India Could Become a Major Paid AI Market, but Risks of Value Leakage and Excess Computing Capacity Are Rising in Tandem

Overview

The report compares usage data from OpenAI and Anthropic to examine how India and other emerging markets use AI, whether users will shift to paid subscriptions, and what this transition means for employment, cross-border fees, IT services, and data center investment. Its core conclusion is that emerging markets are not laggards in AI adoption; instead, they are moving more rapidly into productive and automated use cases. However, whether India becomes an economic beneficiary depends on its ability to build domestic models and applications rather than merely purchasing overseas services.

Core views

AI adoption in India is substantial, but per-capita metrics can easily understate its impact. India surpassed 100 million weekly active ChatGPT users in February, making it the second-largest market after the United States, yet its per-capita usage ranks only 109th among 150 economies. Anthropic data show the same pattern: India contributes 7.1% of absolute usage, ranking second, but ranks only 103rd among 121 economies in per-capita usage. The report attributes this disparity to the fact that much of India's workforce remains in industries with low AI exposure, rather than to a lack of depth among those already using AI. Even with a relatively concentrated user base, Indian paid users use ChatGPT's data analysis tools at four times the global average intensity and coding tools at three times the global average intensity. AI use in emerging markets is more oriented toward work and automation, whereas developed economies use it more for information acquisition and assistance with expression. In India, 35% of ChatGPT use is work-related, above the global average of 31%; work-related use accounts for 32.4% in emerging economies, also above the 29.3% recorded in developed economies. India remains among the leaders in technical assistance across the 28 major economies tracked by the report. The report therefore believes users in emerging markets are more likely to have AI directly complete technical, writing, or similar work tasks, while users in developed economies more often use AI to gather information, express themselves, and obtain practical advice. This also means the employment impact of AI in emerging markets should not be dismissed because current adoption is low-cost and localized; initial enterprise use cases may fund subscription fees by reducing staffing needs. Usage categories are changing rapidly and show that different models are gradually developing specialized roles. The share of “technical assistance” in Indian ChatGPT queries has fallen from approximately 20% in 2024 to 6.3% currently, although it remains among the highest globally. Over the same period, multimedia rose from 1.4% to 9.8%, self-expression from 2% to 6.3%, practical guidance from 26% to 33%, and information search from 15% to 17%. Globally, technical use of ChatGPT also fell from approximately 13% in 2024 to 4% currently, while computer and mathematics use on Anthropic exceeds 25% in emerging economies. Based on this, the report concludes that technical and coding users are migrating to more specialized tools: Anthropic is becoming the preferred choice for coding tasks, while OpenAI continues to lead in information search, personal guidance, and multimedia use cases. The age and use-case mix gives emerging markets relatively high long-term subscription value. The share of global ChatGPT users under age 35 declined from nearly 70% in 2024 to 66%, indicating that older users are joining. In developed economies, approximately 40% of users are over 35 and 18% are over 45, compared with 28% and 11%, respectively, in emerging economies. India's user base is younger: 78% of users are under 35 and 92% are under 45, while approximately 30% of its population is over 45 and it has more than 1 billion internet users. The report defines technical assistance, writing, and practical guidance as “productive use cases,” further categorizes writing and technical assistance as knowledge- or work-related uses, and uses the share of young users as a proxy for sustainable usage duration and productive use as a proxy for willingness to pay. The resulting matrix shows that emerging economies have both longer potential user lifecycles and a greater propensity to pay. The primary constraint is not the value of use, but affordability resulting from income disparities. If subscription prices reflect local income levels, more paid users could emerge in developing economies. AI subscriptions differ fundamentally from the business models of previous free internet services. India was able to accumulate users rapidly during the eras of social media, video, payments, and e-commerce because many services were free, with platforms monetizing through advertising, transactions, and ecosystem expansion. Software also often had free alternatives, while ongoing