Opinion | The Cloud Has A Power Bill: The Terms India Should Set For Its Data Centre Boom

Published on:

India should welcome the wave of AI and data centre investment, but make clean power by the hour, efficiency, water stewardship, grid planning and fairness to host communities explicit conditions of that growth, says Prof. Sundeep K Nayak and Dr Kyawt Yin Min Thein.

Wires plugged into a data server
Opinion | The Cloud Has A Power Bill: The Terms India Should Set For Its Data Centre Boom

Ask a chatbot to summarise a court judgement, stream a cricket match or pay a vegetable vendor by UPI, and the request seems to vanish into a cloud. It does not. It lands in a windowless building in Navi Mumbai, Chennai or Hyderabad, where servers draw electricity around the clock and turn almost all of it into heat. Chillers carry the heat away; diesel generators wait for the grid to falter. The cloud is an industrial plant with a grid connection.

That is not a complaint. India is drawing one of the world's largest waves of digital-infrastructure investment. Global firms have announced multi-billion-dollar campuses, states offer incentives, and the Union Budget for 2026-27 proposed a tax holiday until 2047 for foreign companies that serve cloud customers worldwide from Indian data centres. Done well, it builds computing power for Indian languages, public services and businesses.

So the question is not whether India should host these facilities. It is on what terms. Clean electricity, efficiency, water, grid reliability and fairness to communities that carry the costs should be conditions of this growth, not afterthoughts.

Where The Capacity Is

Cushman & Wakefield counts 1,789 MW of operational data centre capacity across India's seven leading markets in the first half of 2026. Mumbai alone has 890 MW, about half the total. Chennai follows with 214 MW, then Delhi NCR (183 MW), Bengaluru (162 MW), Hyderabad (151 MW), Pune (133 MW) and Kolkata (22 MW). Another 509 MW is under construction, and 3,351 MW is planned for delivery by 2030. JLL, using its own definitions, puts today's capacity at 1.6 GW and expects 6 GW by 2029.

Read the small print. These are consultants' estimates, usually of IT capacity and the power available to servers. A building's total load is higher, and its annual electricity use depends on how full the halls are. A megawatt of capacity is not a megawatt-hour of consumption.

Announcements run further ahead still. In Visakhapatnam alone, more than 6 GW of hyperscale capacity is reported to be at various stages of execution, including Google's announced $15 billion, 1 GW AI hub. Announced is not built, but grid and water planning must begin long before the first server arrives.

Some states now reward performance as well as size. Uttar Pradesh's Data Centre Policy 2026, notified in July, offers capital and interest subsidies, land concessions and up to full relief on transmission and wheeling charges for renewable power. It also promotes 50 per cent renewable energy use in new data centres by 2030, with extra incentives for green certification. Every facility sits in a village or ward whose residents are stakeholders, not bystanders.

Power, Cooling And Water

A data centre never sleeps, and its heat must be removed continuously. The industry measures this overhead with Power Usage Effectiveness, or PUE: the building's total energy divided by the energy used by the computers. A PUE of 1.5 means every unit used by servers needs another half unit for cooling and other systems. The Uptime Institute's 2025 survey found a global average of 1.54, barely changed in six years; newer and larger sites averaged 1.44 to 1.48. In India's heat and humidity, cooling design matters more still.

A simple illustration, not a measurement, shows the scale. Suppose the 1,789 MW above were all IT capacity, running at 70 per cent on average, with a PUE of 1.5. Together these facilities would use about 16 terawatt-hours a year, a little under one per cent of the electricity on India's grid in 2025-26. At the Central Electricity Authority's average emission factor for that year, 0.675 tonnes of CO₂ per megawatt-hour, that comes to roughly 11 million tonnes of CO₂. Change the utilisation, PUE or power source and the answer shifts a great deal, which is why measured, published data matter more than any back-of-the-envelope figure, ours included.

The global picture is moving fast. The International Energy Agency estimates that data centres used about 415 TWh in 2024, roughly 1.5 per cent of the world's electricity. Its latest assessment finds their demand grew 17 per cent in 2025 and expects it to double by 2030.

Water needs care with words. Withdrawal is what a facility takes in; consumption is what does not come back, mostly evaporation from cooling towers. Air-cooled and closed-loop systems consume little water on site but tend to use more electricity; evaporative systems save electricity but use more water. Power stations use water too, elsewhere. In Visakhapatnam, Andhra Pradesh Chief Minister N Chandrababu Naidu has said the proposed facilities would need about 3.5 TMC (thousand million cubic feet) of water a year, and that the Polavaram project could bring the city far more. Whether that holds is for hydrologists and public discussion to settle, not assertion.

Then there is diesel. Almost every data centre installs backup generators. How often they run is not publicly reported, and we assume nothing. But fuel storage, testing and exhaust are real local concerns in cities that already breathe poor air.

Whose Grid, Whose Burden

Here is the arithmetic most debate skips. An average emission factor tells us how much carbon an average unit of Indian electricity carried over a year. It does not tell us what happens when a new 100 MW load switches on at nine in the evening. For that we need the marginal emission factor: the emissions of whichever power plant actually ramps up to meet the extra demand.

India's official data hint at the gap. The same CEA database reports an “operating margin” of 0.963 tonnes of CO₂ per megawatt-hour for 2025-26, which reflects the plants that typically adjust their output, against the 0.675 average. Its “build margin”, reflecting recently built capacity, was 0.446. Neither is a true hour-by-hour figure. Peer-reviewed modelling of 2017-18 data found coal likely to meet changes in demand more than half the time in more than half of India's states, with marginal emissions varying enormously between states and expected to vary more through the day as solar grows. So a data centre running round the clock is neither automatically as clean as the grid average nor automatically coal-fired. It depends on where it is, when it draws power and what contracts back that power.

