The government has shortlisted nearly eight homegrown companies developing artificial intelligence (AI) foundational models to provide incentives under the IndiaAI Mission, according to people familiar with the matter.
Union IT and Electronics Minister Ashwini Vaishnaw is likely to announce the selected beneficiaries from the second round of screening next week.
BharatGen, a consortium of IIT researchers led by IIT Bombay and backed by the Department of Science and Technology, is among the frontrunners expected to receive approval in this round, sources indicated.
In January, the Ministry of Electronics and Information Technology (MeitY) therefore announced an incentive pool of ₹1,500 crore to encourage organisations and individuals who are working on building AI models from the ground up.
By February 15, the ministry had received 67 applications from Indian and global startups and researchers, with another 120 applications submitted the following month. The total number of applications has now crossed 500.
In the first round of approvals, MeitY granted four startups — Sarvam, Soket Labs, Gnani.ai, and Gan.ai — access to subsidised GPU compute to develop indigenous AI models.
In July, BharatGen unveiled a 2.9-billion-parameter bilingual large language model (LLM) named Param 1. Researchers pre-trained the open-source model on five trillion tokens in both English and Hindi.
Senior researchers at BharatGen did not respond to requests for comment.
At the same time, access to GPUs under MeitY’s common compute facility has become an increasing challenge.
“We are facing a chicken-and-egg story,” said one data centre executive who did not wish to be quoted. “Until recently, Indian startups had no access to affordable GPU compute. The government stepped in to address that, and suddenly we saw a surge of interest from startups and researchers building AI models. But now we’re back to a supply crunch, since many of the GPU purchase orders exist only on paper.”
Within the ministry, there is now a growing view that future rounds of GPU procurement should also prioritise inferencing models that deliver higher efficiency, rather than focusing solely on training models, according to government officials.
Out of the proposed 34,333 GPUs under the IndiaAI Mission, the current installed capacity stands at 17,374 GPUs.
