MUMBAI — A consortium of nine Indian institutes, backed by the government’s biggest single grant under a $1.2-billion national AI mission, is building a foundational large language model from scratch. The entity, BharatGen, received ₹ 1,058.82 crore on 18 September 2025. That money makes it the core of India’s push for sovereign AI.
BharatGen is a non-profit. It was established by the Indian Institute of Technology Bombay in December 2024.
Its founding director, Ganesh Ramakrishnan, a chair professor for computer science engineering at IIT-Bombay, argues the country has no choice. “We can only truly understand AI models when we build them from scratch,” he said. The stakes are straightforward.
Most Indian enterprises and startups today rely on foreign foundational models. That makes India a consumer of AI, not a builder.
Ramakrishnan drew an analogy to cooking. Knowing the recipe lets you tailor a dish to your health needs. The same applies to AI models — knowing what goes into them, especially the data, is essential for building effective tools.
“For AI adoption to scale, users have to move from the last mile to the first mile,” he said. “An ecosystem built solely around using AI tools will remain at the consumption layer. We need people who understand, build and improve.” The warning lands in a specific context.
India’s AI mission, announced by the central government, allocated $1.2 billion to build national capabilities. BharatGen is the largest single beneficiary.
The consortium brings together AI labs spread across nine Indian institutes. That structure is deliberate — a distributed network, not a single lab. BharatGen is now developing one of India’s first sovereign large language models.
These are the foundational AI systems that mimic human writing and speech by analyzing vast amounts of data. The project is building its first model from scratch.
That means not fine-tuning an existing foreign model. It means starting at the beginning. That choice carries weight.
Building from scratch is expensive and slow. It requires data, compute power, and expertise. But the consortium argues the alternative — relying on foreign models — leaves India dependent on systems built elsewhere, on data others chose, with priorities others set.
The government grant of ₹ 1,058.82 crore was announced on 18 September 2025. It is a large sum by Indian research standards.
It is also a bet. The bet is that India can build its own AI foundation, not just use someone else’s. Ramakrishnan’s cooking analogy is not casual.
It gets at a deeper point. If you only eat food someone else cooked, you never learn to cook.
If you only use AI models someone else built, you never learn to build them. You stay at the consumption layer. That is the risk BharatGen is trying to avoid.
The consortium is not a startup. It is a non-profit. That means its goals are not commercial in the usual sense.
It is not trying to build a product to sell. It is trying to build a capability the country can use.
That makes it unusual in the current AI landscape, where most big models come from large companies with deep pockets and clear profit motives. India’s AI mission is the framework. BharatGen is the largest single piece of it.
The consortium’s work will determine, in part, whether the mission succeeds or fails. The first model is under development.
No timeline has been given. The work is hard. The stakes are high.
Ramakrishnan put it plainly. “We can only truly understand AI models when we build them from scratch.” That is the logic. That is the plan.
Whether it works is not yet known.
{“@context”: “https://schema.org”, “@type”: “NewsArticle”, “headline”: “BharatGen receives ₹1,058.82 crore grant to build India’s foundational AI model”, “description”: “MUMBAI — A consortium of nine Indian institutes, backed by the government’s biggest single grant under a $1.2-billion national AI mission, is building a foundational…”, “datePublished”: “2026-06-25T11:33:00+00:00T09:30:00+00:00”, “author”: {“@type”: “Person”, “name”: “Sophie King”}, “publisher”: {“@type”: “Organization”, “name”: “InfoPulse Today”, “url”: “https://newsfeed.infopulsetoday.com/”}}
