India’s AI mission portal’s launch soon: 10 firms to provide 14,000 GPUs
IndiaAI Mission will provide 14,000 GPUs for AI compute capacity, with over 10,000 already secured, benefiting various stakeholders at reduced costs.
The IndiaAI Mission will provide common compute capacity through about 14,000 graphics processing units (GPUs) after ten shortlisted companies met the lowest bids for different GPU types, a senior government official said, asking not to be named.
Of this, over 10,000 GPUs are already available with Yotta Data Services, E2E Networks, Tata Communications, and AWS’s managed service providers, while the remaining 4,000 GPUs will be purchased by companies including Jio Platforms and CtrlS Datacenters.
The ten companies had proposed providing 18,693 GPUs for the mission’s AI Compute pillar. After the IT ministry announced the lowest bids on January 30 for different types of GPUs, bidders had until February 14 to meet them. They ultimately met the L1 bids for about 14,000 GPUs.
Of the ₹10,371.92 crore ($1.25 billion) approved for the IndiaAI Mission in March, 44% ( ₹4,563.36 crore) is earmarked for compute capacity through more than 10,000 GPUs over five years. Companies will be empanelled for 36 months, extendable by 12 months.
Seventy percent of the promised GPUs are high-end, such as Nvidia’s H100. The remaining thirty percent are low-end GPUs with lower capacity or older generations.
{{/usCountry}}Seventy percent of the promised GPUs are high-end, such as Nvidia’s H100. The remaining thirty percent are low-end GPUs with lower capacity or older generations.
{{/usCountry}}Yotta Data Services accounts for the bulk of the compute capacity with 9,216 GPUs, including 8,192 Nvidia H100s.
{{/usCountry}}Yotta Data Services accounts for the bulk of the compute capacity with 9,216 GPUs, including 8,192 Nvidia H100s.
{{/usCountry}}AWS, bidding through four managed service providers (CMS Computers, Locuz Enterprise Solutions, Orient Technologies, and Vensysco Technologies), will provide 1,200 low-end GPUs — 800 AWS Inferentia 2 and 400 Trainium 1. Though these companies also bid for 2,400 Nvidia H100 GPUs, they declined to meet the lowest bid.
{{/usCountry}}AWS, bidding through four managed service providers (CMS Computers, Locuz Enterprise Solutions, Orient Technologies, and Vensysco Technologies), will provide 1,200 low-end GPUs — 800 AWS Inferentia 2 and 400 Trainium 1. Though these companies also bid for 2,400 Nvidia H100 GPUs, they declined to meet the lowest bid.
{{/usCountry}}Jio Platforms agreed to provide 208 H200 Nvidia GPUs and 104 AMD MI300X GPUs, where it was already the lowest bidder. The company plans to submit fresh, lower bids for other GPUs during continuous empanelment on April 30, the official said.
{{/usCountry}}Jio Platforms agreed to provide 208 H200 Nvidia GPUs and 104 AMD MI300X GPUs, where it was already the lowest bidder. The company plans to submit fresh, lower bids for other GPUs during continuous empanelment on April 30, the official said.
{{/usCountry}}Also read: Meta enters humanoid robotics race, eyes AI-powered consumer robots
The IndiaAI Compute Portal will also be launched in the coming days by IT minister Ashwini Vaishnaw. Various stakeholders, including union ministries and state governments, can request compute capacity through this portal.
Memo Details Subsidies and Services
The IndiaAI Compute Pillar sent a memo to all union ministries, departments, and chief secretaries about the discounted compute capacity, network, and storage services. The empanelment aims to serve Indian users including academia, researchers, MSMEs, startups, students, industry, and government departments and agencies.
IndiaAI Mission will cover up to 40% of compute costs for “eligible users”, according to the February 7 memo signed by Abhishek Das, general manager of IndiaAI Compute Pillar.
Also read: China's AI firms take spotlight with deals, low-cost models
The lowest bids averaged 42% below market price, costing companies ₹115.85 and ₹150 per GPU hour for low-end and high-end compute capacity, respectively. The additional government subsidy will reduce costs to under ₹100 per GPU hour for certain users.
A committee will be formed to evaluate subsidy proposals, with amount of subsidy varying by user category, the official said. Projects of “national importance” that benefit citizens could receive the maximum 40% subsidy.
The memo also detailed lowest bids for storage and network services: NxtGen Datacenter offered high-speed storage at ₹1.2/GB/month, Yotta Data Services provided object storage at ₹0.84/GB/month, while AWS MSPs and Jio Platforms offered “ingress” (uploading) network services for ₹0/GB/month, while E2E Networks offered “egress” (downloading) network services for ₹0/GB/month.
