India's Data Center Boom Outruns Its Own Accounting

The Binding Constraint Is a Grid Node

Nationally, the footprint still looks modest. CEEW estimates data centers took about 0.5% of India’s electricity in 2024. The Ministry of Power expects associated demand to reach 13.56 GW by 2031-32. The Central Electricity Authority has since cited a figure closer to 17 GW. Even the higher number is manageable in aggregate.

Aggregates are the wrong frame. By September 2025, nearly 90% of India’s operational capacity sat in four cities, according to CBRE. Mumbai held 53%, Chennai 20%, Delhi-NCR 10% and Bengaluru 7%.

Rohit Kumar, chief operating officer at the data platform firm SCIKIQ, treats that concentration as the whole story. “The question is therefore about location rather than national scarcity,” he said. He points to Bengaluru, Visakhapatnam, Hyderabad and parts of the National Capital Region, where water or infrastructure limits deserve close scrutiny.

The arithmetic worsens as projects grow. A facility rated at 100 MW of IT load draws more than 100 MW once cooling, pumps and power conversion are included. Dikshu C. Kukreja, managing principal of CP Kukreja Architects and an urban planner, works the example at gigawatt scale. A 1 GW cluster at a power usage effectiveness of 1.3 needs about 1.3 GW of facility power. At 1.5, it needs 1.5 GW.

The illustration is crude, but the planning problem behind it is real. A large campus needs firm power continuously, at one node, not intermittent electricity averaged across a national grid.

Prasad Rai, chief executive officer of DAAS Labs, calls it a timing mismatch. A data center can be built in 12 to 18 months. Transmission infrastructure to serve a major new load takes three to five years. “The truth is that no Indian data center has yet been refused water or power,” Rai said. “The challenge is an institution’s capacity to build in time.”

He is right about the record.No such refusal appears in public reporting, although India maintains no national record against which Rai’s claim can be checked. The question is whether grid capacity in the largest clusters can expand fast enough when several campuses go online together.

The next wave of data centers will place much heavier demands on local grids. AI servers pack more computing capacity into each rack. They draw more electricity and generate more heat, increasing the load on both power and cooling systems. Wood Mackenzie’s July outlook projects Indian data center electricity use rising from about 10 TWh in 2025 to 191 TWh by 2040. That would equal roughly 7% of national consumption under its scenario. Globally, the International Energy Agency estimates that data centers used about 415 TWh in 2024 and expects consumption to approach 945 TWh by 2030.

India’s renewable fleet helps, but it does not solve the need for power around the clock. Solar generation falls away at night, while data centers keep running. CEEW’s industry consultations found that roughly 80% renewable supply may be technically feasible through a diversified mix of solar, wind and batteries. Matching consumption with carbon-free power every hour is much harder because of storage costs and transmission constraints. A facility that matches its annual consumption with renewable purchases may still draw conventional grid power at 3 a.m.

Tejas Kusurkar, co-founder and chief executive of Offgrid Energy Labs, argues that workloads should influence where data centers are built. Real-time services and inference need low latency and often belong near users. Large training runs are more flexible. “Training is the outlier,” he said. Those workloads can be placed closer to abundant renewable power, land and storage rather than following the existing metro clusters.

Cooling design determines much of the trade-off between water and electricity, and every option shifts the burden somewhere. Air cooling reduces direct water use but consumes more electricity in hot weather. Evaporative cooling uses water to reduce energy demand. Direct-to-chip and immersion systems can manage dense AI racks more efficiently but require specialized equipment. Desalination can reduce dependence on freshwater along the coast, but it consumes power and produces brine that must be managed.

The useful question, then, is not only how many liters a building uses. Kukreja frames the assessment around three issues. Where does the water come from: potable supplies, groundwater, recycled water or desalination? How much is consumed rather than merely withdrawn? What happens during a drought? “A facility using a closed-loop system with treated wastewater has a fundamentally different resource footprint from one dependent on fresh municipal or groundwater supplies,” he said.

Those questions matter most in Bengaluru, Chennai and parts of Hyderabad and the National Capital Region, where new campuses must compete with the needs of large residential and industrial populations.

India’s Data Center Approvals Miss Three Critical Tests

Visakhapatnam is where the abstraction ends.

Google is building a 1 GW AI hub there comprising three campuses and backed by a $15 billion investment between 2026 and 2030. AdaniConneX and Nxtra by Airtel are its partners. Google broke ground on April 28.

