Every technology boom has a layer beneath the applications that grab the headlines: the compute. In artificial intelligence that means graphics processors, the data centres that house them, the power and networking that feed them, and the cloud software that rents them out by the hour. Through 2025 and into 2026, a striking share of the capital flowing into Indian AI has gone not to consumer-facing models or apps but to this foundational plumbing. The picks-and-shovels bet has become one of the largest in the country's technology economy.
The capital is real, and it is large
The clearest signal is in data centres. According to a CBRE report published in April 2026, investment into Indian data centres reached $56.4 billion in 2025, lifting cumulative commitments to roughly $126 billion, a figure CBRE expects to exceed $180 billion in 2026. Operational capacity crossed 1,700 MW during 2025 after a record 440 MW of new supply was added, and CBRE projects capacity will grow another 30% in 2026. AI-driven workloads are named as a central reason for that demand.
Venture and private capital are moving in parallel. In February 2026, a Blackstone-led group committed up to $1.2 billion to Neysa Networks, an AI-cloud startup founded only in 2023. The commitment is split evenly between $600 million of equity and $600 million of debt, with Teachers' Venture Growth, TVS Capital, 360 ONE and Nexus Venture Partners also participating. The money is earmarked to deploy more than 20,000 GPUs, up from roughly 1,200 the company runs today. Yotta, one of the domestic clouds empanelled under the government scheme, is among several racing to stand up tens of thousands of accelerators of their own.
Sovereign compute, with the state as anchor customer
Government policy is amplifying the private bet. The Union Cabinet approved the IndiaAI Mission in 2024 with an outlay of ₹10,371 crore over five years, and building shared compute is a core pillar. The first empanelment round, concluded in early 2025, secured 14,517 GPUs across ten providers including Reliance Jio, Yotta, E2E and NxtGen, surpassing the original 10,000-GPU target. Subsidised pricing landed near ₹115.85 per GPU-hour for lower-tier units, with the state absorbing part of the cost so that researchers, startups and academics can reach hardware they could not otherwise afford. A second round drew bids from Google, Infosys, TCS, Oracle, Airtel and Ola Krutrim, among others.
Still early, and still small
The enthusiasm should be read against scale. Indian AI startups raised $676 million across 57 deals in the first half of 2026, more than four times the same period a year earlier, and Sarvam AI became the country's second AI unicorn on a $234 million round. Yet that entire half-year total is dwarfed by single foreign rounds, with OpenAI and Anthropic each raising tens of billions of dollars in the same window. Blackstone estimates India has fewer than 60,000 GPUs deployed today, a base it expects to expand many times over. The infrastructure story is real and accelerating, but it is starting from a low base.
The logic behind the bet is straightforward. Infrastructure is a wager on the whole sector rather than any single winner: whoever builds the best Indian models or applications will still need somewhere to train and serve them. It is also a question of sovereignty, since depending entirely on foreign clouds and imported chips for a strategic capability makes policymakers uneasy. The risks are equally clear: extreme capital intensity, reliance on imported hardware, heavy geographic concentration in a handful of cities, and the danger of building capacity ahead of paying demand. For now, investors and the government alike are placing some of India's most consequential AI bets not on the models but on the ground they stand on.