Anthropic tells policymakers AI could take US output to $44 trillion by 2030, while measured returns stay thin

Anthropic’s Jack Clark told policymakers that artificial intelligence’s impact is accelerating quickly and could vastly boost US gross domestic product¹. Clark made the case to officials gathered for the International Monetary Fund’s meetings, where he cited a scenario in which US output reaches about $44 trillion by 2030^3,5. Reuters reported his argument that if GDP were to grow that much, standard economic doctrine would imply large consequences for policy^1,2.

The projection is the company’s own. It comes from Anthropic’s internal economic model, released in September 2026 as an economic scenarios explorer⁵. The explorer finds that AI drives GDP growth in all its scenarios, but that a larger share of the resulting prosperity might go to capital rather than to workers⁴.

Why it matters to investors

The figure is a vendor’s forecast presented to the institutions that set macro policy, which makes the provenance part of the story rather than a footnote. Anthropic’s own staff hold strong views: an internal survey found employees expect AI to handle 80% of their jobs by 2030⁷, and a separate exercise found 91% of its staff and 39% of US adults expect an AI Nobel by 2030⁶.

The measured return looks different. UK government research found that 75% of businesses using AI reported improved workforce productivity, but 77% reported no revenue change⁸. That gap, between reported productivity and booked revenue, is the one equity investors are actually underwriting when they pay current multiples for AI capital spending.

Ray Dalio’s bubble indicator puts AI at the level it reached in 1929 and 2000⁹. The same account carries the optimists’ case that AI could raise productivity tenfold⁹. Both views can be held at once: a technology can be transformative and its securities can still be mispriced.

What to watch

Whether the revenue gap in the UK survey narrows as deployments mature, since that is the variable that converts productivity claims into earnings⁸. Whether policymakers who heard Clark’s scenario treat it as a planning assumption or as an input from an interested party^1,3. And whether the distributional point in Anthropic’s own model, that capital may capture more of the gain than labour, becomes a political constraint on the build-out⁴.