India GPU Infrastructure: Data Center Boom, Reliance Jio, and Bangalore Hubs

India's GPU infrastructure market is growing at 35-40 percent annually, the fastest rate of any major economy, driven by a confluence of factors: the IndiaAI Mission's Rs 10,372 crore ($1.25 billion) compute allocation, Reliance Industries' entry into AI infrastructure, and surging demand from India's IT services and startup ecosystems. Total Indian data center capacity reached 950 MW in 2026, with 600 MW additional under construction, and the GPU count is estimated at 15,000-22,000 H100-equivalent units, growing to 40,000-50,000 by 2028 by the India Data Centre Alliance's projections.

The market divides into three segments: hyperscaler cloud GPU capacity (AWS Mumbai, Azure Central India, GCP Mumbai), domestic provider capacity (Yotta, CtrlS, NTT India, STT GDC India), and captive infrastructure built by IT services companies and financial institutions. Tata Consultancy Services (TCS), Infosys, and Wipro have collectively deployed 5,000+ H100 GPUs across their AI CoE (Centers of Excellence) for client workloads. Power costs of INR5-8 per kWh ($0.06-0.10) are competitive with US rates, and labor costs for data center operations are approximately 60-70 percent below developed market levels.

Reliance Industries has placed the largest single GPU infrastructure bet in South Asia. Through Jio Platforms, Reliance purchased 10,000 NVIDIA H100 GPUs in 2025-2026 as part of a broader partnership that includes joint development of an India-specific LLM and AI platform. The GPUs are deployed across Jio's data center in Navi Mumbai (the Jio AI-Cloud Center) and a new facility being built in Jamnagar, Gujarat, adjacent to Reliance's refining complex. The Jamnagar site benefits from Reliance's captive power generation at effective rates of INR3-4 per kWh ($0.035-0.048), among the lowest power costs for any GPU deployment globally.

The Jio GPU cluster is being used for three primary workloads: training the BharatGPT family of Hindi and regional language models, running Jio's internal AI services across telecom (500 million subscribers) and retail (JioMart), and leasing GPU capacity to Indian enterprises and startups through Jio's cloud platform at INR200-280 per GPU-hour for H100. This last component is a strategic move to disrupt the Indian GPU market, which has historically relied on AWS and Azure at INR350-600 per GPU-hour. Jio's pricing has already triggered price reductions from hyperscaler competitors in the Indian market.

Bangalore hosts the densest concentration of AI companies in India, creating strong local demand for GPU infrastructure. The city's GPU capacity is anchored by Yotta's HYDRA supercomputer at the Yotta Panvel facility (physically in Navi Mumbai but serving Bangalore with 8ms latency), plus CtrlS' Bangalore campus, and STT GDC India's Whitefield facility. Yotta's HYDRA was India's first large-scale AI supercomputer at 1,500 H100 GPUs when launched in 2024, and has since expanded to 5,000 H100 and 2,000 H200 GPUs. HYDRA is used by Indian AI startups, research institutions, and enterprise clients for LLM training and fine-tuning at INR250-350 per GPU-hour.

Bangalore faces significant power reliability challenges that affect GPU infrastructure. Karnataka state experiences 5-15 hours per month of grid instability, and monsoon season brings additional risk of flooding in Whitefield and Bellandur areas where several data centers are located. Operators like CtrlS and STT GDC maintain 8-12 hours of battery backup plus on-site diesel generation for their Bangalore facilities, adding 8-12 percent to total cost of GPU operations versus Mumbai. Despite these challenges, Bangalore's talent ecosystem makes it indispensable: 60 percent of India's AI startups are headquartered in Bangalore, and the city's GPU utilization rates consistently exceed 85 percent versus 70-75 percent for Mumbai.

Mumbai is India's primary submarine cable landing point, with 10+ cable systems including the SEA-ME-WE-5, SEA-ME-WE-6, AAE-1, and the 2Africa system connecting India to Europe, Africa, Southeast Asia, and the Middle East. Total international capacity through Mumbai exceeds 200 Tbps as of 2026. This bandwidth abundance makes Mumbai the preferred location for GPU workloads that require cross-border data transfer, such as Indian IT services companies running models for US and European clients. Navi Mumbai's Panvel area has emerged as the specific GPU corridor, with Yotta, Reliance Jio, and NTT all operating within a 5km radius served by redundant fiber from the Mumbai cable landing stations.

Chennai serves as India's second cable gateway, with the Chennai-Andaman and Nicobar Islands (CANI) cable and multiple Southeast Asian connections. Chennai's GPU capacity is smaller but growing, serving the manufacturing AI sector (automotive, electronics in the Chennai-Bangalore industrial corridor) and the emerging AI research ecosystem at IIT Madras. Chennai's advantage is lower power costs (INR5-8 per kWh versus INR6-9 in Mumbai) and the Tamil Nadu government's data center incentive policy, which offers 5-year GST reimbursement and 100 percent power tariff concessions for AI data centers above 10 MW. Sterlite Power is developing a 50 MW renewable power park near Chennai specifically for data center and GPU workloads.

India's government has made sovereign AI infrastructure a strategic priority through the IndiaAI Mission's compute pillar. The mission allocated Rs 10,372 crore ($1.25 billion) to build a common AI compute facility with 10,000 GPUs, deployed across multiple data center sites to ensure redundancy. The first tranche of 4,000 GPUs was awarded in 2025 to a consortium of Yotta and Netmagic, deployed at Yotta's Panvel and Netmagic's Bangalore facilities. The second tranche of 6,000 GPUs is under procurement as of mid-2026, with Reliance Jio, CtrlS, and Tata Communications bidding for the contract.

The government's GPU procurement includes a requirement that at least 30 percent of compute capacity be reserved for academic and startup use at subsidized rates of INR100-150 per GPU-hour, roughly 40-50 percent below market rates. This subsidized capacity has already been accessed by 200+ approved AI startups and 50+ academic institutions as of mid-2026. The remaining capacity is available to enterprises and government departments at market rates. The mission also includes provisions for GPU federated across providers via the India AI Cloud, enabling researchers to access H100 capacity across multiple provider data centers through a unified API, with the India Datasets Platform providing curated training data for Indian languages and contexts.