costs were low once spread across a large user base. Generative AI, by contrast, entails high per-user computing and energy costs; the more it is used, the higher the operating costs. Consequently, ChatGPT, Gemini, Claude, Microsoft Copilot, and AI-native enterprise applications have emphasized paid subscriptions from the outset. Account sharing is also constrained by token limits, privacy, and account permissions. The report therefore believes paid AI could ultimately become widespread in emerging markets, but growth in consumer token usage will be more gradual, while annual subscriptions will crowd out other discretionary spending. The report presents scenario estimates for India's LLM subscription spending. Assuming retail user conversion rates of 0.75%-1.25% for the three major models—Gemini, ChatGPT, and Claude—and enterprise user conversion rates of approximately 1.4% for all three, current adoption levels imply annual LLM subscription spending of nearly $950 million, representing a realistic annual outflow of approximately $1 billion. Although ChatGPT clearly leads with approximately 330 million monthly active retail users, its revenue share will not be proportionately dominant, and Claude is expected to trail only slightly. As enterprise penetration rises and subscription plans move upmarket, the report believes annual outflows could soon reach $4 billion-$5 billion. In ten years, annual outflows from domestic LLM usage fees could reach $20 billion-$30 billion, excluding expenditures offset by export revenue; the application layer could charge tens of billions of dollars more. This would make AI fees equivalent to approximately one-third of India's current oil import bill. If India lacks major domestic foundation models and widely used application software, most of these funds will flow to U.S. technology companies. A substantial reduction in model costs or the development of domestic Indian applications could change this outcome. This leaves India at a critical fork in the road: whether being a major user also makes it a major capturer of value. The report believes that becoming one of the world's largest AI users does not automatically mean India will become one of its largest economic beneficiaries. If India can only purchase overseas models and applications, its young and highly productive user base will translate into persistent cross-border fees. If it can build AI-native companies, domestic foundation models, and application-layer businesses, it could retain more value. The report also believes that extensive use of AI for software tasks in markets such as India and Vietnam will drive a reallocation of IT service capabilities within the countries supplying the relevant labor. AI alternatives and agentic companies could grow locally, but a more evenly distributed capability base may also disrupt some large IT service organizations. India's current AI investment is concentrated mainly in infrastructure and manufacturing rather than frontier large models or mature application products. Commitments associated with the 2026 AI Impact Summit in New Delhi could reach $250 billion, according to government officials cited in the report. Reliance has committed $110 billion over seven years to build sovereign AI infrastructure, while Microsoft plans to invest $50 billion through India by 2030 to extend AI to low-income countries. Data centers and computing power are the largest investment areas: current capacity is approximately 1,500 MW, and 30 projects totaling 3.5 GW of new capacity were announced between March 2025 and April 2026. Some announcement-based estimates project capacity reaching 12 GW by 2030. Semiconductor investment is more heavily focused on outsourced packaging and testing than on wafer fabrication. The IndiaAI Mission seeks to close gaps in sovereign computing power and foundation models. Its shared facilities contain more than 45,000 GPUs, fund 15 domestic large and small language models, and support 20 AI solutions across 12 industries, including agriculture and healthcare. The report is clearly cautious about the 12 GW data center target. Its high-level estimate indicates that a 1 GW data center operating at 70% average daily utilization could process approximately 3 billion inference queries per day, equivalent to serving roughly 55 million-60 million daily active users. In theory, 12 GW could serve more than 700 million daily active users. Even before considering electricity, water resources, contractor capabilities, and GPU availability, demand itself may be insufficient. Therefore, if all announced projects materialize, substantial oversupply could emerge. Unless Indian data centers can serve cross-border AI inference demand, the wave of announcements may gradually turn into delays or cancellations; however, cross-border inference is itself constrained by latency sensitivity and data localization requirements. The report plans to further refine this capacity estimate in subsequent work.

