The Centre is moving to build a pool of government-controlled artificial intelligence (AI) ecosystem by adding 10000 GPUs of computing capacity, sources have said.
The government is working with Centre for Development of Advanced Computing (C-DAC) to install 3,000 GPUs of computing capacity, while also finalising an agreement with the UAE under which an 8-exaflop AI supercomputer, equivalent to about 10,000 GPUs of computing capacity, will be deployed in India.
This is being developed through Abu Dhabi-based G42 and Cerebras in partnership with C-DAC and the Mohamed bin Zayed University of AI, and will be owned by the Indian government as a “UAE government gift”, a senior government official told businessline.
The official further said that the government is also preparing further bids for additional computing capacity, as it seeks to build a government-controlled pool of AI compute alongside capacity sourced from private providers.
“We are working with C-DAC on a 3,000 worth of capacity to be installed and then we are also working on the other G42 termsheet which was signed in the presence of the Prime Minister. We are doing the final agreements to get further compute. That will be worth of about 10,000 GPUs worth of compute,” the official said.
It will be owned by the government because it is a gift from the UAE government to the Indian government, he said.
“We will be setting up the data centres…that was the agreement signed between the two governments,” the official said.
The India AI Mission currently has access to about 30,000 GPUs against commitments of 45,000, with several private providers unable to deliver their pledged capacity amid a sharp rise in equipment costs.
Asked about the pricing of these GPUs, the official said the government has not decided on them yet.
“We will have to see…we will have to adjust the subsidies according to what the startups get. It will take at least three-four months and then we are going to float bids… It is still in RFP stage,” the official added.
According to analysts, the government has to provide more clarity on the ecosystem it is building, especially about whether the GPUs are of latest generation.
“There is currently very little clarity on what specific mix of AI accelerator chips the government plans to install and targeting which cohort and use-cases training/fine-tuning vs inferencing,” Neil Shah, Vice President-Research at Counterpoint Research told businessline.
He said total budget for the AI Mission is set at ₹10,372 crore over five years and the key question now is whether the government will increase this budget to keep up with rising costs and secure advanced chips.
If the government primarily deploys chips that are already three to four years old, they will be inefficient in terms of both power and performance, he noted. The Hindu Businessline
Leave a Reply