Government relaxes eligibility norms for procuring 10,000 GPUs to benefit small players

The government has relaxed eligibility norms for procuring 10,000 graphic processing units (GPUs) which are imperative for training artificial intelligence (AI) models under the IndiaAI mission, allowing smaller companies and start-ups to participate in the bidding process.

The Information Technology (IT) ministry lowered the total turnover threshold for bidding companies and relaxed the specifications of AI compute units. The changes were made after companies complained about the tender’s restrictive provisions for smaller players.

The average turnover requirement for primary bidders has been lowered from Rs 1000 million to Rs 500 million, while for non-primary consortium members, it has been halved to Rs 250 million.

While the requirement for bidders to have at least 1,000 Al compute units on their cloud platform remains, the performance specifications for these units have been significantly relaxed. Initially, bidders were expected to meet performance thresholds of 15 TFLOPS for FP32, 300 TFLOPS for FP16, and provide 40 GB of Al compute memory. The revised requirements now state that the FP16 performance threshold is reduced to 150 TFLOPS, and the Al compute memory requirement has been lowered to 24 GB.

To further accommodate smaller players, bidders can provide a bank guarantee for anticipated Al compute units. This allows them six months after the agreement to fulfil the requirement, providing additional flexibility.