Nvidia H100 chips inside a server room at the Yotta Data Services Pvt. data center, in Navi Mumbai, India, March 14, 2024.
Dhiraj Singh | Bloomberg | Getty Images
India’s Yotta Data Services, which is building a $2 billion artificial intelligence hub using Nvidia‘s chips, said demand for graphic processing units in the country is exceeding supply as domestic AI models prepare to scale and the local user base surges.
At present, India trails the U.S. and China in the race to develop a native AI foundational model and lacks large domestic AI infrastructure. That is beginning to shift.
Last week, during the India AI summit, a few Indian companies launched early or limited versions of their AI models, such as Sarvam AI’s Indus chatbot.
“We’re gradually rolling out Indus on a limited compute capacity, so you may hit a waitlist at first. We will expand access over time,” Pratyush Kumar, co-founder of Sarvam AI, said in a post on X.
Most Indian AI models launched at the AI summit were trained on Nvidia’s GPUs hosted in Yotta’s facilities, Sunil Gupta, co-founder, managing director and CEO of the company told CNBC’s Inside India on Thursday.
The Mumbai-based data center company, which began sourcing Nvidia GPUs in 2023, now owns 60% to 70% of India’s GPU capacity, Gupta said. He added that demand is also expected to come from global AI companies as their user base in India expands.
Push for more data centers
In recent months, U.S. tech majors such as OpenAI, Google, and Perplexity have offered their AI tools at low or no cost to millions of users in India.
Among hyperscalers, Google has firmed up its plans to invest $15 billion to build a data center hub in southern India, while Microsoft will invest $17.5 billion to expand its data center footprint.
Last week, OpenAI became the first customer of India’s Tata Consultancy Services‘ data center business, signing up for 100 MW of capacity, with an option to scale to 1 GW.
“Through OpenAI for India, we’re working together to build the infrastructure, skills, and local partnerships needed to build AI with India, for India, and in India,” said Sam Altman, CEO of OpenAI, in a statement on Feb. 19.
As the Indian user base of leading global AI companies expands, Gupta said they will require local data centers and GPU capacity. Yotta plans to fund additional GPU purchases through a $1.2 billion to $1.5 billion pre-IPO round and aims to list within the next 12 months, Gupta added.
India has a total data center capacity of 1.93 gigawatt in 2025 and is projected to nearly double to 4 gigawatt by 2028, according to a Feb. 20 report by Nomura.
During the AI Summit, many companies announced plans to invest $277 billion over the next five to seven years, most of which will be directed towards building AI infrastructure in India, the brokerage said.
“Majority of these investments will flow into data centers, with domestic and US firms leading hyperscale buildouts and positioning India as a key US technology partner,” the brokerage said.