains will concentrate with those owning the technology and capital.
