The key to AI adoption is not just model supply, but how each country uses AI

Overview

The report discusses differences in AI adoption across over 60 countries, where the core question shifts from model and infrastructure supply to end-user demand and usage patterns. It argues that judging the gap between advanced and emerging economies by simple per-capita AI usage is misleading because emerging markets have much larger populations and higher shares of labor in agriculture and low-end manufacturing, which naturally depress per-capita metrics. When focusing on absolute usage in the information economy and AI task types, emerging markets are not simply lagging; some countries show higher marginal benefits in usage intensity, task complexity, and time savings.

Core views

The report's key views are: first, AI global penetration is rising quickly, but may encounter structural ceilings around 55%-60% over the long term, with some large emerging economies potentially peaking earlier due to occupational structure, local language, legal, and cultural differences. Second, AI use in emerging markets is more skewed toward software, writing, and education, while developed economies have a wider use mix, with higher shares in sales, finance, healthcare, law, and other professional services. Third, as AI adoption matures, per-task time savings may show diminishing returns because a larger share of AI tasks remains within human-capability bounds and human-in-the-loop oversight lengthens workflows. Fourth, lower- and middle-income economies are using AI more for automation, which may affect local routine and process jobs, especially in IT/BPO and process service roles that rely on labor arbitrage.

Analysis framework

The report uses third-party AI usage data for cross-country comparison, mainly referencing Claude usage data from the Anthropic Economic Index and combining it with app usage statistics from sources such as Sensor Tower. The analysis dimensions include AI penetration, per-capita use, absolute usage duration, average session length, usage structure by occupation/sector, distribution across work/learning/personal settings, AI's relative task-performance advantage over humans, time saved per task, and the distinction between automation and augmentation use.

Methodology notes

• Macro thematic researchAI adoption and usage structure comparison

  Compare AI adoption across three levels—per-capita usage, absolute usage, and usage scenarios—rather than relying on a single penetration metric.

  The report argues that emerging-market population and occupational structures distort per-capita metrics, so it is necessary to combine absolute usage in the information economy with specific task types to assess AI impact.

• Productivity analysisSpeed gains and quality enhancement framework

  Differentiate whether AI is used to save time, perform tasks beyond human capability, or improve the quality of existing tasks.

  Emerging markets show more automation and speed benefits, while developed economies show more quality enhancement and broader scenario application.

• Industry impact analysisAutomation risk and sovereign LLM demand

  Infer potential winners and losers in IT services, media, cloud services, data centers, utilities, and local-language models based on usage structure.

  If AI adoption continues to expand, lower-end process jobs may be replaced by local automation, while local language, legal, and cultural localization will strengthen demand for sovereign LLMs.

Asset mapping & comparison

Structured mapping from thesis to named assets (strengths, weaknesses, peers, risks).

• IT services sector

  Facing AI automation pressure with potential business model reshaping

  Strengths

  Existing software and BPO talent pools can adopt AI relatively quickly, and in some countries AI use for software/coding is high.

  Weaknesses

  Labor-intensive delivery models may be weakened by AI automation, with lower-end process jobs at risk of displacement.

  Comparison

  Compared with developed economies, emerging markets are more concentrated in AI use for software, writing, education, and automation.

  Risks

  If task complexity does not upgrade, labor arbitrage advantages may erode and unemployment pressure may rise.

• Sovereign LLMs and local-language models

  Potential beneficiary direction

  Strengths

  Can adapt to local language, legal rules, customs, and occupational structure, helping to break through AI adoption ceilings.

  Weaknesses

  Require capital, data, compute, and policy support, and face competition from global general-purpose models.

  Comparison

  Large emerging economies may need localized AI capabilities more than smaller high-income economies.

  Risks

  If local-model quality is insufficient or costs are too high, uptake may fall short of expectations.

• Media, content, and production platforms

  May become the next industries to be restructured by AI

  Strengths

  AI use in media is already substantial, with room to improve content generation, editing, and production efficiency.

  Weaknesses

  Traditional broadcasters and film/TV studios may face pressure from reshaped cost structures and creative workflows.

  Comparison

  The report views streaming platforms and content/production firms as more likely to benefit than traditional media.

  Risks

  Copyright, content quality, regulation, and employment impacts may slow implementation.

• Cloud service providers

  Long-term beneficiaries of expanding AI demand

  Strengths

  Continued AI usage growth increases demand for compute, model hosting, and enterprise workflow deployment.

