India focuses on GPU subsidies and ‘BharatGen’ to rewrite the AI growth playbook

By Kamal Das, Dean at the Wadhwani Center for Government Digital Transformation (WGDT)

One key takeaway from India Union Budget, is that the government is done treating Artificial Intelligence as a side project. Finance Minister Nirmala Sitharaman’s 2026-27 blueprint puts AI front and centre, not just as a shiny talking point tucked into the last ten minutes of a Budget speech, but as the actual scaffolding for the country’s next chapter of economic growth. For an ecosystem that has spent years begging for affordable compute and local-language models, this feels welcome.

Cheap GPUs, Finally

The headline number is hard to miss. The IndiaAI Mission is being scaled up to 38,000 GPUs available to startups at roughly ₹65 an hour. That is, by any measure, a dramatic move. Until now, a bootstrapped AI lab in Bengaluru or Hyderabad had to either burn venture money on cloud credits or make do with underpowered hardware. At only ₹65 an hour, the economics shift overnight. Not entirely, of course as training a serious model still isn’t cheap but it removes a barrier that has genuinely held people back.

“We are democratising the guts of AI. High-end compute should not be a luxury restricted to Big Tech.”— Nirmala Sitharaman, Finance Minister

There’s more on the infrastructure side. Data centres now have “Strategic Infrastructure Status,” which is bureaucratic jargon for access to long-term, low-interest loans for the kind of patient capital that only highways and power plants used to get. And foreign cloud providers building capacity here get a tax holiday till 2047. The message to hyperscalers is clear enough: park your servers in India, keep Indian data in India, and we’ll make it worth your while.

BharatGen and the Language Question

Then there’s BharatGen, and this is where it gets genuinely interesting. The government wants sovereign foundational models covering all 22 scheduled languages by June 2026. Twelve startups have been roped in to build Large Multimodal Models that actually understand Indian context. A farmer querying crop insurance in Marathi shouldn’t have to translate his question into English first, and frankly, the fact that he still does in 2026 is a bit of an embarrassment for the ecosystem.

Supporting all this is AIKosh, a national dataset platform that now sits on over 5,500 datasets. The data, officials say, is meant to ensure that Indian AI is “trained on Indian reality” not on Reddit posts and Wikipedia articles scraped from the Western internet. It’s a reasonable ambition. Whether 5,500 datasets are actually sufficient for 22 languages, each with their own regional dialects and cultural nuance, is a different conversation entirely.

Skilling Youth on AI

The skilling piece is less flashy but arguably just as important. A High-Powered Committee will study AI’s impact on services, the sector that employs the lion’s share of India’s white-collar workforce. Content Creator Labs in 15,000 schools and AI Centres of Excellence at premier institutions are meant to make sure the next generation doesn’t end up on the wrong side of the automation divide. It’s a sensible start, though the proof, as ever, will be in the implementation rather than the press release.

The Gaps That Worry People

For all the good news, a few things are conspicuously absent. Procurement, for starters. Founders will tell you that selling to the government is still complicated. A “GovTech Fast-Track” for AI pilots from startups could have been a genuinely useful reform.

Data sharing is another sore point. AIKosh is growing, sure, but the valuable datasets PII redated healthcare records, logistics flows, financial transactions remain locked behind corporate firewalls. There was talk of a “Data Dividend Scheme” to incentivise anonymised sharing. It didn’t materialise.

And then there’s the elephant in the room: regulation and AI safety measures. The Budget references “Safe and Trusted AI” as a pillar, but beyond the label, there’s precious little on how India plans to deal with deepfakes, algorithmic bias, or the kind of AI safety questions that the EU is already legislating around.

At a Glance: Key AI Initiatives

 The Verdict

On the whole, this is a Budget that gets the direction right. The money is real, the compute is tangible, and the language-first approach through BharatGen shows a level of strategic thinking that wasn’t always obvious in previous policy documents. The intent and focus on AI deserve a nod.