Anthropic经济指数报告：地域和企业AI采用的不均衡性

Published:September 15, 2025 Authors:Ruth Appel*, Peter McCrory*, Alex Tamkin*Miles McCain, Tyler Neylon, Michael Stern Acknowledgements: Helpful comments, discussions, and other assistance: Alex Sanchez, Andrew Ho, Ankur Rathi, Asa Kittner,Ben Merkel, Bianca Lindner, Biran Shah, Carl De Torres, Cecilia Callas, Daisy McGregor, Dario Amodei, DeepGanguli, Dexter Callender III, Esin Durmus, Evan Frondorf, Heather Whitney, Jack Clark, Jakob Kerr, JanelThamkul, Jared Kaplan, Jared Mueller, Jennifer Martinez, Kaileen Kelly, Kamya Jagadish, Katie Streu, KeirBradwell, Kelsey Nanan, Kevin Troy, Kim O’Rourke, Kunal Handa, Landon Goldberg, Linsey Fields, Lisa Cohen,Lisa Rager, Maria Gonzalez, Mengyi Xu, Michael Sellitto, Mike Schiraldi, Olivia Chen, Paola Renteria, RebeccaJacobs, Rebecca Lee, Ronan Davy, Ryan Donegan, Saffron Huang, Sarah Heck, Stuart Ritchie, Sylvie Carr, TimBelonax, Tina Chin, Zoe Richards *Lead authors. Contributed equally to this report. Introduction AI differs from prior technologies in its unprecedented adoption speed. Inthe US alone, 40% of employees report using AI at work, up from 20% in2023 two years ago.1Such rapid adoption reflects how useful this technologyalready is for a wide range of applications, its deployability on existing digitalinfrastructure, and its ease of use—by just typing or speaking—withoutspecialized training. Rapid improvement of frontier AI likely reinforces fastadoption along each of these dimensions. Historically, new technologies took decades to reach widespread adoption.Electricity took over 30 years to reach farm households after urbanelectrification. The first mass-market personal computer reached earlyadopters in 1981, but did not reach the majority of homes in the US for another20 years. Even the rapidly-adopted internet took around five years to hitadoption rates that AI reached in just two years.2 Why is this? In short, it takes time for new technologies—even transformativeones—to diffuse throughout the economy, for consumer adoption to becomeless geographically concentrated, and for firms to restructure businessoperations to best unlock new technical capabilities. Firm adoption, firstfor a narrow set of tasks, then for more general purpose applications,is an important way that consequential technologies spread and havetransformative economic effects.3 In other words, a hallmark of early technological adoption is that it isconcentrated—in both a small number of geographic regions and a smallnumber of tasks in firms.As we document in this report, AI adoption appearsto be following a similar pattern in the 21st century, albeit on shorter timelinesand with greater intensity than the diffusion of technologies in the 20thcentury. To study such patterns of early AI adoption, we extend the AnthropicEconomic Index along two important dimensions, introducing a geographicanalysis of Claude.ai conversations and a first-of-its-kind examination ofenterprise API use. We show how Claude usage has evolved over time, how adoption patterns differ across regions, and—for the first time—how firms aredeploying frontier AI to solve business problems. Changing patterns of usage onClaude.ai over time In the first chapter of this report, we identify notable changes in usage onClaude.ai over the previous eight months, occurring alongside improvementsin underlying model capabilities, new product features, and a broadening ofthe Claude consumer base. We find: •Education and science usage shares are on the rise.While the use of Claudefor coding continues to dominate our total sample at 36%, educational taskssurged from 9.3% to 12.4%, and scientific tasks from 6.3% to 7.2%.•Users are entrusting Claude with more autonomy.“Directive”conversations, where users delegate complete tasks to Claude, jumped from27% to 39%. We see increased program creation in coding (+4.5pp) and areduction in debugging (-2.9pp)—suggesting that users might be able toachieve more of their goals in a single exchange. The geography of AI adoption For the first time, we release geographic cuts of Claude.ai usage data across150+ countries and all U.S. states. To study diffusion patterns, we introduce theAnthropic AI Usage Index (AUI) to measure whether Claude.ai use is over- orunderrepresented in an economy relative to its working age population. We find: •The AUI strongly correlates with income across countries.As withprevious technologies, we see that AI usage is geographically concentrated.Singapore and Canada are among the highest countries in terms of usage percapita at 4.6x and 2.9x what would be expected based on their population,respectively. In contrast, emerging economies, including Indonesia at 0.36x,India at 0.27x and Nigeria at 0.2x, use Claude less.•In the U.S., local economy factors shape patterns of use.DC leads per-capita usage (3.82x population share), but Utah is close behind (3.78x). We see evidence that regional usage patterns reflect distinctive features of thelocal economy: For