Analysis framework

The report first uses OpenAI and Anthropic data to compare absolute usage, per-capita usage, task categories, and age structures across India, emerging economies, and developed economies. It then treats the share of young users as a proxy for potential years of use and productive uses such as technical assistance, writing, and practical guidance as proxies for willingness to pay. Next, the report constructs subscription-spending scenarios using the monthly active users and retail and enterprise paid-conversion rates of the three major LLMs, extending the results to cross-border fees and domestic value capture. Finally, it tests whether India's planned data center capacity matches potential demand by using the number of queries that a unit of computing capacity can process.

Methodology notes

• (Method Outside the Taxonomy)

  Cross-Platform Usage Data Comparison

  The report cross-compares absolute usage, per-capita rankings, task mix, and age structures from OpenAI and Anthropic to test whether the AI usage characteristics of India and emerging markets recur across different platforms.

• (Method Outside the Taxonomy)

  User Lifetime Value and Willingness-to-Pay Matrix

  The report uses the share of young users as a proxy for the duration of sustainable future usage and the share of productive tasks as a proxy for willingness to pay, thereby comparing potential subscription lifetime value across economies. Affordability is treated separately as a limiting factor.

• Industry/Sector Analysis FrameworkVolume-price decomposition

  LLM Subscription Spending Scenario Analysis

  The report estimates current, near-term, and ten-year annual spending and cross-border capital outflows based on the active-user scale, retail and enterprise paid-conversion rates, and subscription pricing of major LLMs.

• Industry/Sector Analysis FrameworkUpstream-Midstream-Downstream Industry Chain Transmission

  AI Value Chain and Capital Flow Analysis

  The report distinguishes among data centers and computing power, semiconductors, foundation models, and the application layer, analyzing where Indian investment is concentrated and whether subscription revenue ultimately accrues to domestic companies or overseas technology firms.

• Industry/Sector Analysis FrameworkSupply-demand framework

  Data Center Capacity Supply-Demand Test

  The report converts the daily query-processing capacity of a 1 GW data center at 70% utilization into the number of users it can serve, then compares this with the planned 12 GW capacity for 2030 to assess whether oversupply could emerge.

Asset mapping & comparison

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

• India's AI Economy

  India has a globally leading absolute user base, a young user demographic, and a high share of productive use cases, but may primarily bear subscription fees for overseas LLMs and applications.

  Strengths

  More than 100 million weekly active ChatGPT users, high levels of work, technical, and coding use, and a young user base supporting a longer potential lifecycle.

  Weaknesses

  Per-capita usage remains low, affordability is constrained by income levels, and domestic frontier models and mature application-layer offerings are relatively limited.

  Comparison

  Absolute usage is second only to the United States, but India's per-capita ChatGPT and Anthropic usage rankings are only 109th/150 and 103rd/121, respectively.

  Risks

  Job displacement, subscription spending crowding out discretionary consumption, growing cross-border fees, and insufficient domestic value capture.

• Global LLM Subscription Platforms

  India and other emerging markets could become sources of long-term paying users, directing most subscription revenue to overseas model providers.

  Strengths

  Emerging-market users are younger and have higher levels of productive and work-related usage, supporting a stronger propensity to pay.

  Weaknesses

  High per-user variable costs make a free model difficult to sustain over the long term, while local income levels constrain affordability.

  Comparison

  Anthropic is more oriented toward coding and technical tasks, while OpenAI is stronger in information search, personal guidance, and multimedia use cases.

  Risks

  Slower adoption after subsidies are withdrawn, pricing mismatches with local incomes, and user migration between models.

• India's IT Services Industry

  High use of AI for software tasks in markets such as India and Vietnam could drive a local reallocation of service capabilities and foster domestic AI alternatives and agentic companies.

  Strengths

  A globally oriented software talent pool and relatively high levels of technical AI use.

  Weaknesses

  The diffusion of capabilities enabled by AI may weaken the concentration advantages of some large organizations.

  Comparison

  Emerging economies are more inclined than developed economies to use AI to directly complete work and automation tasks.

  Risks

  Reduced staffing needs and disruption to large service providers.

• India's Data Centers and Computing Infrastructure

  Investment commitments and project announcements are driving rapid capacity expansion, but the report believes the 12 GW target for 2030 could materially exceed domestic demand.

  Strengths

  Global cloud providers, Indian conglomerates, and the government are all committing resources, resulting in large-scale current plans.

  Weaknesses

  Demand, power supply, water resources, contractor capabilities, and GPU availability may limit project execution.

  Comparison

  Current capacity is approximately 1.5 GW, while announcement-based forecasts reach 12 GW by 2030.

  Risks

  Overcapacity, project delays, or cancellations; cross-border inference is also constrained by latency and data localization.

