In February 2026, Pratyush Kumar and Vivek Raghavan, co-founders of Bangalore-based Sarvam, unveiled India's first large language models trained from scratch in India, Sarvam 30B and Sarvam 105B, at the India AI Impact Summit in New Delhi. The launch marked the culmination of three years of work building a full-stack sovereign AI platform: foundational models for 22 Indian languages, voice agents deployed across enterprise and government programs, and an enterprise conversational platform that had processed over 140 million conversations. Sarvam had been selected by the Indian government as its primary institutional bet for a domestic foundation model under the IndiaAI Mission and had signed state-level infrastructure agreements with Tamil Nadu and Odisha.

As the summit concluded, the founders faced strategic questions. Could Sarvam realistically own every layer of the AI stack, models, infrastructure software, and applications, simultaneously, and was this full stack ambition a defensible moat or an overextension? Second, what did sovereign AI mean in practice for a company with a national mandate, an open-source philosophy, and the commercial need to scale? What would it take for Sarvam to become the gold standard for sovereign AI in India?