In August, Reuters reported court challenges and local objections over water availability and the proximity of one site to the Kambalakonda Wildlife Sanctuary, 860 meters away. Visakhapatnam receives about 410 million liters of water a day against demand of 480 million. The Andhra Pradesh High Court asked the state government to respond to allegations that the project could strain water supplies and scheduled the next hearing for August 24. Google says it will use advanced air cooling to protect local water resources. The state says no residential or rural water supplies will be diverted.

None of this proves that the project has caused a shortage. It is not yet operating. It shows what happens when questions about water and infrastructure emerge after a site has been selected and construction has begun.

Kukreja calls the result infrastructure lock-in. Once a campus has land, fiber, substations, transmission connections and cooling equipment, moving it is no longer an economically realistic option. Approval, rather than operation, is therefore the decisive point.

A serious approval process would apply three tests. The node test would ask whether the local grid can supply firm power before the campus opens, taking account of every other project awaiting connection. The basin test would ask whether the watershed can meet cooling demand in a dry year and whether the water will be potable, groundwater, recycled or desalinated. The ledger test would establish what the operator must publish, in what format and how often, so that its power and water claims can later be checked.

Call it the three-gate test. In Visakhapatnam, Google and the state have provided only a partial answer to the basin question. Neither has published a facility-level water commitment or a schedule for reporting actual consumption.

India is closer to creating the ledger than it might appear. In February, the government notified several BIS standards aligned with the ISO/IEC 30134 framework. They cover measures including water usage effectiveness, carbon usage effectiveness, cooling efficiency ratio and energy reuse factor. The methods are internationally comparable, but compliance remains voluntary unless the government issues a quality control order.

Sumeysh Srivastava, partner at The Quantum Hub, describes the underlying pattern. India has rules covering energy efficiency, groundwater, renewable procurement and other parts of data center development, he said. What it lacks is a consistent way to determine whether a particular location can support a project of a given size. “The pressure will not be evenly spread,” Srivastava said.

Kumar makes the same point from the other direction. Regulation is not absent. Building codes, energy standards, groundwater rules and renewable energy regulations all apply. The weakness is that these requirements have never been assembled into a single resource test for a hyperscale campus.

The incentives, meanwhile, are clear. India has granted infrastructure status to qualifying data centers, while states offer a range of concessions. The Finance Act, notified on March 30, 2026, created a tax exemption through March 2047 for notified foreign companies procuring services from specified Indian data centers. It also provided a 15% cost plus safe harbor for an Indian data center operator serving a related foreign cloud company. Income Tax Department

A draft National Data Centre Policy circulated in 2020 was never notified. CEEW reviewed 15 state policies and found that only five explicitly included sustainability provisions.

Other markets already make grid access conditional. Ireland’s regulator decided in December 2025 that new data centers must provide generation or storage capacity matching their requested maximum import demand. They must also meet at least 80% of annual demand through additional renewable projects in Ireland and pass a location-specific grid assessment. Ireland’s Commission for Regulation of Utilities

Singapore allocates new capacity through competitive applications. Its December 2025 round offered at least 200 MW. Applicants must achieve a PUE of 1.25 or better at full IT load and source at least half their power through eligible green-energy pathways.

Neither model can simply be transplanted to India. Kusurkar notes that India has a different grid and renewable-energy base. Pairing flexible AI workloads with renewable generation and long-duration storage may achieve more than importing restrictions designed for another market, he argues. The transferable lesson is narrower. When demand outruns supporting infrastructure, governments become selective about which projects can connect and on what terms.

Implications by Role

C-suite. Give the grid node and watershed the same weight in site selection as land prices and fiber connectivity. A campus built in a constrained node or stressed basin carries regulatory and reputational risks that later efficiency improvements cannot fully erase. Operators planning Indian campuses through 2030 should complete node-level and basin-level assessments for every site, not only contested projects such as Visakhapatnam.

Functional leaders. Adopt the BIS metrics before they become mandatory. Report water withdrawal and consumption separately, identify the source and disclose the cooling technology. Facilities and energy teams should model both a dry-year water scenario and hourly carbon-free power availability. Annual renewable procurement figures can conceal the hours when a facility depends on carbon-intensive grid electricity.

Boards and governance. Where public incentives have been claimed, ask what was promised in return and whether the result can be measured. Separate employment projections into construction jobs, permanent operational staff, contractors and wider supply-chain roles before including them in an annual report. Boards should also consider what happens to expansion plans and asset values if future grid connections become conditional on resource performance.

India’s data center capacity has roughly quadrupled in five years and could quadruple again by 2030. Yet the country still cannot show which grid nodes and watersheds can absorb the next round of projects. Nor can it produce consistent national figures for facility-level water use or permanent employment.

India adopted useful performance metrics in February. It has not attached them to the most consequential and difficult decision to reverse: where a campus is approved before the concrete is poured.