Anthropic’s latest Economic Index report reveals that artificial intelligence is being adopted at an unprecedented rate, with significant geographic and sectoral disparities, raising concerns about potential economic divergence.
The study, titled “Uneven Geographic and Enterprise AI Adoption,” draws from extensive data on Claude.ai usage and enterprise API traffic, documenting how AI is transforming workflows in concentrated pockets. The report builds on prior iterations by incorporating geographic breakdowns across more than 150 countries and all U.S. states, alongside a pioneering examination of first-party (1P) API usage. This expansion allows researchers to track not only consumer patterns but also how businesses are programmatically integrating frontier AI models like Claude into operations.
The report’s findings are grounded in anonymized, aggregated data from millions of interactions, mapped to occupational taxonomies such as O*NET, and emphasize the dual nature of AI as both an automation tool and a productivity enhancer. At the core of the report is the observation that AI’s rollout is accelerating faster than historical precedents. In the United States, employee AI usage at work has nearly doubled, climbing from 20% in 2023 to 40% by September 2025, according to Gallup data cited in the report.
This surge outpaces the diffusion of transformative technologies like electricity, which took over 30 years to reach rural U.S. households after urban adoption, or personal computers, which required two decades to penetrate a majority of homes following their 1981 debut. Even the internet, often hailed for its rapid spread, needed about five years to achieve similar penetration levels. Such velocity stems from AI’s inherent advantages: its broad applicability across tasks, seamless integration with existing digital tools, and intuitive interfaces that require no specialized training—merely typing or speaking prompts.
The report attributes further momentum to the swift advancements in frontier models, which continually expand capabilities and attract a wider user base. However, this early-stage enthusiasm masks underlying concentrations: AI use remains focused on a limited set of tasks within firms and is geographically clustered, echoing patterns seen in 20th-century innovations but compressed into shorter timelines. To quantify these dynamics, the report introduces the Anthropic AI Usage Index (AUI), a metric that compares Claude.ai conversation volumes to working-age populations in specific regions.
This index reveals a strong correlation between per-capita AI adoption and economic income levels, signaling potential risks for global inequality. High-income nations like Singapore and Canada lead with AUI scores of 4.6 times and 2.9 times expected usage, respectively, based on population size. In contrast, emerging economies lag significantly: Indonesia registers 0.36 times expected use, India 0.27 times, and Nigeria just 0.20 times. Within the U.S., adoption hotspots mirror local economic strengths. Washington, D.C., tops the list at 3.82 times expected usage, driven by demands in document editing and career assistance amid its policy and professional services hub.
Utah follows closely at 3.78 times, benefiting from a burgeoning tech ecosystem. California exhibits elevated IT-related applications, while Florida sees heavier reliance on financial services tasks. These regional variations illustrate how AI deployment is tailored to sectoral needs, with coding dominating in tech-heavy areas and administrative functions prominent in service-oriented ones. Diving deeper into usage patterns, the report charts an evolution in Claude.ai interactions over the past eight months, coinciding with model upgrades and feature enhancements.
Coding remains the largest category at 36% of total usage, underscoring AI’s role in software development. However, non-technical applications are gaining ground: educational tasks have risen from 9.3% to 12.4%, reflecting students and professionals leveraging AI for learning and research. Scientific tasks have similarly increased from 6.3% to 7.2%, pointing to growing integration in data analysis, simulations, and hypothesis testing. A notable shift is the rise in “directive” conversations, where users delegate full tasks to Claude rather than engaging in iterative exchanges.
These automation-oriented interactions have jumped from 27% to 39% of sessions. Within coding, this manifests as a 4.5 percentage-point increase in program creation and a 2.9 percentage-point decline in debugging requests, suggesting users are achieving outcomes more efficiently in single interactions. This trend aligns with AI’s maturation, enabling higher autonomy and reducing the need for human oversight in routine processes. Geographic disparities extend beyond raw adoption rates to the diversity and style of usage.
In low-AUI countries like India, coding accounts for over 50% of interactions—far exceeding the global average of about one-third—indicating a narrow focus on technical applications amid limited access to broader tools. High-adoption regions, conversely, display more varied portfolios: education, science, and business tasks each claim significant shares, fostering comprehensive productivity gains. After adjusting for task composition, the report uncovers divergent collaboration modes.