Key data

• Weekly Active ChatGPT Users in IndiaMore than 100 millionSurpassed this threshold in February, with absolute scale second only to the United States
• India's Per-Capita ChatGPT Usage Ranking109th/150Large absolute scale, but a relatively low per-capita ranking
• India's Share of Anthropic Usage7.1%Absolute usage is second only to the United States
• India's Per-Capita Anthropic Usage Ranking103rd/121Consistent with ChatGPT's relatively low per-capita usage pattern
• Data Analysis Tool Usage Intensity Among Paid Users in India4 times the global averageIndicates substantial usage depth among the concentrated user group
• Coding Tool Usage Intensity Among Paid Users in India3 times the global averageReflects a relatively high level of technical use
• Share of Work-Related ChatGPT Use in India35%Above the global average of 31%
• Share of Work-Related Use in Emerging vs. Developed Economies32.4% vs 29.3%Emerging economies are more oriented toward work-related use cases
• Technical Assistance Use in IndiaDeclined from approximately 20% to 6.3%A significant decline since 2024, but still relatively high globally
• Global Technical Use of ChatGPTDeclined from approximately 13% to 4%Since 2024, indicating that technical users are migrating to other tools
• Multimedia Use in IndiaRose from 1.4% to 9.8%The usage mix is rapidly shifting toward content such as images and video
• Practical Guidance Use in IndiaRose from 26% to 33%One of the more rapidly growing use cases
• Share of Indian Users Under Age 3578%Above the global figure of 66%
• Share of Indian Users Under Age 4592%The user age distribution is notably young
• Estimated Current Annual LLM Subscription SpendingNearly $950 millionBased on retail conversion rates of 0.75%-1.25% and an enterprise conversion rate of approximately 1.4%, corresponding to roughly $1 billion in annual outflows
• Estimated Near-Term LLM Fee Outflows$4 billion-$5 billion/yearBased on higher enterprise penetration and adoption of premium subscription plans
• Estimated LLM Fee Outflows in Ten Years$20 billion-$30 billion/yearExcluding expenditures offset by export revenue; the application layer could charge tens of billions of dollars more
• India's Current Data Center CapacityApproximately 1,500 MWEquivalent to approximately 1.5 GW
• New Data Center Projects30 projects totaling 3.5 GWAnnounced between March 2025 and April 2026
• Estimated Data Center Capacity in 203012 GWBased on announced projects; the report considers the plan aggressive
• IndiaAI Mission Shared Computing CapacityMore than 45,000 GPUsFunding 15 domestic large and small language models and 20 AI solutions across 12 industries
• Inference Capacity of a 1 GW Data CenterApproximately 3 billion queries/dayEstimated at 70% utilization, sufficient to serve approximately 55 million-60 million daily active users
• Potential Service Capacity of 12 GW of Data CentersMore than 700 million daily active usersUsed to illustrate the potential demand shortfall if all planned capacity comes online

Impact & implications

The report believes emerging markets could become important AI subscription markets due to their young users and productive use cases, while enterprise adoption may also support spending by reducing staffing needs. However, if India cannot establish domestic capabilities in foundation models and the application layer, its substantial usage will manifest more as persistent fees flowing to overseas technology companies than as local value creation. Meanwhile, AI could reshape the supply structure of IT services in countries such as India and place pressure on some large institutions. Infrastructure investment also faces issues including insufficient demand, electricity and water constraints, GPU availability, latency, and data localization.

Risks

• If subscription prices do not adapt to income levels in emerging markets, user affordability may constrain paid conversion.
• Enterprises may finance AI use by reducing staffing needs, potentially amplifying the employment impact.
• If India lacks domestic foundation models and a widely used application layer, substantial LLM and application fees will flow overseas.
• Annual AI subscriptions may crowd out other discretionary consumer spending.
• The planned 12 GW of data center capacity by 2030 may exceed actual inference demand, leading to oversupply, delays, or project cancellations.
• Electricity, water resources, contractor capabilities, GPU availability, latency, and data localization may constrain data center construction and cross-border services.
• A more even distribution of AI capabilities may disrupt some large IT service providers.

What to watch

• Track the path of declining unit costs for LLMs and the withdrawal of subsidized pricing.
• Monitor whether subscription plans aligned with local income levels can improve paid-conversion rates in emerging markets.
• Watch enterprise-user penetration, token usage, and adoption of premium subscription plans in India.
• Monitor whether India can develop major domestic foundation models, AI-native companies, and widely used application-layer products.
• Track the actual commencement, utilization, delays, and cancellations of announced data center projects.
• Watch whether Indian data centers can serve cross-border inference demand and whether latency and data-localization constraints ease.
• Monitor the impact of AI on IT services employment in India and the competitive position of large service providers.