• •  More than 34,000 GPUs installed under IndiaAI Mission to fuel indigenous AI development.
• •  Three new AI startups Soket AI, Gnani AI, and Gan AI chosen to develop Indian foundation models.
• •  Emphasis on inclusivity: AI infrastructure and datasets tailored for Indian languages, sectors, and social equity.

India has made a big jump towards AI self-reliance, with the Ministry of Electronics and Information Technology (MeitY) saying that India now has more than 34,000 GPUs in its national AI compute infrastructure. This achievement, announced at the 'IndiaAI – Make AI in India, Make AI work for India' event in New Delhi on May 30, is part of a broader effort to democratize access to AI, promote indigenous innovation, and establish a sovereign AI ecosystem.

Union IT Minister Ashwini Vaishnaw underscored the government's aim of making state-of-the-art technology mainstream. "The technology should not continue to be in the hands of a select few. The IndiaAI Mission is about engineering level-playing field access to AI resources and catalysing innovation at scale," he reiterated.

Originally with a goal of 10,000 GPUs, the IndiaAI compute program already surpassed 34,000 GPUs, 18,417 of which have been installed as part of the first phase and 15,916 in the second. There is a third phase under consideration as well.

It is powered by this strong compute backbone to train and infer at scale for AI models, which will allow India to create its own large language models (LLMs) specific to Indian languages and needs.

367 datasets have been uploaded to the AI Kosh platform so far to act as fuel for basic AI building.

The ministry further announced the selection of three new startups under the IndiaAI Mission to build indigenous foundation models:

•Sarvam AI will spearhead India's sovereign LLM effort with an open-source 120-billion-parameter model for governance and citizen services.

•Soket AI is developing India's first multilingual 120-billion-parameter LLM tailored for industries such as defense, healthcare, and education.

•Gnani AI will develop a 14-billion-parameter Voice AI foundation model with multilingual, real-time speech processing capabilities.

•Gan AI is developing a 70-billion-parameter multilingual TTS (text-to-speech) model to reach "superhuman" capability.

Vaishnaw pointed out that this ecosystem will not just fuel the technological progress but also ensure reverse brain drain, talent attraction from across the world, and position India as a talent hub for AI innovation with Indian values and priorities.