Previous entries in this digest covered India's GPU story from the deal side -- pledges, vendor partnerships, headline stockpile counts. What's newly public is the pricing mechanism underneath those deals. According to the Press Information Bureau, subsidised GPU compute under the IndiaAI Mission is priced at Rs 65 per GPU-hour, funded by a government subsidy of roughly 40% that brings the effective charge down to somewhere between Rs 67 and Rs 116 an hour depending on GPU class -- with the government paying infrastructure providers directly to allocate that discounted compute to local model builders, researchers, and academic institutions rather than subsidising end users after the fact.
The budget behind that subsidy is a multi-year commitment, not a one-off: the total approved IndiaAI Mission outlay is Rs 10,371.9 crore, with Rs 1,000 crore specifically earmarked for FY27 in the most recent Union budget. That's a meaningfully different signal than a single splashy pledge -- it's a standing line item that infrastructure providers and startups can actually plan against.
On raw capacity: officials are targeting 100,000 GPUs by late 2026, contingent on budget approvals and vendor delivery timelines actually holding. February's India AI Impact Summit pledged 20,000 additional sovereign GPUs "within weeks," pushing total public capacity past 58,000 units at the time. Industry projections suggest private-sector deployments will roughly mirror the public number, which would put cumulative national GPU capacity somewhere past 200,000 units within two years if the trajectory holds.
The governance posture around all of this remains deliberately light relative to comparable efforts elsewhere: where the EU AI Act leads with risk classification and compliance obligations, India's approach leads with infrastructure and capital, organized around three stated principles -- inclusivity, responsibility, and sovereignty, with sovereignty specifically meaning the capacity to build and run AI models on domestic data and domestic compute rather than importing both.
India's AI compute story has moved from pledges to an actual priced subsidy mechanism (Rs 65/GPU-hour, ~40% subsidised, Rs 10,371.9 crore multi-year outlay) -- the number to watch isn't the headline GPU count, it's whether the 100,000-GPU-by-late-2026 target survives contact with actual vendor delivery schedules.