India's AI Moment: Sarvam, Krutrim, BharatGPT and AI4Bharat

For years the story of large language models was written largely in English and largely outside India. That is changing. A homegrown Indian AI ecosystem has taken real shape, aimed squarely at the country's linguistic diversity and its ambition for technological self-reliance. This is a moment worth marking — and, told honestly, it is genuinely impressive.

Sovereign models and startups

Indian startups have moved from talk to shipping. Sarvam AI has focused on models built for Indian languages and voice, positioning itself around the practical realities of a multilingual population. Krutrim, backed by the Ola group, has pursued its own family of models and the compute to train them, becoming one of the more visible names in the space. Alongside them, BharatGPT and related efforts have targeted conversational systems tuned for Indian languages and use cases. These are not translations of foreign models but attempts to build for the country's needs from the ground up.

The AI4Bharat foundation

Underpinning much of this is AI4Bharat, the research initiative associated with IIT Madras, which has done foundational work on Indic-language datasets, benchmarks and open models. Building good models for Indian languages is hard precisely because high-quality digital text in many of them is scarce; AI4Bharat's work on corpora, speech data and evaluation has been an enabling layer that others build on. It is a reminder that a lot of the most important AI work is unglamorous infrastructure rather than flashy product.

The policy push

The state has put weight behind the effort. The IndiaAI Mission, the government's flagship programme, has been framed around expanding compute capacity, supporting datasets and encouraging the development of foundational models suited to India. Reported ambitions include subsidised access to GPUs and support for building sovereign models, part of a broader goal of not depending entirely on foreign providers for a strategically important technology. As with any large public programme, execution will determine how much of the ambition is realised, but the direction is a serious one.

Why it matters

The case for Indic-first AI is practical, not just symbolic. India's languages span an enormous population that global models often serve poorly, and the country's use cases — from agriculture to public services delivered in many tongues — reward models that genuinely understand local context. Voice matters too, in a market where many users are more comfortable speaking than typing. Models built with these realities in mind can reach people that English-centric systems simply do not.

There is real work still ahead, and honest observers will note that matching the global frontier on raw capability remains a challenge. But the ecosystem is no longer aspirational. With capable startups, a strong research base in AI4Bharat and a national mission behind it, India's AI moment is under way — and it is being built, deliberately, for India.