Anthropic’s economists model three scenarios for AI’s macroeconomic impact through 2030 — Modest, Substantial and Extreme — against a no-AI counterfactual. Output rises in every case. The distributional results are where it gets interesting: average wages climb throughout, yet wages in cognitive occupations turn negative, and aggregate labor income barely moves despite an economy up to a third larger.
The scenarios
All figures below are drawn from Table 3 of the modelling. Percentages are deviations from the no-AI counterfactual path; the GDP index sets 2024 = 100.
Output
| No AI | Modest | Substantial | Extreme | |
|---|---|---|---|---|
| GDP, % above no-AI path | 0 | 1.6 | 8.3 | 32.4 |
| GDP, index (2024 = 100) | 112.7 | 114.5 | 122.1 | 149.3 |
| GDP growth, % per year | 2.0 | 2.4 | 5.4 | 15.4 |
Factor prices and income shares
| No AI | Modest | Substantial | Extreme | |
|---|---|---|---|---|
| Average wage, % above no-AI path | 0 | 0.7 | 2.1 | 9.7 |
| — cognitive occupations | 0 | 0.4 | −0.3 | −11.5 |
| — all other occupations | 0 | 1.1 | 5.9 | 33.6 |
| Net return to capital, % per year | 6.5 | 6.6 | 7.0 | 8.3 |
| Capital stock, % above no-AI path | 0 | 2.3 | 13.8 | 56.3 |
| Labor share, % of income | 60.0 | 59.4 | 56.1 | 45.2 |
| Capital share, % of income | 40.0 | 40.6 | 43.9 | 54.8 |
| Labor income, % above no-AI path | 0 | 0.6 | 1.4 | 0.5 |
| — wage bill, cognitive occupations | 0 | −0.3 | −4.6 | −31.0 |
| Capital income, % above no-AI path | 0 | 3.1 | 18.9 | 81.4 |
What the numbers say
Growth scales non-linearly
GDP gains run 1.6% to 8.3% to 32.4% above the no-AI path. Annual growth rises from 2.0% to 15.4% — roughly seven times the baseline rate in the Extreme scenario.
Cognitive wages diverge from everything else
Average wages rise in every scenario, from 0.7% to 9.7% above baseline. That average conceals a split. Wages in non-cognitive occupations rise sharply, reaching 33.6% above the no-AI path under Extreme, while cognitive wages turn negative from the Substantial scenario onward and end at −11.5%. The two series cross and never reconverge.
Aggregate labor income barely moves — by coincidence, not stability
Total labor income sits at 0.6%, then 1.4%, then 0.5% above baseline. The flatness of that last figure is the product of two large opposing movements: the cognitive wage bill falls 31% while non-cognitive earnings surge. The aggregate looks calm because the components nearly cancel. This is the most easily missed result in the table.
Income shifts from labor to capital
Labor’s share of income falls from 60.0% to 45.2%; capital’s rises from 40.0% to 54.8%, crossing the halfway mark only in the Extreme case. Capital income ends 81.4% above the no-AI path against labor’s 0.5%. The capital stock ends 56.3% higher, and the net return to capital rises from 6.5% to 8.3% per year — returns climb even as the stock expands.
A falling share is not the same as a falling income
Labor’s share collapses by nearly fifteen points, yet labor’s absolute income is roughly unchanged, because total output grew so much. Both readings are true. Treating either one alone as the headline misrepresents the result — and most charts of this data force exactly that choice, since standard chart types can show shares or levels but rarely both at once.
Caveats
These are model outputs under stated assumptions, not forecasts. The scenario labels describe assumed AI capability and diffusion, not assigned probabilities, and nothing in the table indicates which is more likely. The distributional results depend on how cognitive and non-cognitive occupations are defined and on the substitutability assumptions in the underlying production function. Transition dynamics and policy responses fall outside the table’s scope.