Quick Answer: India's government deployed 38,000+ GPUs at ₹65/hour—roughly 40x cheaper than commercial cloud rates. Eligible Indian startups, researchers, and MSMEs can register via the IndiaAI compute portal. The catch? It requires project approval, and demand will spike before the February 2026 AI Impact Summit.
The government just gave Indian startups what costs ₹50 lakh a month on commercial clouds—for practically nothing.
That's not an exaggeration. If you're training an AI model in India and you're still paying AWS or Azure rates, you're either uninformed or allergic to free money. The IndiaAI Mission has quietly built one of the world's largest public-sector GPU clusters, and as of January 2026, it's open for business.
Speaking at the AI Convergence Summit 2026 in Lucknow today, Prasad Menon, CEO of CIBA Centre for Incubation and Business Acceleration, put it bluntly: startups no longer need to build expensive infrastructure from scratch. Incubators can now provide access to world-class computing resources—something unthinkable just a few years ago.
The numbers tell the story.
The 38,000 GPU Reality Check
India's original IndiaAI Mission target was 10,000 GPUs. They've deployed over 38,000—more than triple that goal, according to official government statements. This includes high-end NVIDIA H100 and H200 GPUs, plus Google Trillium TPUs added in the third procurement tender.
The subsidized rate? ₹65 per GPU hour.
Let's put that in perspective. According to AceCloud's August 2025 pricing comparison, AWS charges roughly ₹330+ per hour for H100 access. Azure sits even higher at approximately ₹590 per hour for comparable compute. That makes IndiaAI's offering somewhere between 5x and 9x cheaper than commercial alternatives—and that's before accounting for enterprise pricing tiers and hidden data egress fees that can inflate cloud bills by 20-40%.
IT Minister Ashwini Vaishnaw called it the cheapest compute facility in the world when launching the IndiaAI Compute Portal in March 2025. For once, the ministerial hyperbole might actually be accurate.
Who Can Actually Access This?
Here's where the government bureaucracy kicks in—but it's surprisingly reasonable.
The IndiaAI compute portal accepts registrations from researchers and academic faculty, IndiaAI fellowship awardees, students at recognized institutions, DPIIT-registered startups, MSMEs, and government entities.
You'll need to register through Meri Pehchaan using DigiLocker, Parichay, or ePramaan credentials. After that, you submit a project proposal outlining your technical approach, intended impact, and estimated compute requirements.
The subsidy can cover up to 40% of compute costs for approved projects. According to a February 2024 MeitY memorandum, projects of national importance get priority access to maximum subsidies.
The process isn't instant. Requests go through a Project Management Evaluation Committee. But compared to negotiating enterprise cloud contracts? This is refreshingly straightforward.
The Seven Pillars Behind the Push
The ₹10,300 crore IndiaAI Mission isn't just about throwing GPUs at the problem. It's structured around seven pillars that reveal the government's actual strategy.
Compute Infrastructure gets the lion's share—₹4,563 crore earmarked specifically for GPU capacity expansion over five years. Seven industry partners including Yotta Data Services, E2E Networks, and Tata Communications provide the actual infrastructure.
Foundation Models is where things get interesting. India wants indigenous large language models trained on Indian languages and cultural context. Twelve startups have been selected across two phases—Sarvam AI, Gnani AI, Soket AI, and Gan AI among them—to build these homegrown models.
FutureSkills supports 13,500 scholars with AI training, including 500 PhD fellows. AI and Data Labs are being rolled out in Tier 2 and Tier 3 cities through NIELIT partnerships—31 labs operational as of recent government updates.
Startup Financing includes the IndiaAI Startups Global Acceleration Programme, running in partnership with Station F and HEC Paris, pushing 10 Indian AI startups into European markets.
Safe and Trusted AI covers the responsible deployment angle—machine unlearning, bias mitigation, privacy-preserving ML. Thirteen projects selected so far through expressions of interest.
The remaining pillars cover application development (30 India-specific AI applications approved in sectors like healthcare, agriculture, and cybersecurity) and the AIKosh national dataset platform hosting over 3,000 datasets.
Why This Matters Beyond the Summit
You might be tempted to dismiss this as pre-summit optics. Fair enough—governments love announcing things before big events.
But the structural shift here is real.
