India's Compute Capacity To Surge: Yotta CEO Predicts 1 Lakh GPUs Within 18 Months
Speaking to NDTV Profit, Gupta said the India AI mission has already aggregated 38,000 GPUs through public-private partnerships, with demand far outstripping supply.
India’s compute capacity is set for a dramatic scale-up, with the country poised to cross 1 lakh GPUs within the next 18 months, according to Sunil Gupta, Co-Founder and CEO of Yotta Data Services.
Speaking to NDTV Profit at the sidelines of the Assocham AI Conclave in New Delhi, Gupta said the India AI mission has already aggregated 38,000 GPUs through public-private partnerships, with demand far outstripping supply.
"If you look at the 500-plus proposals received by the government for training large and small language models, the requirement easily crosses 1–2 lakh GPUs. In my view, 1 lakh GPUs will be operational in the next 18–24 months, possibly sooner," he said.
Gupta highlighted that Yotta began investing in GPU infrastructure ahead of the global AI boom, tying up with Nvidia in late 2023 and going live with its first batch by March 2024. "We are among only six providers globally operating on NVIDIA’s reference architecture, capable of training multiple LLMs simultaneously while supporting inference platforms like Bhashini," he noted.
Beyond compute, Gupta said India’s data centre industry is witnessing exponential growth. From just 200 MW capacity until 2017, the sector has scaled to 1.4 GW today, and is projected to hit at least 3 GW by 2030. But with the AI wave, he expects even higher acceleration. "If you go by US benchmarks, India could need 50 times more capacity. The demand curve is massive," he said.
Gupta added that India’s population-scale adoption of digital services means the need for GPU power will remain long after model training. "Inference for applications catering to tens of crores of users will dwarf even training requirements," he said, calling India’s AI and data infrastructure buildout “one of the fastest globally."
