India’s AI expansion will depend on where data centres are built next

For most of the last decade, India’s data centre map was a story of two cities, Mumbai and Chennai. Mumbai benefitted from cable landing stations and strong demand from the financial sector, while Chennai combined its own connectivity advantages with an early concentration of global offshoring and IT services. Between them, they accounted for the bulk of the country’s capacity, and for good reason. Connectivity and customer proximity largely determined where operators chose to build.

That logic still holds, but it is no longer sufficient. The workload mix in Indian data centres is changing faster than at any point in the market’s history, and AI is the driving force behind that shift.

An AI-ready data centre is not simply a cloud facility with a different label. Floor loading requirements that once sat comfortably at 12-15 kilonewtons per sq. m now increasingly need to support 25 kilonewtons and beyond, as GPU infrastructure is significantly denser and heavier than the server racks it is replacing. Power consumption per rack has risen accordingly, while cooling requirements have evolved just as quickly. Liquid cooling, once discussed primarily as a future possibility, has now become a core design consideration for AI-ready facilities.

What this means for site selection is straightforward, even if the implications are not always openly discussed. A location that worked well for a cloud-era data centre, where land availability and connectivity were the primary considerations, will not automatically work for an AI-era facility. Increasingly, the deciding factor is power, particularly the amount available, the reliability of supply, and the speed at which capacity can be expanded as customer requirements grow.

Structural constraint

This is where India’s established hubs encounter a structural constraint. India is not short of power generation capacity. The challenge is ensuring that power can be delivered to the specific locations where operators want to build. In parts of Mumbai, utilities can deliver power directly to a site’s boundary. In some Chennai locations, operators may still need to invest in dedicated connections to nearby substations, an expense that can rival the cost of the land itself. Those differences can materially alter the economics of a project long before a single server is switched on.

The natural consequence is that growth is expanding outward from the original hubs rather than moving away from them altogether. Around Mumbai, that increasingly means looking towards the broader Mumbai-Pune corridor, where industrial zones already possess the zoning frameworks, transport infrastructure, and the grid capacity needed to support large-scale development.

The pattern is consistent with how the industry has evolved in India over the past decade. Data centres tend to cluster in designated industrial areas rather than residential zones, following the path created by power infrastructure, connectivity, and long-term government planning.

Government support for the sector has been genuine. Central ministries have consistently encouraged investment, while individual states have become increasingly active in competing for data centre projects. However, policy intent at the top does not always translate into immediate execution on the ground. While governments have made it easier to secure approvals for new projects, many designated data centre campuses still require additional investment in power distribution, fibre connectivity, and supporting infrastructure before they become genuinely deployment-ready. The gap between announcement and execution remains one of the most significant variables determining how quickly emerging locations can absorb demand.

The timing is particularly important when viewed in a global context. The U.S. is grappling with growing power constraints and increasing local resistance to new data centre developments. Other markets face their own grid limitations as well. As a result, a growing share of global demand is looking towards regions such as India, Japan, and parts of Southeast Asia.

The India opportunity

When hyperscalers decide where to deploy capacity, they evaluate regions against one another. India’s combination of available land, expanding renewable energy capacity, and improving policy support places it in a genuinely competitive position. To capitalise on that opportunity, however, infrastructure gaps in emerging locations must be addressed over the next three to five years. Otherwise, a significant portion of that demand could be captured by competing markets.

India’s AI infrastructure story over the next five years will be determined less by who builds the most capacity and more by who builds in locations where power, connectivity, and execution align. In that environment, success will depend not on finding the right piece of land, but on finding the right ecosystem to support data centre development.

(The author is President and Country Head India, Colt Data Centre Services. Views are personal.)