New Delhi, July 22 -- India's developers have cumulatively announced close to 15 GW of data centre capacity for 2030. We estimate that around 9.6 GW will actually get built. Even this figure, roughly six times today's 1.6 GW, is remarkable and among the largest infrastructure buildouts of the decade. But more telling is the gap between the two.
Announcements have run well ahead of what the ground can support, and closing that gap has little to do with demand or capital, both of which are abundant here. It comes down to execution. An announcement is a statement of intent and not a measure of capacity. Turning one into a live, revenue-earning facility means securing land, power, water, fibre and equipment, building over several years, and then operating the site reliably. This is where the market is won or lost.
The scale is easy to underestimate. India has been adding around 200 MW a year, yet reaching 9.6 GW by 2030 means adding close to 1.6 GW annually, roughly the size of the country's entire operational fleet today. This, we estimate, at a build cost of Rs.50-60 crore per megawatt. The ability to do this repeatedly, at scale, is what separates a business from a pipeline.
Grid connection is the first and hardest hurdle. The constraint is largely physical rather than procedural, since the transmission lines that deliver power take years to build. This is a global problem, and India is better placed than most. A new 100 MW connection takes about two years here, against four or more in Australia, China and Indonesia, up to six in South Korea and close to a decade in Japan. Even the biggest markets struggle. Connections in Northern Virginia can take up to seven years, Dublin now approves new sites only if they bring their own power, and Singapore reopened only to projects with a power usage effectiveness (PUE) below 1.25. The lesson is the same everywhere: power, not land, decides who goes live first.
Power is where the economics are won as well. The cost of electricity is a decisive competitive lever, since energy makes up close to half of a facility's running costs and mostly flows through to the customer. Two things move that cost: the tariff itself, and how much of each unit reaches the servers rather than the cooling. An operator on group-captive renewable power can run 30 to 50% below grid tariffs once cross-subsidy and wheeling charges are waived, and a more efficient site, with a lower PUE, supports a higher IT load from the same connection. That is why the biggest builds increasingly source power directly through open access, captive plants, and behind-the-meter generation rather than the state distribution utility.
The same economics are redrawing India's data centre map. Data centres use approximately 0.5% of the country's electricity today and are on course to reach 3% by 2030, concentrated in a handful of cities. Mumbai and Navi Mumbai lead the country on data centre capacity, but their share is set to slip from more than 40% today toward 30% by 2030, as growth moves to wherever electricity supply is firm and affordable. The fastest-rising sites may be greenfield locations chosen for power rather than proximity to users, among them Jamnagar in Gujarat and Visakhapatnam in Andhra Pradesh, where gigawatt-scale campuses are being paired with dedicated captive renewable supply. Connectivity is maturing, but it still sets the strong platforms apart. Subsea capacity remains concentrated: around 18 international cables serve India, and more than two-thirds of that capacity comes ashore at just two cities, Mumbai and Chennai. That is starting to change, with around 6 new landing stations planned across 4 coastal states, spreading the points of entry and easing the concentration risk, and access to a cable landing station is therefore a real differentiator. On land, the edge comes from diverse routes, low latency, and fast repair. A new set of neutral-host providers now offers point-to-point, high-bandwidth, low-latency links as a managed service, giving operators resilience without building the network themselves. Connectivity is no longer the moat it once was, but done well it still wins business.
AI is where strategy and execution meet. Estimates suggest that a data centre built for AI must deliver power and cooling on a completely different scale (40 or more kW per rack) from one running ordinary workloads (6-12 kW per rack). Air cooling is approaching its limits at higher rack densities, so liquid cooling is increasingly adopted for AI workloads. The capex shifts as specifications shift from conventional Tier III builds (from Rs.50-55 crore per MW) to AI-ready designs with immersion cooling or Tier IV redundancy (to up to Rs.75 crore per MW).
So, the question is how far to go in on AI is a design choice made before construction starts? Aim too high, and capital sits idle. Aim too low, and the site cannot take AI tenants without an expensive rebuild. Either way, the design sets the bar. Delivering to it, on time and on budget, is the execution test that follows.
None of this risk is hypothetical. Data centre projects in India routinely slip 6 to 18 months, but disciplined project management is what turns a likely delay into a delivered asset. The demand itself is real and huge, with roughly Rs.7-8 lakh crore of data centre investment announced for India, alongside spending on power infrastructure and AI compute. Everyone can see that demand, so the winning edge lies elsewhere: in execution, the discipline to secure power, build efficiently, design for the next wave of hardware, and do it again and again across a portfolio.
The platforms that manage that, and the investors who back them, will turn India's announced gigawatts into capacity that is actually powered, connected, and ready for AI.
Published by HT Digital Content Services with permission from TechCircle.