Sarvam AI: The Indian Startup on a Mission to Build Bharat's Own Large Language Model

In a country of 1.4 billion people, fewer than 10% speak English as a primary language. Yet virtually every major AI model — from ChatGPT to Claude to Gemini — is built primarily on English-language data, with other languages treated as an afterthought. Sarvam AI, a Bengaluru-based startup founded in 2023, is setting out to change that. Its mission: build a sovereign Indian AI stack, starting with large language models that genuinely understand the languages, cultural nuances, and communication patterns of India's diverse population.

The Founding Story

Sarvam AI was co-founded by Vivek Raghavan and Pratyush Kumar, two AI researchers who had spent years studying the gap between AI's global progress and its relevance to Indian users. Raghavan previously led AI work at EkStep, the non-profit behind the DIKSHA education platform, and was deeply involved in building India Stack — the digital public infrastructure that powers Aadhaar, UPI, and DigiLocker. Kumar, a former researcher at Microsoft Research India, brought deep expertise in natural language processing for low-resource languages. Together, they identified a structural problem: India was being left behind in the AI race not because of a lack of talent, but because of a lack of data infrastructure and focused investment in Indian-language models.

The Technology: Sarvam-1 and Beyond

Sarvam's first model, Sarvam-1, was released in late 2023 and immediately drew attention for its performance on Indian-language benchmarks. Unlike multilingual models that spread their capacity thinly across dozens of languages, Sarvam-1 was purpose-built for 10 Indic languages — Hindi, Tamil, Telugu, Kannada, Malayalam, Gujarati, Marathi, Bengali, Punjabi, and Odia — with a training corpus curated specifically for Indian linguistic patterns.

The follow-up, Sarvam-2B, was a 2-billion-parameter model that the company open-sourced, making it freely available for Indian developers, researchers, and startups to build on. This decision — to open-source rather than commercialise the base model — was deliberate. The founders believe that India's AI ecosystem needs a shared foundation on which thousands of applications can be built, rather than a centralised gatekeeper that charges for access.

The India AI Mission Partnership

In 2024, Sarvam AI was selected as one of the anchor partners for India's ₹10,372 crore AI Mission, administered by the Ministry of Electronics and Information Technology. Under this partnership, Sarvam is working with the government to build a foundational Indian AI model — colloquially referred to as "India's own GPT" — that will be hosted on public cloud infrastructure and made available to government services, startups, and enterprises at subsidised rates. This is arguably the most significant public-private AI partnership India has ever attempted, and Sarvam is at its centre.

Why It Matters Beyond Technology

The significance of Sarvam AI extends beyond model benchmarks. If India is to realise the economic potential of AI — estimated at over $500 billion in GDP impact by 2030 — that benefit must reach its 900 million non-English-speaking citizens. A farmer in Andhra Pradesh asking about crop disease, a small business owner in Punjab seeking GST advice, or a patient in rural Maharashtra trying to understand a prescription: these use cases are only possible with AI that genuinely understands Indian languages. Sarvam AI is building the infrastructure layer that makes these applications possible. In that sense, it is not just an AI company — it is a piece of India's digital public infrastructure for the next decade.