Sarvam likely to launch India’s first large language model next year: IT minister

Indian AI startup Sarvam is likely to be the first company to roll out the country’s homegrown large language model (LLM) early next year, according to IT Minister Ashwini Vaishnaw.

In an interview with Moneycontrol, Vaishnaw said the government has scaled up its GPU capacity under the ₹10,371.92 crore IndiaAI Mission to 40,000 units, four times the original target.

“One of them, most probably Sarvam, will be the first to come off the mark with a reasonably sized, well-trained, Indian language-trained model, with data that is not having the bias of many other models,” Moneycontrol quoted the minister as saying.

Sarvam, founded in July 2023 by former AI4Bharat researchers Vivek Raghavan and Pratyush Kumar, secured the largest GPU allocation so far under the mission, receiving 4,096 NVIDIA H100 SXM GPUs via Yotta Data Services along with nearly ₹99 crore in subsidies, according to Moneycontrol.

The company was earlier named the first startup chosen to build India’s sovereign AI model.

Raghavan, who played a key role in building Aadhaar and other digital public goods, said Sarvam plans to co-develop domain-specific AI models with Indian enterprises and layer generative AI capabilities onto India’s digital infrastructure for public-good applications.

In December 2023, Sarvam raised $41 million in Series A funding led by Lightspeed Ventures, with participation from Peak XV Partners and Khosla Ventures.

The IndiaAI Mission, approved by the cabinet in March 2024, aims to establish a sovereign AI ecosystem by offering subsidies, a shared compute facility, an AI marketplace, and a national datasets platform. It also envisions indigenous large multimodal models and sector-specific foundational models for healthcare, agriculture and governance.

In March, Vaishnaw launched the IndiaAI compute portal, offering access to high-end Graphics Processing Units (GPUs) at rates as low as ₹67 per hour, in a bid to accelerate AI innovation and research in the country.

The IndiaAI compute portal–designed to build a scalable AI computing ecosystem for the country’s rapidly growing AI startups and research community–will provide access to compute more than 18,000 GPUs, cloud storage and other AI services to students, startups, researchers, academia, and government departments.

The portal will facilitate easy access to high-end and mid-range GPUs such as NVIDIA H100, H200, A100, L40S, and L4, AMD MI300x and 325X, Intel Gaudi 2, AWS Tranium and Inferentia along with network and storage services.