India's Data Centre Capacity Boom
Why India's server farms are expanding faster than almost any other piece of its infrastructure right now
Think of a data centre's size the way you'd think of a power plant's, measured not in square feet but in megawatts, MW, of electricity it's built to consume, because servers are fundamentally electricity-hungry machines and a facility's true capacity is really a statement about how much power it can draw and cool reliably, not just how much floor space it occupies.
By that measure, India's data centre industry is expanding at a genuinely striking pace, total capacity reached roughly 1,700 MW in 2025 on the back of a record 440 MW of new supply added that year alone, with the industry projecting a further climb to roughly 2,073 MW by the end of 2027, an 85% increase from current levels, and CBRE forecasting around 30% year-on-year growth specifically in 2026.
Longer-range forecasts get considerably more dramatic still, industry consensus points to 4 to 5 GW of capacity by 2030 in base-case scenarios, and 8 to 9.2 GW in AI-accelerated scenarios where artificial intelligence workloads specifically drive demand well beyond traditional cloud and enterprise computing needs, a genuinely wide range reflecting just how uncertain AI's actual compute appetite remains even to industry forecasters themselves.
This capacity expansion isn't evenly spread across the country, Mumbai and Chennai together account for nearly 70% of India's colocation data centre capacity, Mumbai-MMR alone holding 47%, a geographic concentration reflecting where power availability, submarine cable connectivity and existing digital infrastructure already cluster, exactly the kind of infrastructure advantage that's hard for other cities to quickly replicate.
Related concepts
IndiaAI Mission
The government programme renting out supercomputer access to Indian startups for less than the price of a cup of coffee per hour
India's Hyperscale Data Centre Megaprojects
Why Reliance, Adani and global players are suddenly committing tens of billions of dollars to Indian server farms
GPU-as-a-Service
Why AI compute is increasingly rented by the hour rather than bought outright, and what that means for who can build AI in India