CIO Insider

CIOInsider India Magazine

Separator

IndiaAI Mission to Install 14,000 GPUs for Compute Capacity

CIO Insider Team | Tuesday, 18 February, 2025
Separator

Ten businesses fulfilled the lowest bid requirements for different types of graphics processing units (GPUs), allowing the IndiaAI Mission to increase its compute capacity by about 14,000 GPUs. This action is a component of the government's larger plan to improve the nation's artificial intelligence infrastructure.

The mission will also roll out a subsidy plan covering up to 40 percent of compute costs for eligible users, aiming to support research, academia, startups, and government agencies.

More than 10,000 of the 14,000 GPUs obtained for the project are currently available through managed service providers Yotta Data Services, E2E Networks, Tata Communications, and AWS.

Companies like CtrlS Datacenters and Jio Platforms will purchase the remaining 4,000.

The ten businesses' first offers included 18,693 GPUs; however, the final number was decided when bidders met the lowest bid rates established by the IT ministry on January 30. Businesses had until February 14th to match the L1 pricing with their bids.

Seventy percent of the GPUs will be high-end, such as Nvidia's H100, with the other thirty percent being lower-end or older versions. 8,192 of the 9,216 GPUs owned by Yotta Data Services are Nvidia H100s, giving them the greatest stake.

AWS will supply 1,200 low-end GPUs, including 800 AWS Inferentia 2 and 400 Trainium 1, via its managed service providers CMS Computers, Locuz Enterprise Solutions, Orient Technologies, and Vensysco Technologies. AWS MSPs decided not to match the lowest bid even though 2,400 Nvidia H100 GPUs were up for auction. Jio Platforms, meanwhile, has acquired 208 Nvidia H200 GPUs and 104 AMD MI300X GPUs. They want to place fresh, reduced bids in the next empanelments on April 30.

IT Minister Ashwini Vaishnaw is scheduled to launch the IndiaAI Compute Portal shortly, which would make GPU resources more accessible to state governments and union ministries alike. The platform would streamline the allocation of AI resources by enabling users to request compute capacity.

The IndiaAI Compute Pillar's pricing structure and subsidy scheme were described in a memorandum signed by General Manager Abhishek Das on February 7. The lowest bids for GPU compute services, which cost Rs 115.85 per GPU hour for low-end GPUs and Rs 150 per GPU hour for high-end GPUs, averaged 42% less than the market price.

Companies will be impaneled for an initial period of 36 months, extendable by another 12 months

For some customers, such as researchers, startups, MSMEs, and academic institutions, government subsidies will further lower costs to less than Rs 100 per GPU hour. Projects of national significance will be eligible for the maximum 40 percent subsidy, and a committee will be formed to evaluate bids for subsidies.

The mission also obtained competitive bids for network and storage services: AWS MSPs and Jio Platforms will offer free data ingress services, Yotta Data Services will offer object storage at Rs 0.84 per GB per month, and NxtGen Datacenter will offer high-speed storage at Rs 1.2 per GB per month. Free egress (downloading) network services will be provided by E2E Networks.

The IndiaAI Mission, backed by Rs 10,371.92 crore ($ 1.25 billion) approved in March, has earmarked 44 percent of its budget—Rs 4,563.36 crore—specifically for compute capacity expansion over the next five years. Companies will be impaneled for an initial period of 36 months, extendable by another 12 months.



Current Issue
Smartphones Are Ruling The Genai Era



🍪 Do you like Cookies?

We use cookies to ensure you get the best experience on our website. Read more...