AI data centers in India

A pipeline, not yet a fleet

Our census counts 51 AI data center campuses in India, and 40 of them publish a capacity. Together those figures add up to 6.4 GW, which is 7.2% of everything the census counts worldwide. The headline hides the timing. Operational campuses account for only 573 MW; the rest is 2.1 GW under construction and 3.7 GW announced.

That split matches what industry research describes. Wood Mackenzie, in research reported in July 2026, put AI-dedicated data center capacity in India at 275 MW in 2025 and expects it to reach 6,546 MW by 2030, against total operational capacity rising from 2.2 GW to 12 GW over the same years. Our figures are not an estimate of that market: they are the sum of named projects whose own sources publish a number, so they undercount wherever a developer has not disclosed megawatts, and they count announced projects that may never be built.

Three gigawatt announcements carry the total

The census holds 3 Indian campuses rated at a gigawatt or more, and none of them is running yet. The largest is Colt DCS / RMZ Visakhapatnam Digital Campus, a project of Colt Data Centre Services / RMZ in Visakhapatnam (Vizag) rated at 1,250 MW. AdaniConneX’s Navi Mumbai campus and the HyperVault Future City campus in Hyderabad, backed by TCS, complete the group.

Projects of that size are statements of intent. Each depends on transmission lines, substations and, for liquid-cooled AI halls, water and cooling plant that take years to deliver. A buyer planning GPU capacity in India for the next twelve months should look past the top of Table 2: the operational and nearly finished mid-sized campuses further down the list are where capacity can actually be contracted today.

Developers rather than hyperscalers

India’s list is led by developers that lease space and power to tenants, not by the cloud companies themselves. AdaniConneX leads with 4 campuses and 2.0 GW of published capacity. Princeton Digital Group, AirTrunk, ST Telemedia, CtrlS and Yotta follow, each with several campuses. In all, 25 named operators or developers appear.

The hyperscalers are present, but mostly without a published figure. Microsoft’s Hyderabad region and AWS’s campus in Bharat Future City are in the census with no megawatts in their sources, and Google’s AI hub in Visakhapatnam is listed as announced, also without a number. Meta’s capacity appears through a built-to-suit campus that Reliance Industries is building in Jamnagar. For anyone counting India’s AI capacity, those gaps matter: the largest buyers of GPUs are the least likely to publish what they are building.

Hyderabad, Navi Mumbai and Chennai

Three metros hold most of the capacity. Hyderabad leads with 7 campuses and 1.9 GW, driven by two very large announcements and a cluster of builds by AdaniConneX, AirTrunk and Princeton Digital Group. Navi Mumbai and Mumbai together form the densest cluster by count. Chennai has the most campuses of any single city in the list, most of them mid-sized builds by colocation developers.

The newer names point to where growth is going next. Visakhapatnam appears through two large projects, and Raipur and Indore through RackBank’s liquid-cooled builds. Both sit outside the established hubs, which is where the next round of capacity is being announced. Across the country, 16 campuses record liquid cooling, the clearest sign that a site is being built for dense GPU racks rather than for conventional enterprise servers.

Reading this list

Three cautions apply. First, “announced” means a developer has said it will build, usually with a capacity and a city; it does not mean land, power and permits are secured. Second, megawatts are as published, and some sources quote IT load while others quote grid connection, so two campuses of the same headline size can differ in usable capacity. Third, 11 Indian campuses publish no capacity at all and add nothing to the totals, so the real figure is higher than the sum above.

For the full picture of data centers in India, including conventional enterprise and colocation facilities that do not meet our AI definition, see the megawattlas.com country page for India. This page counts only campuses that carry an AI-ready flag in their records or are pipeline projects of at least 50 MW.