IndiaAI Mission: Govt relaxes norms for 10,000 GPU tender, to benefit smaller players

IndiaAI mission was launched in March 2024 to boost India's artificial intelligence capabilities

The government has relaxed eligibility norms for procuring 10,000 GPUs, which are critical for training AI models, under the IndiaAI mission, opening the bidding process to smaller companies and startups.

The information and technology (IT) ministry lowered the total turnover threshold for the bidding companies and relaxed the specifications of artificial intelligence (AI) compute units, the government said in revised tender requirements sent to the companies. Moneycontrol has reviewed a copy of the document.

The changes were made after companies during a pre-bid meeting in September said the provisions of the tender were restrictive for smaller players such as startups.

The average turnover requirement for primary bidders has now been lowered from Rs 100 crore to Rs 50 crore. For non-primary consortium members, the requirement has been halved to Rs 25 crore, the September 27 documents show.

Many small players, particularly new entrants, often struggle to meet the high turnover thresholds in tenders. By lowering the threshold, these companies can now participate in the procurement process.

In the original tender, the government had said bidders must have at least 1,000 AI compute units on their cloud platform.

While the government has retained that requirement, it has considerably relaxed the performance requirements of these units.

Originally, the IT ministry had said the 1,000 AI compute units should have a performance threshold of 15 TFLOPS for FP32, 300 TFLOPS for FP16, and 40 GB AI compute memory.

TFLOPS is a unit that measures the computing power of a system. Floating-point numbers or FP are used to handle very large or very small values with decimal points.

For example, if a system has 10 TFLOPS of FP16 performance, it means it can carry out 10 trillion FP16 calculations every second.

In the corrigendum, the government has reduced FP16 to 150 TFLOPS (from 300 TFLOPS) and reduced AI compute memory requirement from 40 GB to 24 GB.

Bidders can also provide a bank guarantee to account for anticipated AI compute units, giving them six months post-agreement to fulfil the requirement.

The reduction of FP16 from 300 TFLOPS to 150 TFLOPS and AI memory from 40 GB to 24 GB lowers the cost barrier for smaller players.

Smaller companies can now participate because they don’t need to invest in the most powerful, expensive hardware, industry insiders said.

In March, the government approved the Rs 10,732-crore IndiaAI Mission to create computing infrastructure in the country, AI centres for multi-modal LLMs, procuring 10,000 graphics processing units and more.