Anthropic Working Paper Maps Three AI Futures: 1.6% to 32% GDP Uplift by 2030
The Anthropic Institute published a working paper this week that puts hard numbers on three possible AI futures between now and 2030. The authors — Anton Korinek, Charles I. Jones, and three co-researchers — built a task-based economic model that converts a handful of parameters into quantitative paths for GDP growth, wages, the labor share of income, and unemployment.
The paper is not a forecast. The authors are explicit: no probabilities are attached to any scenario. The point is a structured comparison tool — a way to read the wide range of published predictions using a single consistent framework.
The Three Scenarios
Modest change: AI adds under half a point to annual GDP growth. By 2030, GDP is 1.6% higher than the no-AI baseline. Unemployment rises by one tenth of a percentage point. Cognitive wages are barely moved. This is roughly consistent with Acemoglu (2025) and current OECD figures.
Substantial change: GDP is 8.3% higher by 2030 — consistent with the Goldman Sachs and McKinsey-era 2023 forecasts that circulated widely. Cognitive employment falls 4%. The median US adult surveyed by the authors expects this scenario.
Extreme change: GDP is 32% higher by 2030, with annual growth running near 15%. The labor share of income falls from 60% today to 45%. Nearly one in five cognitive workers is unemployed. The cognitive wage is 11.5% below its no-AI path; wages in manual occupations are 34% above it.
Key Numbers
| Scenario | GDP uplift vs. baseline | Cognitive employment | Labor share |
|---|---|---|---|
| Modest | +1.6% | ~flat | ~flat |
| Substantial | +8.3% | -4% | modest decline |
| Extreme | +32% | ~-20% | 60% → 45% |
The Model Structure
The framework treats cognitive workers — management, professional, sales, office — as directly exposed to AI automation and augmentation. Construction workers, electricians, and other physical occupations are not. Displaced cognitive workers face search frictions when moving to other sectors, which generates sustained unemployment rather than instant reabsorption.
Capital demand rises as AI substitutes for labor, which raises capital’s return and share of total factor income. That is the mechanism behind the falling labor share in all three scenarios.
One counterintuitive finding: the boost from AI accelerating research and innovation is modest even in the extreme change scenario. The model follows semi-endogenous growth theory, which predicts that physical research bottlenecks cap this channel. The growth acceleration comes almost entirely from automation of cognitive work, not from AI-generated ideas compounding faster.
The timing matters: “almost all of the divergence comes after 2027.” In all three scenarios, 2026 and 2027 look similar. The scenarios only spread apart in the final three years.
The Survey
The authors surveyed US adults on their AI expectations. Results tracked the substantial scenario most closely — median respondents expect AI to raise GDP by roughly 8% and cut cognitive employment by around 4% by 2030. Views varied widely across the full sample, with meaningful clusters expecting both the modest and extreme outcomes.
Anthropic released an interactive scenario explorer at anthropic.com/institute/econ-scenarios where users can adjust diffusion and productivity parameters and trace the implied economic outcomes directly.
The paper is Working Paper No. 2026-02 from the Anthropic Institute, published September 2026.