About 1.8 GW is operating today, and forecasts for around 2030 run from 5.6 to 12 GW. Even the lowest forecast is more than three times today's capacity. Not every announcement will be built.
Capacity today. Cushman & Wakefield counts 1.8 GW at hyperscale cloud and colocation sites at the end of H1 2026, in IT megawatts. Other counts differ: Axis Capital has 1.3 to 1.4 GW across all types and the government 1.57 GW. The gaps are unexplained.
Forecasts for around 2030.
- Wood Mackenzie: 12 GW (July 2026)
- Jefferies: about 10 GW over five years (September 2026)
- Government: nearly 8 GW (reported 3 November 2026)
- Crisil Ratings: 7.5 GW by March 2030 (3 November 2026)
- Cushman & Wakefield: 5.6 GW from the current pipeline, rising to about 8 GW in its AI acceleration case
One cautious view is below all five: an expert cited in an Axis Capital report, as reported by ANI, puts operating capacity at 2.8 to 3.6 GW by mid-2030. The report itself was not seen.
Money. More than US$170 billion has been announced since 2021 (CBRE has about US$173 billion, Cushman & Wakefield over US$180 billion), and about three quarters of it is still early stage. Hyperscalers took more than 80% of H1 2026 absorption.
Cost. Cushman & Wakefield puts a mid-spec India build at US$7.4 million a megawatt, in a range of US$5.9 to 9.0 million, construction only, plus or minus 15%. Turner & Townsend's 2026 index has Mumbai at US$6.83 a watt, the second-lowest of 62 markets, against US$9.67 in Bangkok, US$10.06 in Kuala Lumpur, US$11.80 in Northern Virginia and US$12.95 in London, for a typical 60 MW air-cooled build that excludes the site, utility works and IT equipment. Mumbai is the only Indian city in that index.
Who is building. Reliance and Adani have each announced around US$100 billion for AI and digital infrastructure. L&T's Vyoma.AI has an order of Rs 10,000 to 15,000 crore from Together AI, and Microsoft plans US$17.5 billion in India over 2026 to 2029. Beyond the large groups the work spreads across engineering, procurement and construction firms: power, transformers, electrical equipment, cooling, cabling, fit-out and base building. Of Goldman Sachs' 42 listed Indian AI enabler stocks, 30 had beaten the Nifty 50 from 31 December 2025 to 6 November 2026. That list is a Goldman theme and not a buy list.
What could slow it. A power sanction takes around 18 months (Axis Capital), backup generators are booked about two years ahead, specialist engineers in electrical, cooling and commissioning are short, and water, sites, policy and local support are open issues.
Analysis of public sources, for information only. Not investment advice. Full post with the deck and all 42 company tables: https://www.tigzig.com/post/india-data-centres-engineering-construction-power-nov2026. Related: is India becoming a global AI data-center hub? and how the hyperscalers are paying for the build.
Building something like this? How I work covers the rates, the availability and what I take on.