What Happened
- IndiaAI Mission CEO Abhishek Singh announced that India's GPU (Graphics Processing Unit) capacity is expected to triple from the current approximately 38,000 to 1,00,000 GPUs by the end of 2026.
- Of the current 38,000 GPUs, approximately 24,000-25,000 have been installed and deployed, with another 20,000-25,000 units expected in the next three months.
- The expansion is part of the Rs 10,372 crore IndiaAI Mission, which subsidises GPU usage for domestic startups, researchers, and academic institutions at rates significantly below commercial cloud providers.
- The goal is to ensure at least 100 million Indians benefit from AI-powered services, particularly in healthcare, agriculture, and education.
Static Topic Bridges
IndiaAI Mission — Structure and Components
The IndiaAI Mission, approved by the Union Cabinet in March 2024, is India's flagship national programme for building sovereign AI capacity. With an outlay of Rs 10,372 crore over five years, it represents the government's most comprehensive intervention in the AI ecosystem.
- Seven pillars: (1) IndiaAI Compute Capacity, (2) IndiaAI Innovation Centre, (3) IndiaAI Datasets Platform, (4) IndiaAI Application Development Initiative, (5) IndiaAI FutureSkills, (6) IndiaAI Startup Financing, (7) Safe & Trusted AI
- Compute capacity receives the largest allocation: Rs 4,563.36 crore (44% of total budget) over five years
- GPU procurement through competitive bidding: 10 companies qualified in initial rounds; the third tender round added approximately 3,850 compute units
- Subsidised rates: GPUs available at Rs 65 per GPU hour — approximately one-third of global average commercial cloud cost
- Subsidy structure: 100% compute subsidy for foundational AI model development; up to 40% for application building by startups and MSMEs
- Operational target: 100 million Indians using AI-enabled government services
Connection to this news: The tripling from 38,000 to 1,00,000 GPUs represents the rapid scaling phase of the Mission's compute pillar, moving from initial procurement to building what would be one of the largest publicly accessible AI compute pools in the developing world.
GPU Computing — Why It Matters for AI
Graphics Processing Units (GPUs) are specialised processors originally designed for rendering graphics but now critical for training and running AI models. Their ability to perform thousands of parallel mathematical operations simultaneously makes them far more efficient than CPUs for the matrix computations that underpin modern AI.
- A single modern AI GPU (e.g., Nvidia H100) can perform approximately 4 petaflops of AI computation; training large language models requires thousands of GPUs working together
- Nvidia dominates the AI GPU market with approximately 80-90% market share; its H100 and H200 GPUs are the industry standard for AI training
- Alternative AI accelerators include Google TPUs (Tensor Processing Units), AMD Instinct MI300X, and custom ASICs by companies like Cerebras
- GPU scarcity has become a geopolitical issue: the US has imposed export controls on advanced AI chips to China (October 2022, updated 2023) under the Bureau of Industry and Security (BIS) framework
- India has not been subject to these restrictions but is building sovereign compute capacity to avoid dependency on any single supplier or geopolitical bloc
Connection to this news: India's push to reach 1,00,000 GPUs is fundamentally about AI sovereignty — ensuring that Indian researchers, startups, and government agencies can access the computational resources needed for AI development without being entirely dependent on foreign cloud providers.
AI Sovereignty and Compute Infrastructure — Global Context
AI sovereignty refers to a nation's ability to develop, deploy, and control AI technologies using its own infrastructure, data, and talent. As AI becomes a strategic technology, countries are investing heavily in domestic compute infrastructure to reduce dependence on foreign cloud providers and maintain control over sensitive data and models.
- France launched a EUR 2.5 billion AI strategy including a sovereign cloud; the UK invested GBP 1.3 billion in AI computing (including the Isambard-AI supercomputer at the University of Bristol)
- Saudi Arabia's Project Transcendence aims to invest $100 billion in AI infrastructure
- The EU Chips Act (2023) allocates EUR 43 billion for semiconductor self-sufficiency, complementing the European AI Act's governance framework
- At Davos 2026, Minister Vaishnaw stated that India aims to run most AI workloads on sovereign models within a year
- India's approach emphasises affordable compute access (subsidised rates) rather than proprietary national champions, making AI infrastructure a public good
Connection to this news: The 1,00,000 GPU target is central to India's AI sovereignty strategy. By building publicly accessible compute at subsidised rates, India aims to democratise AI development — enabling local startups and researchers to compete globally without depending on expensive commercial cloud infrastructure.
IndiaAI Compute Procurement — Public-Private Partnership Model
The IndiaAI Mission uses a unique procurement model where private companies build and operate GPU data centres while the government provides demand aggregation, subsidies, and guaranteed utilisation. This PPP model reduces the financial risk for private providers while ensuring affordable access for end users.
- Procurement via competitive bidding: GPU service providers bid on cost per GPU hour; lowest bids averaged 42% below market price
- Winning bidders included major Indian and global companies (Google, Infosys, TCS, Yotta, among others)
- Cost to users: Rs 115.85/GPU hour for low-end GPUs, Rs 150/GPU hour for high-end GPUs at bid rates; government subsidies bring costs below Rs 100/GPU hour for eligible users
- Current availability: 38,000+ GPUs at Rs 65/hour after subsidies
- The mission plans to set up 600 data labs across the country to extend access beyond metropolitan areas
Connection to this news: The tripling of GPU capacity relies on this PPP procurement model — the government does not build data centres itself but incentivises private operators to deploy GPUs that are then made available to Indian users at subsidised rates.
Key Facts & Data
- Current GPU capacity: ~38,000 (24,000-25,000 installed and deployed)
- Target: 1,00,000 GPUs by end of 2026 (approximately tripling current capacity)
- IndiaAI Mission budget: Rs 10,372 crore over 5 years (approved March 2024)
- Compute pillar allocation: Rs 4,563.36 crore (44% of total budget)
- Subsidised GPU rate: Rs 65/hour (approximately one-third of global average)
- Subsidy: 100% for foundational model development; up to 40% for applications
- 600 data labs planned across India
- IndiaAI Mission CEO: Abhishek Singh
- User target: 100 million Indians using AI-powered services