The Indian government's IndiaAI Mission has deployed over 38,000 GPUs — H100s, H200s, and Google Trillium TPUs — at subsidized rates of ₹65–92 per GPU-hour. Commercial equivalents on AWS Mumbai or Azure India run ₹300–600/hr. That gap is real and the programme is live. Here's what it actually means for Indian startups and researchers.
What the subsidy actually costs
The headline rate is ₹65/GPU-hour for standard allocation. Projects in areas of national importance — healthcare AI, agriculture, Indic language models — can apply for a further 40% reduction, bringing effective costs below ₹100/hr. For context: E2E Networks' on-demand H100 rate is ₹362/hr and NeevCloud's is ₹191/hr. The subsidized rate undercuts even the cheapest Indian commercial provider by roughly 30–50%.
Who qualifies
The IndiaAI compute programme is open to Indian startups registered under DPIIT, academic institutions, MSMEs, and individual researchers affiliated with recognized institutions. Foreign-incorporated entities, even with Indian offices, are not eligible. There is no revenue cap — early-stage founders with zero revenue can apply alongside series-B companies. The stated goal is to lower barriers for indigenous AI model development rather than to subsidize existing cloud users.
How the application process works
Access is through the IndiaAI Compute Portal at indiaai.gov.in. You register your organization, describe the workload (model type, dataset size, compute hours needed, intended outcome), and submit. MeitY's evaluation committee reviews applications on a rolling basis and approves compute quotas rather than open-ended access. Approved projects receive compute credits usable against any of the 14 empaneled service providers — the list includes E2E Networks, Cyfuture, Yotta, and others already listed on IndiaGPU. You are not locked into a single provider after approval.
The practical catch
This is not self-serve. You cannot sign up and start a training run the same day. The application review takes weeks, and quota allocation is tied to the specific workload you described. If your use case changes materially, you may need to re-apply. There are also per-project caps — the system is designed to spread access widely rather than giving one team unlimited H100 time. For teams with deadline-driven experiments, commercial providers with immediate on-demand access may still be necessary even if you eventually qualify for the subsidy.
Is it worth the effort
For a research project running 500 GPU-hours per month: at commercial rates that's ₹1.8–3L/month; at IndiaAI rates it's ₹32,500–46,000/month. Over six months, the savings exceed ₹10–15 lakh. For early-stage startups where compute cost is a genuine constraint on experimentation, the approval process is worth the effort. For teams that need immediate access or are already past their experimentation phase, commercial providers with instant provisioning are the more practical choice — and Indian providers like E2E Networks and NeevCloud now offer prices competitive enough that the gap matters less than it did in 2024.