The Indian Govt Is Planning To Build A 25,000 GPU Cluster To Support AI Startups

An AI revolution has been quietly brewing on foreign shores with innovations including ChatGPT and Stable Diffusion, but India has been largely left out of the action. But India’s AI startups might get a helping hand from an unlikely source to help kickstart its own AI industry.

The Indian government considering a proposal to build a 25,000 GPU cluster to support AI startups. GPU stands for Graphics Processing Unit, which is a high-powered version of the CPU, and used to create AI models and applications. There is currently a worldwide shortage of such GPUs owing to the high demand from the AI industry, but a massive 25,000 GPU cluster that’s built by the Indian government could help provide much-needed computing power for Indian AI startups.

The GPU cluster will cost around Rs. 8000-10,000 crore to build. The project will likely be built through a public-private partnership. “We will create in the first phase a 25,000 GPU cluster of AI compute capacity,” an official told ET. “It will be done as a PPP and the AI compute capacity will be offered as a service to startup.” The proposal may be finalized as early as December this year, the person added.

Indian entrepreneurs hailed the move. Snapdeal founder Kunal Bahl, who also heads the CII National Startup council, called the move a “visionary step that will accelerate innovation” and technological progress. “By demo-cratising AI infrastructure, we are empowering the next generation of entrepreneurs to dream big, create boldly, and transform our nation,” he said.

GPUs are the innovation that have powered much of the recent AI revolution. They were initially built to allow advanced computer games to run on personal machines, but AI researchers found that they were very useful for computing large amounts of data, which are vital for AI applications. AI researchers used much of the theory developed in the late 2000s on modern GPUs, and were able to get incredible new results, such as LLMs and image-generation program like Stable Diffusion and Midjourney. These companies use GPU clusters, like the one the Indian government is planning to build, and train their models over many weeks and months to be able to generate the results they do.

This has meant that there’s currently a shortage of such GPUs. NVIDIA, the company that makes most of these GPUs, has seen its stocks soar in recent times, and GPUs are even being sold on the black markets for several times their price. In India, GPUs are made available through foreign providers, or through a handful of Indian companies like E2E Networks, which have some GPU capacity available. But there aren’t enough GPUs to go around, especially for training large AI models. If the Indian governments can build its own GPU cluster and make it available to Indian companies, it could both reduce GPU costs for Indian companies, and reduce their reliance on foreign firms for what are essentially the building bricks of the AI space.

And it’s this self-reliance that’s been touted by the Modi government in recent times. This self reliance could end up being particularly crucial in the world of AI — AI is likely to upend most sectors, including critical sectors like computing, healthcare, warfare and defence. And having its own AI cluster could be a crucial component of protecting Indian interests as the AI revolution sweeps through the world in the years to come.