India’s data-centre boom is often described through the language of investment announcements: gigawatts of planned capacity, billions of dollars committed, new campuses and incentives offered by competing states. But the economics of this industry are more demanding than the headline numbers suggest. A data centre is a capital-intensive asset that must keep expensive equipment running around the clock, secure reliable electricity, cool increasingly dense computing systems and persuade customers to commit to capacity over long contracts.
- A ₹4 lakh crore build-out—and a much larger customer bill
- Power is the operating-cost variable that can change the model
- AI changes the cost per megawatt
- Utilisation is the quiet determinant of returns
- India has a cost advantage—but it is not guaranteed
- Who captures the value?
- The investment question India needs to ask
That is why the next phase of India’s AI infrastructure story is not simply about how much capacity gets announced. It is about who can finance it, keep it utilised, procure power at a competitive price and earn an adequate return after accounting for the full cost of operating the facility.
A ₹4 lakh crore build-out—and a much larger customer bill
CareEdge Ratings estimates that India’s data-centre capacity could rise from around 1.5 GW in FY26 to about 8 GW by FY31, requiring nearly ₹4 lakh crore of investment by colocation operators over five years. The same analysis indicates that client-side spending could be substantially larger, particularly for AI workloads, where the cost of compute equipment adds significantly to the facility bill. These are industry estimates, not committed expenditure or a guaranteed investment outcome. (CareEdge estimates and the infrastructure cost squeeze.)
The distinction between the data-centre shell and the computing equipment inside it matters. Operators build or lease the facility, power systems, cooling, security and network infrastructure. Cloud providers, AI companies and enterprise tenants may separately fund servers, accelerators, storage and software. The total capital committed to an AI data-centre ecosystem is therefore not the same as the capital invested by the facility operator.
For investors, this creates two linked but different businesses: the infrastructure platform that sells capacity and the compute business that uses that capacity to sell cloud or AI services. Their risk, margins and asset cycles are not identical.
Power is the operating-cost variable that can change the model
Electricity is not a marginal input for a data centre. It is a continuous operating requirement. CareEdge’s analysis puts power at roughly 70–75 per cent of operating expenses for Indian data centres. The exact share varies with facility design, utilisation, tariff structure, cooling technology and whether costs are passed through to customers. But the broad point is clear: a small difference in delivered electricity cost can have a material impact on operating economics.
That makes the 26.3 GW additional AI data-centre load projected by the Ministry of Power by FY2031–32 more than a capacity story. It is a test of whether generation, transmission and distribution can deliver reliable power where facilities are being built. As we examined in India’s AI data-centre power problem, around 17 GW of the projected demand had entered the grid-connectivity application process in the Ministry’s August clarification. That is not the same as approved or operational capacity.
Developers are responding by looking beyond a conventional utility connection. Renewable power procurement, open access, captive generation, battery storage and long-term energy contracts can help manage price and sustainability exposure. But each introduces its own complexity: intermittency, transmission charges, banking rules, storage economics, regulatory variation and the need for firm backup.
AI changes the cost per megawatt
AI workloads are changing what a megawatt of data-centre capacity actually costs to build and operate. Higher rack densities require more advanced power distribution, liquid cooling in some configurations, stronger networking and more sophisticated thermal management. CareEdge’s analysis notes that the cost of setting up a data centre has risen to more than ₹80 crore per MW from historical estimates of ₹40–50 crore per MW, reflecting higher infrastructure requirements.
That does not mean every facility has the same cost profile. A colocation building designed for conventional enterprise workloads is not directly comparable with a high-density AI facility. Nor should a cost-per-megawatt figure be confused with the cost of the GPUs and other IT equipment installed inside the building.
The distinction is important because AI capacity can command higher pricing, but only if demand materialises and customers are willing to pay for it. An operator that builds ahead of demand may carry expensive assets with low utilisation. An operator that waits too long may lose customers to a competitor with ready capacity and power access.
Utilisation is the quiet determinant of returns
Data centres are often valued for long contracts and recurring revenue. Those characteristics can make the business attractive to infrastructure investors. But the quality of the return depends on how quickly a facility fills, the price at which capacity is contracted, the cost of financing and the amount of capital required to maintain or upgrade the asset.
A building with substantial available power and server space is not automatically a productive asset. The operator must secure tenants, deliver the required service levels and manage the gap between construction expenditure and revenue generation. AI adds another uncertainty: technology generations move quickly. Power and cooling infrastructure may have a long useful life, while servers and accelerators can become commercially outdated much faster.
That creates a potential mismatch between the life of the building, the life of the electrical and cooling systems, and the replacement cycle of compute hardware. Investors need to understand who bears the cost of hardware refreshes, whether the customer or operator owns the equipment, and how contracts handle changes in rack density and power requirements.
India has a cost advantage—but it is not guaranteed
India’s relative cost position is one reason global operators and cloud providers are looking at the market. Mint, citing Nomura research, reported that data-centre construction costs in India were estimated to be 30–50 per cent lower than in developed markets across Asia, Europe and the United States. The comparison excludes land and computing equipment, and the advantage varies by location and project specification. (The reported India cost advantage and its limits.)
Lower construction costs do not automatically translate into lower total cost of ownership. Power tariffs, grid-connection delays, land prices, water availability, financing costs, imported equipment, taxes and utilisation all influence the economics. A cheaper site that cannot secure reliable power on time may be more expensive than a premium location with established infrastructure.
The market is also geographically concentrated. Mumbai, Navi Mumbai, Chennai, Hyderabad, Bengaluru, Delhi-NCR and Jamnagar have emerged as important hubs. As capacity expands, operators may consider newer locations where land and incentives are attractive. But the site must also have fibre connectivity, transmission access, water planning, skilled operations teams and a credible customer pipeline.
Who captures the value?
The AI data-centre value chain includes landowners, utilities, transmission companies, renewable developers, equipment suppliers, cooling specialists, colocation operators, cloud providers and the businesses that consume AI services. Their economics differ. A utility may earn regulated returns on network investment. A colocation operator may earn recurring rent and service revenue. A cloud provider may capture higher-value compute and software revenue—but also carry the risk of hardware depreciation and demand volatility.
This is why the headline value of the AI infrastructure market should not be mistaken for the profit pool available to every participant. The strongest position may belong to companies that control a scarce input: reliable power, ready-to-use capacity, high-quality connectivity, specialised cooling or customer relationships that support sustained utilisation.
Our earlier analysis of optical connectivity and the next AI infrastructure battle explored how data movement is becoming a strategic layer around compute. Power and cooling are equally consequential constraints. The infrastructure stack is becoming more interdependent, and the economics of one layer can affect returns across the others.
The investment question India needs to ask
India should welcome investment in data centres, but the policy conversation must move beyond the size of announced commitments. The more useful questions are whether projects are financed, whether grid connections are secured, whether customers have signed contracts, whether capacity is commissioned on schedule and whether the facility can operate efficiently over its life.
For investors, the checklist should include contracted versus speculative capacity, customer concentration, power procurement terms, debt levels, utilisation ramp-up, capex per MW, hardware ownership, cooling design, water exposure and the cost of future upgrades. For state governments, incentives should be assessed against measurable outcomes—not simply the value of investment proposals.
India’s data-centre opportunity is large, but it is not a guaranteed infrastructure windfall. The winners will not necessarily be the companies announcing the biggest campuses. They will be the ones that convert capital into dependable capacity, secure customers at sustainable prices and manage the cost of power, cooling and technology refreshes.
The AI boom may create demand for more data centres. The harder business is making those data centres earn their keep.
FutureIsNow | Signals over noise. Intelligence for what comes next.
