Yotta to invest $2bn for AI hub with top Nvidia chips

New Delhi: India is set to host one of Asia's largest AI superclusters, with Yotta Data Services unveiling plans to deploy more than 20,000 Nvidia Blackwell Ultra GPUs in a $2 billion investment. The facility is expected to go live by Aug 2026. The infrastructure, to be housed at Yotta's Greater Noida hyperscale campus and supported by its Navi Mumbai operations, marks a significant escalation in the country's AI compute ambitions.

For Sunil Gupta, co-founder & CEO of Yotta, the bet on scale is rooted in a structural shift in demand. He told TOI that India's AI ecosystem has always had talent, data and startups, but lacked compute. That gap, he said, is something Yotta identified years ago as a potential bottleneck. Today, Yotta runs around 10,000 GPUs, with more being added, and Gupta argues the next phase will be even more compute-intensive.

Much of the industry's attention has focused on training frontier models. Gupta, however, insists inferencing (where the trained model generates content, predictions and decisions) will ultimately require even larger volumes of infrastructure. As Indian-built models such as those serving agriculture, healthcare and education move into production and reach population scale, "millions and billions of people will come concurrently" to access them, he said, requiring "thousands and thousands of GPUs purely for inferencing."

Gupta suggested that India could serve large parts of the APAC region and Global South from its data centres. Yotta is in discussions with multiple countries exploring whether their AI capacity could be delivered from India.