Sarvam may get first IndiaAI Mission grant, of Rs 220 crore

Sarvam AI may be the first to receive funding from the govt's IndiaAI Mission , as the country seeks to build an indigenous large language model (LLM) like OpenAI's GPT and China's DeepSeek. Sources told TOI that Sarvam is expected to receive the equivalent of Rs 220 crore, mainly as free access to some 4,000 Nvidia H100 chips for about six months. These chips are crucial for training AI models.

Sarvam is expected to use the Nvidia chips to improve its existing language model. The startup, founded by tech professionals and researchers Vivek Raghavan and Pratyush Kumar, and incubated in IIT Madras, launched a 2 billion parameter model late last year, specifically designed for Indian languages. When TOI asked S Krishnan, Secretary in the Ministry of Electronics and Information Technology (MeitY), about Sarvam being selected for the grant, he declined to comment. An email sent to Sarvam didn’t elicit a response by the time of going to press.

Last year, the govt allocated over Rs 10,300 crore towards the IndiaAI Mission. These funds are set to support multiple elements of the mission, encompassing crucial projects such as the IndiaAI Compute Capacity, IndiaAI Innovation Centre (IAIC), and IndiaAI Datasets Platform. A significant aspect of the mission involves establishing a sophisticated shared computing facility containing 18,693 GPUs.

Sarvam says its current model beats global models in several benchmarks related to Indian languages. The company is seen to have tokenised Indian languages more efficiently, enabling the model to generate faster and more accurate responses. It's not clear what other benchmarks it will now try to beat with the compute capacity it receives with the govt grant.

IndiaAI Mission has received over 190 proposals for creating a domestic AI model, both large language models (LLMs) and small language models (SLMs).