Fairness matters too. A data centre's benefits, such as tax revenue, skilled jobs and digital services, gather in big cities. Some of the costs fall elsewhere: in districts where coal is mined and burnt, and along the corridors where transmission lines run. Planners should map who gains and who pays, using verifiable data. It cuts both ways: well-designed renewable contracts bring investment and lease income to solar and wind districts.

Rules Still Catching Up

In August, the Environment Ministry told the Rajya Sabha that AI data centres do not, in themselves, need environmental clearance under the EIA Notification, 2006. Clearance is needed only when they form part of a building project above 20,000 square metres of built-up area, or a township of 50 hectares or 1,50,000 square metres, and is granted by state authorities. Water discharge is monitored by State Pollution Control Boards, and groundwater use follows Ministry of Jal Shakti guidelines.

In practice, then, a very large computing campus is assessed much like a large building. In a pending challenge to a proposed 512 MW data centre near Visakhapatnam, the National Green Tribunal's Southern Zone bench has observed that the EIA Notification has no specific category for mega data centres, and that heat, e-waste and battery waste deserved attention. It has added the Environment Ministry as a party; its reply is due on 20 November, with a hearing on 7 December. These are interim remarks in a pending case, and we draw no conclusion about the project. Meanwhile the Draft Data Centre Policy of 2020 is still only a draft.

Electricity rules are further along: the Green Energy Open Access Rules, 2022 let consumers with 100 kW of demand buy renewable power directly, though states apply them unevenly. India's climate commitments set the frame. The 2030 target, set in 2022, is a 45 per cent cut in emissions intensity from 2005 levels and about half of installed power capacity from non-fossil sources. The 2035 target, approved by the Cabinet in March 2026, raises these to 47 per cent and 60 per cent. Net zero comes by 2070. Non-fossil sources already exceed half of installed capacity but a smaller share of electricity generated, which is why new round-the-clock demand matters.

At COP31 in Antalya, which is scheduled to close on 20 November, nothing has been decided yet. A first draft of the main decision, released on 16 November, contains contested language on fossil fuels. The same day, the Turkish presidency launched a voluntary Antalya AI pledge, with 80 early endorsements, mostly from companies and civil society. UN climate chief Simon Stiell set out the test plainly: “AI benefits must outweigh its growing costs.”

What Good Looks Like

Six measures would make a real difference.

Measure, don't promise. A design PUE is a promise; a PUE measured over twelve months is evidence. Sealing hot and cold aisles, running servers slightly warmer and liquid-cooling dense AI racks already deliver gains.

Match cooling to place. Use treated wastewater in cooling towers where it is available, as the Bureau of Energy Efficiency's District Cooling Guidelines recommend. Choose air-side or hybrid cooling where the climate allows, and avoid drawing on stressed groundwater.

Buy clean power that is real, not paper. Buying as many green units over a year as one consumes is a start, but it does not show clean power at nine at night. Matching every hour with solar, wind, storage and steady sources such as hydro is harder and costlier but closer to the truth. Contracts that add new capacity and storage count for more than certificates bought from existing plants.

Plan the grid early. Transmission lines and substations take longer to build than data halls; states and utilities should plan cluster connections and price them so other consumers do not quietly pick up the bill.

Reward flexibility. Some AI training and batch work can shift in time or place. Where it is technically safe, encouraging that shift can soak up midday solar power instead of adding to the evening peak.

Clean up backup and use the heat. Batteries for short outages and compliant generators cut local pollution. Waste heat is harder to use in the tropics than in Scandinavia, but absorption chilling and industrial uses are worth studying.

Each option has costs and limits. Hour-by-hour clean power costs more than an annual certificate, and batteries cannot yet replace days of diesel backup. These trade-offs should be argued with numbers, not adjectives.

A Fair Bargain

Four policy steps would turn principle into practice. First, sensible disclosure: annual electricity use, IT and total capacity, measured PUE, whether clean-power claims are annual or hourly, emissions, water withdrawn and consumed, and cooling method, with genuine commercial secrets protected. Second, location planning that coordinates grid connections, checks local water and rewards new clean supply. Third, a fairness test covering both the host site and where its power is produced, with real public consultation on large projects and benefit-sharing where burdens fall heavily. Fourth, incentives tied to verified performance on efficiency, water, flexibility and reporting, not to megawatts alone.

Someone also has to check the numbers. At IIT Goa, a proposed elective on energy transition would ask students to compare a baseline with low-carbon options on cost and emissions, test a key assumption, and weigh reliability and fairness, including the energy footprint of digital services. Regulators, utilities and companies will increasingly need that literacy.

India has a rare opportunity: it is building digital infrastructure while its grid is still growing and its rules are still being written. It can make the data centre a showcase of a just energy transition: efficient, transparent, clean by the hour and fair to the places that host it, rather than one more load the system must simply absorb. The cloud is coming down to earth. India can make sure it lands well.

About the Authors: Prof Sundeep K Nayak, IAS (Retd.), is Professor of Practice and Chair, Mission LiFE Cell, IIT Goa. Dr Kyawt Yin Min Thein is a former Researcher at CRDT, IIT Delhi, and Research Assistant, School of Planning and Architecture (SPA), New Delhi. Views are personal.

Disclaimer: The views and opinions expressed in this article are those of the author(s) and do not necessarily reflect the official policy or position of the publisher. While every effort has been made to ensure the accuracy of the information, the publisher is not responsible for any errors or omissions, or for the results obtained from the use of this information.

×