What was published
Anthropic’s economics team released an interactive model on Thursday that projects the US economy to 2030 from two inputs: how much of knowledge work AI can do, and how fast firms actually adopt it. It sits alongside a technical report, Economic Scenarios for Transformative AI by Korinek and colleagues, and a survey of 10,980 US adults run in August.
The framework converts parameters — the fraction of tasks AI touches, how widely it is used, the productivity gain per task, and how far AI substitutes for labour rather than complementing it — into paths for productivity, growth, wages, the labour share and unemployment.
The three scenarios
In the modest case, AI has an effect on the scale of the internet. US GDP reaches $34.1 trillion in 2030, about 1.6% above the baseline projection, and knowledge-worker wages rise slightly.
In the substantial case, GDP reaches $36.3 trillion, 8.3% above baseline, and wages for knowledge workers are, in Anthropic’s phrase, essentially flat. Growth roughly doubles while pay does not move.
In the extreme case, GDP reaches $44.4 trillion — 32.4% above baseline — and knowledge-worker wages fall by more than 10% by 2030, with unemployment that Anthropic says could spike to historic levels. That path requires AI systems capable of improving themselves without human help.
The pattern across the three is the point. The faster the growth, the worse the outcome for the people whose work is being automated. Nothing in the model bends that relationship.
What the public expects
The survey found that respondents’ median expectations lined up with the substantial scenario — neither the internet-scale case nor the self-improving one. That is a striking result to publish next to the model, because it means the median American already expects the version in which the economy grows and their pay does not.
Read the caveats
Anthropic calls the model “a stark simplification of a complex reality” and lists what it leaves out: policy responses, business cycles, aggregate demand and financial-market disruptions, and catastrophic risks. It does not follow individual workers, and it stops at 2030 by design. The company also notes it may understate the demand effects of the data centre buildout.
Those omissions matter most in the extreme case, where the policy response is the whole question and the model does not model one. Read the scenarios as a way of making assumptions comparable, which is what Anthropic says they are, rather than as a forecast from a company with a direct commercial interest in the answer.
What to watch
The useful next step is other economists running the same dials and publishing where they disagree — on adoption speed above all, which is the parameter the scenarios are most sensitive to and the one furthest from settled.