  Weaknesses

  Short-term capex is high, and some LLM services remain unprofitable.

  Comparison

  Compared with standalone data-center buildout, long-term value is more likely to stay at cloud and platform layers.

  Risks

  Price competition, per-token cost compression, and self-hosted/local sovereign model alternatives may squeeze margins.

• Utilities

  Indirect beneficiaries of AI infrastructure expansion

  Strengths

  AI training and inference increase electricity demand from data centers.

  Weaknesses

  Benefits depend on data-center buildout pace, transmission/distribution capability, and tariff regimes.

  Comparison

  The report expects data-center construction to possibly peak in the mid-cycle, while long-duration electricity demand benefits from AI may remain.

  Risks

  Construction cycles, regulatory approvals, and energy-structure constraints may affect realization.

Key data

• Global AI usage rate among working-age populationaround 18%As of March 2026, up from about 15% nine months earlier.
• Long-term AI adoption peak assumptionaround 55%-60%The report believes global AI adoption may face a structural long-term ceiling, with some large emerging economies potentially reaching the peak earlier.
• Average time saved per task in emerging markets4.6 hoursHigher than about 3.8 hours in developed markets.
• Share of tasks AI can handle beyond users' own abilitiesabout 16% in emerging markets, about 12% in high-income economiesThis indicates that although penetration is lower, some emerging markets allocate a higher share of more complex tasks to AI.
• Share of tasks that humans can complete independently but are assigned to AIabout 86%about 88.1% in developed economies and about 83.8% in emerging economies.
• Total India GPT+Claude usage durationmore than 500,000 hoursFrom Jan to Jun 2026, significantly higher than about 215,000 hours in the United States and slightly below about 30,000 hours on average globally.
• India average LLM session durationabout 3 minutes 57 secondsAbove the global average of about 3 minutes 47 seconds, ranking 11th among 76 economies.
• Share of India AI use related to workabout 44%Near the global average of about 42%; course-related use is about 20%, also near the global average.
• AI concentration areas in emerging marketsComputing 26.1%, arts/writing 14.2%, education 13.7%Corresponding shares in developed economies are about 23.5%, 13.3%, and 13.0%, respectively.
• AI usage advantage areas in advanced economiessales 9.6%, business and finance 5.7%Above emerging economies' 7.4% in sales and about 5.0% in business and finance.

Impact & implications

For markets and industries, the report implies that the AI value pool may be shifting from pure model and infrastructure supply to adoption depth, industry scenarios, workflow transformation, and localization capability. The IT services sector is likely to become less labor-intensive in the future, with competitors coming not only from traditional outsourcing companies but also from domestic software engineers and consulting firms that adopt technology/AI earlier. Media may become the next cluster of industries to be restructured by AI; streaming platforms, content platforms, and production-efficient companies may benefit, while traditional broadcasters and film/TV studios face pressure. As AI demand continues to have room to expand, LLM subscription prices may rise; mid-cycle data-center buildout may peak, while long-term gains are likely to accrue more to cloud providers and utilities.

Risks

• Per-capita AI usage can mislead cross-country comparisons; without accounting for population and occupational structures, emerging markets may appear more behind than they are.
• Claude usage data can proxy usage patterns, but in lower-income economies, free or low-cost alternative models may cause Anthropic data to underestimate overall AI adoption.
• Wider AI use may bring diminishing marginal productivity, and especially under human-in-the-loop review processes, time savings may not be significant.
• Long-term reliance on superficial manual reviews may erode professionals' ability to detect subtle but high-impact AI errors.
• As emerging markets use AI more for automation, local routine process jobs and labor arbitrage-based models may be hit.
• There is high uncertainty around AI adoption ceilings, subscription pricing, sovereign LLM rollout speed, and the peak timing of data-center demand.

What to watch

• Whether global AI penetration among the working-age population can continue rising from 18% toward the long-term 55%-60% ceiling.
• How AI usage intensity evolves in emerging markets such as India, Vietnam, and the Philippines across software, education, writing, and work settings.
• How the share shifts of AI tasks moving from assisting within human capability toward handling more complex tasks.
• Whether automation use continues to exceed augmentation use, especially in IT/BPO, media, and process-service sectors.
• The pace of sovereign LLM progress in adapting to local language, legal frameworks, and occupational norms.
• LLM subscription pricing, per-token costs, cloud provider margins, and data-center buildout cycles.
• Whether AI adoption in sectors like media, healthcare, and finance is moving from pilot use into core workflows.