Technology · NEWS EXPLAINER
Anthropic’s $44 trillion AI figure is a scenario, not a forecast—five assumptions to check
Jack Clark highlighted an extreme 2030 outcome in Bangkok. Anthropic’s own paper says the scenarios carry no probabilities and shows why output, jobs and labor’s share can move in very different directions.
Anthropic co-founder Jack Clark used a striking number at the IMF and World Bank meetings: US output could approach $44 trillion by 2030 under certain AI assumptions. The company’s own technical paper makes the necessary qualification explicit—the scenarios are not predictions, and the extreme result depends on rapid capability gains, adoption and automation.
The $44 trillion figure is a conditional extreme scenario
Anthropic co-founder Jack Clark told officials in Bangkok on October 17 that an internal model could put US output near $44 trillion in 2030 under certain conditions. The underlying working paper is more precise about what that means: its scenarios are not predictions, no probabilities are attached, and the purpose is to compare the consequences of different assumptions. The largest number comes from the extreme-change case, not a base forecast for the economy.
Five assumptions drive the result
The model asks how much cognitive work AI can reach, how widely firms use it, how much productivity improves on affected tasks, whether use replaces or assists workers, and how quickly displaced workers find work elsewhere. In the extreme case, AI affects about 30% of all task value—roughly half of 2025 cognitive work—diffuses to 60% of those affected instances, more than doubles productivity on them, and automates 90% rather than augmenting people. Changing any of those inputs changes the output path.
Higher output does not mean evenly shared gains
The extreme scenario puts 2030 GDP about 32% above the no-AI path and annual growth at 15%, but it also shifts income toward capital. The paper’s modeled labor share falls from 60% to 45%, while cognitive employment is 21.5% below its mid-2026 level and unemployment among workers who began in cognitive occupations reaches 17.9%. Those figures are model outputs tied to extreme assumptions, not observed results or a promise that transfers will offset losses.
The less extreme cases look very different
Under modest change, the paper says AI adds less than half a percentage point to GDP growth by 2030 and raises unemployment by about one tenth of a point. Answers from a survey of 10,980 US adults line up more closely with the substantial-change scenario: by 2030 GDP is about 8% above the no-AI path and cognitive employment is about 4% lower. The wide spread is the central finding; treating only the largest number as the forecast removes the model’s main warning.
What readers should watch between now and 2030
The authors identify measurable signals that can show which path is becoming more plausible: the share of tasks models can perform, real adoption across firms, productivity gains per task, the balance of automation and augmentation, and the time it takes displaced workers to move into other occupations. They also note important limits: the framework focuses on cognitive work, largely excludes fast robotics progress, and does not model every feedback between research and automation. The next useful evidence is therefore measured diffusion, wages, hiring and task-level productivity—not another headline-sized GDP number.
