About Rs 5,000 crore earmarked for AI chips: MeitY’s Abhishek Singh

The government will not take the voucher approach to give the AI chips' access to startups as it can be abused, said Singh.

The government has earmarked about Rs 5,000 crore of the total India AI Mission corpus of Rs 10,000 crore for making access to AI chips available to the country's startups and researchers, a top official said.

"We have almost Rs 5,000 crore earmarked to make access to 10,000 GPUs available to startups and researchers," Abhishek Singh, Additional Secy, Ministry of Electronics and Information Technology (MeitY) said at the Global IndiaAI Summit in Delhi on July 4.

"We will not buy chipsets. We will subsidise a part of the cost of the compute. The way we are building the procurement model is users will decide which AI chips they want. If anyone wants H100, they will get that. If anyone wants Gaudi 3, they will get that," he added.

Moneycontrol reported last week that the India AI programme will not bet the farm on Nvidia's chips, and instead be open to other companies like Intel and AMD.

Being the AI chip supplier to Meta and Microsoft-backed OpenAI has not only helped Nvidia corner almost all of the AI chip demand as yet, but also maintain its pricing power much ahead of its competitors in the market. While chipmarkers like Intel and AMD's gross margins are lower than 50 percent, Nvidia sits atop the rest at around 80 percent.

During the summit, Singh also said that the government will soon float a request for proposal for the AI chip scheme.

Moreover, the government will not take the voucher approach to give the AI chips' access to startups as it can be abused, said Singh.

The forthcoming request for proposal (RFP) will outline the government's initial GPU capacity requirements and the timeline for deployment.

The RFP is also expected to clarify whether data centres/cloud service providers in other countries will be eligible to participate in the bidding process.

The industry has advised the government that it should not invest directly in the capex for GPUs. It has suggested that the government should subscribe to the GPU capacity put up by domestic data centre companies and allocate it to startups, rather than fund capex.

Apart from other hindrances in building home-grown foundational models, India lacks domestic compute capacity. GPUs, which are primarily designed to render images and videos quickly and efficiently, are the hardware on top of which AI models can be trained. They are primarily housed in a data centre.