Sarvam AI to receive access to more than 4,000 Nvidia GPUs to build India’s first homegrown AI model

Bengaluru-based AI startup Sarvam has been selected by the Indian government to build the country’s first indigenous large language model (LLM), marking a significant milestone under the Rs 10,370-crore IndiaAI Mission. 

Chosen from a pool of 67 applicants, Sarvam becomes the first startup to receive support under the programme, with access to substantial compute resources to build the model from scratch.

Indian govt to provide high-end GPUs

The government will provide Sarvam access to 4,096 high-end Nvidia H100 GPUs for six months, sourced from companies empaneled to set up AI data centres in India, including Jio Platforms, Yotta, Tata Communications, E2E Networks, NxtGen Datacenter, CMS Computers, Ctrls Datacenters, Locuz Enterprise Solutions, Orient Technologies, and Vensysco Technologies. 

The startup’s model will be trained, deployed, and optimized entirely within India, using local infrastructure and talent, a move that underscores India's ambitions for strategic autonomy and leadership in the global AI landscape.

"This model will have 70 billion parameters and many innovations in programming as well as engineering," said IT Minister Ashwini Vaishnaw. "With these innovations, a 70-billion-parameter model can compete with some of the best in the world."

Who are the founders?

Founded by Vivek Raghavan and Pratyush Kumar, both former members of AI4Bharat, Sarvam plans to build a suite of models tailored for India’s unique linguistic and operational needs. These include Sarvam-Large for advanced reasoning and generation, Sarvam-Small for real-time interactive applications, and Sarvam-Edge for lightweight, on-device use cases.

In a statement, Sarvam said its model will be capable of sophisticated reasoning, designed for voice-first interactions, and fluent across Indian languages, aiming for secure and population-scale deployment.

"This is a crucial step toward building critical national AI infrastructure," said Raghavan. "Our goal is to build multi-modal, multi-scale foundation models from scratch. When we do, a universe of applications unfolds. For citizens, this means interacting with AI that feels familiar, not foreign. For enterprises, it means unlocking intelligence without sending their data beyond borders."

Rising global AI competition

The development comes at a pivotal moment in the global AI race, following the rise of China's DeepSeek, a low-cost open-source foundational model that has gained significant attention for its performance and efficiency.

India’s strategic push to create homegrown AI capabilities through initiatives like the IndiaAI Mission reflects both a need for technological sovereignty and an ambition to position itself as a global innovation hub.

While Sarvam’s LLM is not expected to be open-sourced, it will be heavily fine-tuned for Indian languages and cultural contexts. The government, meanwhile, continues to evaluate hundreds of other proposals under the IndiaAI Mission, which originally aimed to procure 10,000 GPUs but has since expanded to acquire 18,693 GPUs to support AI startups and researchers.