The bigger pie, the same slice
Anthropic's new Economic Scenario Explorer models three paths for the US economy through 2030. In the most extreme, output is a third larger than it would otherwise be, unemployment nears 12 percent, and labour's share of national income falls so far that total wages are no higher than in a world without AI at all.
TL;DR
- Anthropic's economics team published "Economic Scenarios for Transformative AI" on September 10, alongside an interactive Econ Scenario Explorer and a survey of 10,980 US adults, modelling three paths for the US economy through 2030.
- The scenarios range from "modest" (GDP just 1.6 percent above a no-AI baseline) to "extreme" (GDP 32.4 percent higher, unemployment near 12 percent, and white-collar employment down more than 20 percent).
- In the extreme case, labour's share of national income falls from about 60 percent to roughly 45 percent, meaning total wages paid to workers are no higher than they would be in a world without AI, even though the economy itself is a third bigger.
Most AI economic forecasts land on a single number. Anthropic's economics team instead built a model that maps how far AI capabilities advance, and how widely they are adopted, into three distinct paths for US GDP, employment, wages, and the split between labour and capital through 2030. The result, published alongside a public interactive tool and a survey of nearly 11,000 US adults, is less a prediction than a way to reason about a wide range of outcomes that all remain plausible from where the technology stands today.
Three paths, one shared shape
| Scenario | AI's share of tasks by 2030 | GDP vs. no-AI baseline | Unemployment rate | Knowledge-work employment |
|---|---|---|---|---|
| Modest | Limited, internet-scale disruption | +1.6% | Up about 0.1 point | Little change |
| Substantial | ~12% of all tasks | +8.3% (~$36.3T) | ~4.6% | Down ~4% |
| Extreme | ~1/3 of all tasks | +32.4% (~$44.4T) | Nears 12% | White-collar down 20%+ |
Every scenario shares a structure, not just a headline number: AI automates a rising share of cognitive tasks, productivity and output climb, and displaced workers must search for jobs elsewhere in the economy. What changes between scenarios is the speed and breadth of that automation, and therefore how much of the gain shows up as growth versus how much shows up as displacement. Physical-world occupations, the model finds, see wage gains in all three paths; the disruption concentrates in screen-based, cognitive work.
The pie grows. The workers' slice doesn't.
The paper's most striking finding isn't a growth number at all. Today, roughly 60 percent of US output flows to labour and 40 percent to capital. In the extreme scenario, that split moves to roughly 45 percent labour and 55 percent capital, even as the overall economy grows by a third relative to the no-AI counterfactual.
Under the extreme scenario, the US economy in 2030 is about a third larger than it would otherwise have been, but the total wages flowing to workers are no greater than they would be in a world where transformative AI never arrived at all.
That is the mechanism worth sitting with: growth and worker income can diverge, not just slow down together. A bigger economy is not, on its own, evidence that ordinary compensation is keeping pace with it.
Where the public already stands
The accompanying survey suggests the public isn't waiting for 2030 to form a view. Asked when AI will match a skilled professional on eight specific tasks, how widely it will be used, and how long a displaced worker would take to find new work, the median respondent's answers implied a 2030 outlook close to Anthropic's "substantial" scenario: GDP about 10 percent above baseline and unemployment around 5 percent. Roughly one in ten respondents already hold views consistent with the extreme case. Over half, 53 percent, said AI can already write routine business emails and documents as well as a professional; only 24 percent said the same of building and maintaining a working software product.
What this means for leaders
- Plan against a range, not a point estimate. Stress-test workforce and capital plans against the "substantial" and "extreme" cases, not only the "modest" case that most resembles today.
- Track the labour-capital split inside your own organisation, not just headcount or output. A productivity gain that flows entirely to margin looks identical to a productivity gain that flows to workers on a revenue chart, but not on a compensation one.
- Weight workforce risk toward screen-based, cognitive roles; the model finds physical-work wages hold up or rise across all three scenarios.
- Revisit comp and retention strategy against public expectations, not just internal forecasts. The median US adult's own forecast already resembles Anthropic's middle scenario, not its mildest one.
Related reading
- Anthropic says AI could bring both 15% growth and mass unemployment by 2030 (Euronews)
- When AI capex becomes macro: what the buildout means for buyers
Source: Scenarios for our Economic Future, Anthropic.