A very large number, offered by its own author with a warning attached
Jack Clark did not present $44 trillion as a modest estimate. Speaking to the IMF and World Bank annual meetings in Bangkok on Saturday, the Anthropic co-founder told hundreds of government and finance officials that one of his company's internal economic models puts US output at $44 trillion by 2030, against a current level of just over $30 trillion. Then he supplied the review of his own figure: it "sounds very outlandish, I know."
That self-awareness is doing real work in the story, and it should not be mistaken for verification. Clark's position and the setting give the claim prominence, but neither supplies the missing methodology. The model behind the number is internal and unpublished. As reported, he gave no methodology, no range, and no statement of whether the projection is in real or nominal terms. A reader who wants to check the claim has nothing to check. There is only the number, the company that produced it, and the man presenting it.
The conditions leave the hardest questions unanswered
The conditions Clark did state are worth reading slowly. The forecast assumed "the technology continuing to get better." It also requires, in his words, that new industries and inventions be created to power such a major economic expansion. What the reported remarks do not establish is where those conditions sit in the reasoning: whether they are inputs the model was fed, outputs the model predicts, or Clark's own explanation of what growth on that scale would take. Nor do they show how the developments were modelled, how much growth each one contributes, or which assumptions produce the $44 trillion endpoint rather than a smaller one.
Put the two published numbers next to each other and the scale becomes concrete. Going from just over $30 trillion to $44 trillion by 2030 lifts the annual level of output by roughly $14 trillion in four years. On an approximate $30 trillion starting point, that is compound growth of about 10 percent a year on average — and because the baseline is just over $30 trillion, the implied average is slightly under that. This is our arithmetic on Clark's reported endpoints, not a growth rate he supplied. An average also says nothing about the path between the endpoints: individual years could run slower, faster, or even negative and still arrive at the same total. He offered the endpoints. The distance between them is the part of the claim that needs defending, and it is the part that went undefended on Saturday.
None of this makes the direction wrong. AI is plainly raising output in some kinds of work, and Clark pointed to scientific discovery, where he said researchers are being supplemented by hundreds or thousands of AI assistants. A direction is not a magnitude, though. The distance between "AI will add materially to growth" and "$44 trillion by 2030" is where forecasts go to be tested, and testing requires the model.
The rest of the lecture was more specific than the headline
The GDP figure will travel furthest, which is a pity, because several other things Clark said were narrower and easier to act on. He acknowledged job losses in fields with "high AI exposure," naming call centres and customer service, and said scientific research was being enhanced rather than reduced, with the jobs in entirely new industries a major unknown. He called for international cooperation to build "unusual and serious rules," said the window for imposing safeguards was now "a small number of years," and singled out biological research as the area needing the strictest human oversight and transparency; in his telling, AI has already created new viruses there. That last point is his claim about emerging capability and risk, reported as his claim. It was not presented with independently verified findings. He also said there was growing evidence of a "singularity" in which AI development outpaces human control, while maintaining that AI remains under human control for now — again, his assessment, offered as such.
Those claims can be argued with, measured, or legislated against. The $44 trillion figure, as delivered, cannot. It can only be repeated.
Reuters noted in the same report that Anthropic is preparing for an initial public offering expected in November, which could value the company at up to $2 trillion. That is context, not an accusation. Executives speak at the IMF for many reasons, and a projection of US GDP is not a forecast of Anthropic's own market, revenue or valuation. The commercial setting neither validates nor discredits the number, and the criticism does not need it: an unpublished projection cannot be evaluated from outside, whoever presents it.
Sources
- [1] Reuters (David Lawder), "AI rapidly improving, could vastly boost US GDP, Anthropic's Clark tells policymakers", Bangkok, October 17, 2026 (published 2026-10-17T09:14:49Z, updated 11:50:01Z; primary source for the facts in this article: the $44 trillion figure and the just-over-$30-trillion baseline, the "outlandish" qualifier, the stated conditions, the jobs remarks scoped to high-exposure fields, the safeguards window, the biological-research remarks, the singularity remark, and the November IPO context); read in full via Reuters syndicationRead source
- [2] Reuters via TradingView / Seeking Alpha syndication, "Anthropic co-founder Clark projects U.S. GDP to hit $44T by 2030 thanks to AI", October 17, 2026 (cross-check of the $44T figure, the baseline, the "outlandish" quote and the new-industries condition)Read source
- [3] AI Affairs, "Anthropic's Clark tells IMF and World Bank AI could lift US output", October 17, 2026 (secondary cross-check, including Clark's head-of-public-benefit role and the job-loss remarks)Read source