India Wants NVIDIA To Provide Indian Startups With AI-Capable GPUs

India is planning to join forces with chip-making giant NVIDIA and provide graphics processing units (GPUs) and Neural Processing Units (NPUs) to Indian startups. The plan is to offer these GPUs and NPUs at lower prices to local startups, researchers, academic institutions, and other users. This move will aim to strengthen the AI infrastructure in the country. 

This initiative is said to cost India around ₹10,000 crore and is still at a nascent stage. The Economic Times has reported that the final decision is expected to come after the 2024 general elections.

NVIDIA holds the majority share in the GPU market globally and has become the first option for India if the country wants to achieve its AI computing ambitions. Countries and companies across the globe are betting big on AI computing infrastructure and are lining up to get their hands on NVIDIA GPUs, especially the H100 chips.

Also Read: PS5 Pro Details Leaked: Significant GPU Upgrade Expected

India’s Strategy To Get And Distribute AI NPUs And GPUs

India seems to have two plans to get access to and distribute these AI-powered NPUs and GPUs. One of the approaches includes renting and subletting, where the government will provide GPUs to startups, researchers, and others at a considerably lower price.

Another approach involves a marketplace model where companies will be encouraged to negotiate rental or subletting deals with NVIDIA directly. Similar to India’s PLI scheme, incentives will be handed out upon achieving increased productivity using the GPU.

Getting access to NVIDIA GPUs isn’t easy for companies, especially the ones that have just started operations, due to the limited supplies and high prices. Usually, NVIDIA’s H100 GPUs cost around $50,000 each, while the latest Blackwell card sells for around $40,000.

Also Read: OpenAI Launches GPT-4 Turbo With Notable Upgrades

How Many GPUs Will Startups Need?

Even for the simplest AI data center that is capable of handling existing AI models, at least 100 to 300 GPUs are required. If a company is looking to set up more robust computing power, which is capable of handling complex tasks like developing, and training LLMs, it would require around 5,000 to 10,000 GPUs.

To put things in context, India’s supercomputer ‘AIRAWAT,’ situated at the Centre for Development of Advanced Computing (C-DAC) in Pune, needs around 640 GPUs to function and is the 75th fastest supercomputer in the world. However, when compared to the fastest supercomputer in the world, which packs in over 30,000 GPUs, the difference is huge and India needs to narrow it down to brush shoulders with other nations in AI research and development.

Last month, the Union Cabinet announced a Rs. 10,372-crore fund for the country’s AI Mission. The initiative aims to deploy 10,000 GPUs through public-private partnerships. With the rent-and-sublet model, the government intends to provide GPUs to eligible startups.