Low-AUI areas lean toward automation, with users more frequently offloading complete tasks to AI. High-AUI regions, however, favor augmentation—patterns involving learning, iteration, and human-AI teamwork—which may amplify long-term skill development and innovation. This bifurcation raises equity concerns: while automation streamlines efficiency in resource-constrained settings, augmentation in affluent areas could widen knowledge gaps and economic divides.
Shifting to enterprise contexts, the report provides unprecedented visibility into 1P API traffic, which represents programmatic access to Claude by businesses and developers. Unlike the chat-based Claude.ai, API usage reveals specialized, scalable deployments. Coding again dominates, but API patterns diverge: they show higher concentrations in coding and office/administrative tasks, while Claude.ai skews toward educational and writing activities.
This reflects enterprises prioritizing backend automation over consumer-facing creativity. Automation prevails in API scenarios, comprising 77% of business uses compared to roughly 50% on Claude.ai. The programmatic interface facilitates seamless integration into workflows, such as generating reports or processing data without user intervention. Yet, the report notes that cost does not appear to be a primary barrier; frequently used tasks often incur higher expenses due to computational demands, indicating low price sensitivity.
Instead, deployment decisions hinge on model capabilities and the tangible value of automating specific functions, such as reducing manual labor in high-stakes domains. A key bottleneck identified is contextual data curation. For complex enterprise applications—like legal analysis or supply chain optimization—AI’s effectiveness depends on providing rich, relevant context.
The report suggests that many firms face hurdles in data modernization and organizational restructuring to supply this input, potentially stalling broader adoption. Investments in these areas could unlock AI’s potential in sophisticated sectors, but they represent significant upfront costs, particularly for smaller enterprises. These insights are bolstered by the report’s open-sourcing of its dataset, a commitment to transparency that invites independent scrutiny.
The release includes task-level classifications for both Claude.ai and 1P API data, collaboration breakdowns, and geographic details for consumer usage. Researchers can now explore pressing questions: How does AI adoption impact local labor markets? What policies can democratize access in low-adoption regions? Does task cost influence enterprise strategies, and which worker profiles benefit most from automation versus augmentation?
Historically, transformative technologies like electrification and the internal combustion engine drove modern economic growth but initially exacerbated global inequalities, as documented in works by economists Robert Gordon and Lant Pritchett. AI risks a similar trajectory: if productivity boosts accrue primarily to high-adoption economies, recent trends of growth convergence—evidenced by studies from Michael Kremer and others—could reverse, entrenching divides between rich and emerging nations.
Within firms, uneven task adoption could reshape employment landscapes. Automation may displace entry-level roles in coding or admin, while augmenting experienced workers with organizational knowledge, potentially elevating wages for the latter. The report cites research by David Autor and others on technology diffusion, emphasizing that early concentrations often precede widespread transformation as complementary innovations emerge.
Anthropic’s analysis arrives at a pivotal moment, as frontier models like Claude continue to evolve. The report’s authors—led by Ruth Appel, Peter McCrory, and Alex Tamkin—stress that while technical progress is inevitable, societal outcomes depend on deliberate choices. Policymakers could promote equitable access through infrastructure investments, subsidies for data tools in developing regions, or education programs blending AI literacy with human skills.
Business leaders, meanwhile, stand to gain from addressing contextual barriers early. By modernizing data pipelines and fostering human-AI collaboration, companies can extend AI beyond coding silos into diverse operations, enhancing competitiveness. The report’s findings on weak price sensitivity suggest that as capabilities advance, adoption will likely accelerate, but targeted interventions are needed to ensure inclusivity.
Looking ahead, Anthropic plans ongoing monitoring of these patterns, providing empirical anchors for navigating AI’s economic ripple effects. As the third installment of the Economic Index, this edition expands the framework with API insights and global granularity, underscoring the technology’s dual potential: to amplify prosperity or deepen disparities. In the concluding remarks, the authors caution that “the economic effects of transformative AI will be shaped as much by technical capabilities as by the policy choices societies make.”
History demonstrates that adoption trajectories are malleable—evolving with maturity, innovations, and intentional deployment. Today’s concentrated patterns may broaden, capturing AI’s full productivity potential across sectors and borders. Yet, proactive steps now, from public advocacy to corporate strategy, will define whether AI fosters convergence or divergence in the global economy. This report not only illuminates current trends but also equips stakeholders with data-driven tools to influence AI’s trajectory.
As adoption intensifies, the interplay of geography, enterprise needs, and usage modes will be critical in harnessing AI for equitable growth.