Stanford University's 2025 Global AI Vibrancy Tool ranks India third globally in AI competitiveness, behind only the United States and China. India scored 21.59, ahead of South Korea at 17.24 and the UK at 16.64. That's a jump from seventh place in 2023—a four-position climb in a single year.
The government's framing has also shifted. Nivedan Rathi, Founder of Future & AI, captured it at the AI Convergence Summit: for decades, India has largely been a consumer in the global economy, importing more than it exports. The IndiaAI Mission reflects a larger vision to transform the country from consumer to producer of knowledge and innovation.
Whether you buy the rhetoric or not, the compute is real. The subsidies are real. And the 89% of new Indian startups that integrated AI into their products last year, according to government statistics, suggest the ecosystem is already moving.
The India AI Impact Summit: What's Actually Happening
The main event runs February 19-20, 2026 at Bharat Mandapam in New Delhi. MeitY is positioning it as a follow-up to the Bletchley Park AI Safety Summit (2023), Seoul AI Summit (2024), and Paris AI Action Summit (2025)—but with a distinctly different focus.
As professional services firm Crowell noted, this summit shifts the conversation from AI safety and governance to implementation and measurable outcomes, particularly for the Global South.
Pre-summit events are already underway. The AI Convergence Summit that wrapped up today in Lucknow brought together 70+ AI experts from five countries, alongside India's first AI Healthcare Hackathon with 1,500 teams and 5,000 participants.
Three flagship global challenges—YUVAi for youth, AI by HER for women-led solutions, and AI for ALL—have attracted over 15,000 registrations from 135 countries with roughly 4,700 submissions. Awards total up to ₹6 crore.
The Summit will feature 300+ exhibitors from over 30 countries, research symposiums, an AI Expo focused on responsible intelligence, and the AI Pitch Fest (UDAAN) for startups with a focus on women leaders and differently-abled founders.
How to Actually Get Started
Enough background. Here's what you do if you want GPU access.
Step 1: Go to the IndiaAI Compute Portal and register using your DigiLocker or ePramaan credentials.
Step 2: Complete your profile with required documents. Startups need DPIIT recognition certificates. Researchers need institutional affiliations. Students need enrollment verification.
Step 3: Submit a project proposal with technical details, expected impact, and a draft Bill of Materials estimating your compute requirements and subsidy request.
Step 4: Wait for PMEC approval. Timelines vary, but approved projects get 30 calendar days to begin using services before the allocation expires.
Step 5: Access compute through empaneled service providers—Yotta, E2E Networks, Tata Communications, and others on the approved vendor list.
The registration for the AI Impact Summit itself is separate—available at impact.indiaai.gov.in. If you're building anything AI-related in India, showing up in February makes strategic sense.
The Catch (Because There's Always One)
Let's be honest about the limitations.
First, approval isn't guaranteed. Not every project gets the full 40% subsidy. Projects of genuine national importance get priority; your side project chatbot might not qualify.
Second, demand will spike. With the Summit weeks away and GPU access becoming widely known, expect competition for resources. Early movers have an advantage.
Third, this isn't unlimited compute. The 38,000 GPU number is total deployed capacity, not what's available at any given moment. High-end H100s will be in demand.
Fourth, the voucher-based model—designed to avoid market distortion—means you're still paying something, even with subsidies. The ₹65/hour rate is subsidized, not free.
But here's the thing: even at full unsubsidized rates, IndiaAI's pricing beats commercial clouds by a factor of 2-3x. The subsidy just makes an already compelling offer absurd.
What This Means for Indian AI
The cynical take is that this is infrastructure theater—announcements designed to generate headlines before a global summit. There's probably some truth to that.
But infrastructure theater with 38,000 actual GPUs is different from infrastructure theater with PowerPoint slides. The compute exists. The subsidies are funded. The registration portal works.
If AI development requires three things—talent, data, and compute—India has spent years building the first, is actively constructing data platforms like AIKosh for the second, and has now addressed the third more aggressively than perhaps any government outside China.
The country hosting a global AI summit in February isn't a coincidence. It's a statement.
Whether Indian startups capitalize on this moment will determine if the government's "global AI producer" ambition becomes reality—or just another announcement that faded after the summit ended.
The GPUs are waiting. What's your excuse?