Anthropic models three AI economies for 2030

Anthropic has published an interactive scenario explorer that pointedly refuses to forecast. Move the sliders for capability, adoption and autonomy, and US GDP in 2030 lands at $34.1 trillion, $36.3 trillion or $44.4 trillion — 1.6, 8.3 or 32.4 percent above a path with no AI at all.

How the model is built

Every job is treated as a bundle of tasks, drawing on the Department of Labor O*NET taxonomy. AI can make a task faster, take the task over, or create tasks that did not exist before. Which of those three dominates is what separates one scenario from the next.

The modest case is the internet all over again: real gains, inside historical norms. The substantial case has AI handling roughly half of knowledge work largely on its own by 2030, while adoption still lags across the economy. The extreme case assumes recursively self-improving systems, near-total automation of knowledge tasks, almost no new task creation, and 15 percent annual GDP growth.

Follow the labor share, not the GDP line

Labor takes about 60 percent of output today, capital 40 percent. By 2030 labor holds 59.4 percent in the modest case, 56.1 percent in the substantial one and 45.2 percent in the extreme one. Capital gains 0.6, 3.9 and 14.8 points respectively.

Wages tell the story more sharply than output does. They rise modestly in the first case. In the second, knowledge workers go flat while other workers gain. In the third, knowledge-worker pay falls by more than 10 percent by 2030 and unemployment spikes to levels normally seen only in recessions.

What people actually expect

The same sliders were put to two groups: 10,980 members of the US public and 11,811 visitors to the page, who were asked about capability, adoption, autonomy, productivity multipliers and how fast jobs adjust. The median set of answers implies GDP 10 percent above the no-AI path by 2030 and unemployment near 5 percent — close to the substantial scenario. About 10 percent of respondents land in extreme territory.

The caveats are load-bearing

Anthropic calls its own picture a stark simplification. Policy responses, business cycles, financial disruption and highly capable robots all sit outside the model, and it tracks occupations rather than individual workers. Whether heavily exposed occupations shrink or expand is left open.

Two framings belong with the numbers. They come from a model built by an AI lab with a stake in how this debate goes. And for this article they could only be checked against the publication itself — no second, independent newsroom had covered it by the time of writing.