The deal will also include ordering another 45,000 of Nvidia’s GB-300 GPUs (part of the Blackwell family architecture), which are currently the latest on offer by the company.
Confirming the development, Yotta Data Services Cofounder, Managing Director (MD) and Chief Executive Officer (CEO) Sunil Gupta said: “We are in the process of ordering 50,000 Vera Rubin GPUs and 45,000 GB 300s. The overall cost of acquiring the Vera Rubin chips would be around $7.5 billion, which includes all networking and associated equipment. An additional $750 million will also be required for data centre construction.”
The Yotta order under process will be the largest till now in the country. Until now, Hyderabad-based AMI Intelligence — part of the Greenko group — has made a binding order for 9,000 Vera Rubin GPUs for its Hyderabad intelligent factory.
Gupta said that the new GPUs will power Yotta’s upcoming 120 megawatt (Mw) D4 data centre in Delhi. “Our project timelines to go live with this is May-August 2027, for which deliveries of GPUs to our data centre shall be made around March-June 2027,” he said.
This next-generation GPU has 10x higher efficiency than the previous Blackwell architecture. It delivers 5x faster inference and 3.5x improved training compared to Blackwell. Rather than just generating basic text, it is custom-engineered to power agentic artificial intelligence (AI), enabling advanced models to seamlessly utilise tools and execute complex, multi-step problem-solving. The graphic chips use a 3 nanometre manufacturing process, which can push data memory speeds past 1,500 terabytes per second, lowering cost per token dramatically.
The demand for Nvidia’s GPUs has been on a roll — Amazon this week announced that it will deploy another 2 million of Nvidia’s high-end GPUs to run AI models in 2027 and 2028. The GPUs, apart from Vera Rubin, will also include Blackwell. Before this, Noetra Corporation, a consortium in Japan led by Softbank, announced in July that it will buy 27,500 Nvidia Rubin chips to build its own AI model for robots.
Nvidia officially announced in June that the Vera Rubin is going under mass production. By July 21, it again made a statement that it was officially shipping volume production of the chips to hyperscalers such as Google Cloud, Microsoft, and Oracle. Microsoft confirmed on August 21 that it has taken delivery of the first Vera Rubin systems from Nvidia.
The others who have received delivery of the chips include Open AI and SpaceX AI. SpaceX AI will use the chips to scale infrastructure for Grok and plans to launch a first-generation Starmind AI satellite powered by an orbital Vera Rubin system.
—Yotta is in process of ordering 50,000 Vera Rubin GPUs. Once closed, it will be the largest order from India till now
—The chips will power its upcoming 120 Mw data centre in Delhi, which will go live between May and August 2027
—The price tag will be around $7.5 billion, which will also include networking and other associated equipment costs
—Expecting deliveries of the chips between March and June 2027
