Published 17 October 2026
Anthropic co-founder Jack Clark has presented policymakers with a deliberately arresting possibility: under an extreme set of assumptions, an internal company model puts US economic output near $44tn by 2030, compared with a little over $30tn today. The number is not a consensus projection, and it should not be treated as one. Its real value is as a stress test for institutions built around gradual change.
Speaking during the IMF and World Bank annual meetings in Bangkok, Clark said the scenario depends on AI capabilities continuing to improve and diffusing rapidly through the economy. He linked the upside to faster scientific discovery, engineering and drug development, while warning that the same capabilities could amplify biological and cyber risks.
The scale of the scenario makes its assumptions decisive. Productivity gains do not appear automatically when a model becomes more capable. Firms must redesign workflows, workers need complementary skills, infrastructure must expand, and new products must create demand rather than simply displace existing activity. Bottlenecks in energy, data centres, chips and organisational adoption could slow the path considerably.
Distribution also matters. Even if aggregate output rose quickly, gains could accrue disproportionately to owners of compute, models and intellectual property. Workers in highly exposed occupations may face disruption before new roles emerge. Countries without reliable power, digital infrastructure or access to advanced systems could fall further behind. GDP alone would not reveal those transition costs.
That is why the policy response cannot wait for certainty about a single growth number. Governments need measurement systems capable of tracking AI-related productivity, labour displacement and market concentration in near real time. Competition authorities need to watch whether control of infrastructure becomes a barrier to entry. Education and social-insurance systems need mechanisms that can adjust faster than conventional legislative cycles.
Safety governance must develop on the same timetable. Clark argued for international cooperation and stronger human oversight, especially where AI touches biological research. The IMF’s broader discussion likewise framed AI as both a source of growth and a possible financial-stability risk if promised productivity fails to materialise and investment expectations unwind.
The sensible interpretation is neither technological euphoria nor dismissal. A scenario this extreme is useful precisely because it reveals which institutions break when change accelerates. The central leadership task is to build rules and public capacity that remain credible whether AI produces modest gains, a historic productivity surge or a volatile mixture of both.