India likely to possess 46K graphics processing units in a fortnight

India is expected to have around 46,000 graphics processing units (GPUs) in place by June 10, upon conclusion of the third round of bidding under the IndiaAI Mission, government officials said.

 

After the third round of the bidding, the Ministry of Electronics and Information Technology (Meity) will shift to a continuous empanelment process. Under this system, any company willing to match the lowest bid or per-hour cost for GPUs discovered during the first round of bidding will be eligible to supply these high-performance computing machines to participants in the ₹10,372 crore IndiaAI Mission, the officials said.

 

Results for the second round of GPU bidding were announced on Friday by Union Minister for Electronics and Information Technology Ashwini Vaishnaw. 

 

“Developing a common compute is a very important part of the principle of democratising technology for all. When we started, we targeted 10,000 GPUs. We had then thought that this would be a very big and ambitious target. But the kind of response we have seen is phenomenal,” he said.

 

In the first round, the government received bids from 10 companies to procure and supply 18,693 GPUs, well above the initial 10,000-unit target. The bidders included CMS Computers, CtrlS Datacentres, E2E Networks, Jio Platforms, Locuz Enterprise Solutions, NxtGen Datacentre and Cloud Technologies, Orient Technologies, Tata Communications, Vensysco Technologies, and Yotta Data Services. 

 

The second round of bidding added 15,640 GPUs, secured by seven firms: Cyfuture India, Ishan Infotech, Locuz Enterprise Solutions, Netmagic IT Services, Sify Digital Services, Vensysco Technologies, and Yotta Data Services.

 

Of the GPUs secured in the first round, 4,000 have been allocated to Sarvam for six months to develop its indigenous large language model (LLM), and another 3,000 have gone to the Meity’s Bhashini division for language translation work, according to a senior government official.

 

The ministry continues to evaluate applications and allocation requests. Because projects such as LLM development require large GPU clusters from a single vendor, some companies that succeeded in the first round were also selected in the second, the official added.

 

On Friday, Meity also announced that three new indigenous LLMs would be developed by Gnani AI, Gan AI, and Soket, in addition to the one already underway at Sarvam.

 

Soket is set to build India’s first open-source 120 billion-parameter foundation model, optimised for linguistic diversity and aimed at sectors like defence, health care, and education. Gnani AI will develop a 14 billion-parameter voice artificial intelligence model designed for real-time speech processing and advanced reasoning. Gan AI, meanwhile, will work on a 70 billion-parameter multilingual foundation model with a focus on “superhuman text-to-speech” capabilities, according to the ministry.

 

“Like Sarvam, these three teams also have a very big target in front of them. Whichever sector they focus on, they must be among the top five in the world. That’s a clear target. That is something which is achievable also because today foundational models are becoming, in a sense, one large base, but simultaneously many focused sector-focused models,” Vaishnaw said.

 

LLMs use vast data repositories and compute power to first learn and then be able to generate human-like responses to a user’s inputs, including generating a new idea from scratch.

 

Both the GPU procurement effort and the LLM development initiatives are being carried out under the IndiaAI Mission.