The Server Needs Water
India is building AI infrastructure on the same physical foundation it never finished fixing.
On the morning of 19 June 2019, Malar woke up and turned on her kitchen tap.
Nothing came out.
She was not surprised. The taps in her Velachery neighbourhood in Chennai had been running dry for weeks. She picked up her two-litre plastic containers and walked to the street corner where a water tanker was expected — sometime in the morning, nobody knew exactly when. She joined a queue that stretched around the block. Women in salwar kameez, men in lungis, a few office workers clutching their ID cards who would miss their morning train. Everyone holding containers.
Chennai that summer had reached what planners call "Day Zero." All four of the city's main reservoirs — Poondi, Chembarambakkam, Cholavaram, Red Hills — had run completely dry. The 11.2 million people of one of India's largest cities had no piped water supply. Families in slums received as little as 30 litres per day per household. Restaurants shut down. Hospitals scrambled. Hotels charged a premium for rooms that still had running water.
The city had two failed monsoons behind it. But it also had decades of mismanaged groundwater extraction, encroached lake beds, and a water infrastructure that had not kept pace with a city that tripled in size.
Malar filled her containers. She walked home. She did this every day for months.
Five years later, the Tamil Nadu government announced something historic.
Between 2024 and 2026, the state signed MoUs worth more than ₹18,000 crore to build data centres in and around Chennai. The Adani Group committed ₹2,500 crore. A firm called Henox announced Project Horizon — ₹50,000 crore in data centres, cable landing stations, and solar projects across the state. Chennai's data centre capacity, which stood at 202 MW in 2025, is on track to reach 551 MW by 2030, growing at 22 percent a year.
The government offered generous incentives to attract this investment: stamp duty exemptions, electricity duty waivers, capital subsidies of 25 to 50 percent on land cost, concessional power tariffs, single-window clearances.
What the policy did not include: any cap on water usage.
THE RUSH
India is in the middle of a data centre construction boom that has few parallels in its economic history.
Nationally, data centre capacity stood at roughly 1.2 gigawatts in 2025 and is projected to reach 10 GW by 2030. Cumulative investment commitments from global and domestic players crossed $126 billion by the end of 2025. Microsoft, Amazon, and Google have together pledged tens of billions of dollars to build AI infrastructure in the country. The Union Budget 2026–27 offered a twenty-year tax holiday to foreign cloud providers serving global customers from Indian soil.
India now holds the second-largest internet user base in the world. Its AI ambitions are explicit government policy. And Chennai, along with Mumbai, accounts for 70 percent of India's total data centre absorption. Chennai specifically holds 21 percent of India's installed data centre capacity — the second-highest of any Indian city.
The logic is straightforward: India needs AI capability. AI capability requires computing power. Computing power requires data centres. Tamil Nadu has good connectivity, land, a skilled workforce, and a government that wants the investment.
All of this is true. The problem is what the calculation leaves out.
WHAT A DATA CENTRE ACTUALLY CONSUMES
A modern 100-megawatt data centre — a mid-sized facility — consumes two things in staggering quantities.
Electricity. A 100 MW facility runs continuously at near-full load. That is roughly the electricity consumption of 100,000 Indian homes, or the equivalent of a large aluminium smelter. It needs uninterrupted supply, 24 hours a day, 365 days a year, with near-zero tolerance for fluctuations. By 2030, India's data centre sector will consume 40–45 terawatt-hours of electricity annually — up from 10–15 TWh today. Tamil Nadu already faces unserved energy demand approaching 20 TWh, which is 10 percent of its total electricity requirement. The Central Electricity Authority has only recently begun directing states to factor data centre demand into adequacy planning. For years, this demand was invisible in grid forecasts.
Water. A 100 MW data centre consumes approximately 20 lakh litres of water per day — two million litres — to cool its servers through evaporative cooling towers. This is not grey water. It is clean, treated, potable-grade water, drawn from the same supply that residents drink, bathe in, and cook with. Nationally, India's data centres consumed roughly 150 billion litres of water in 2025. By 2030, that figure is projected to more than double — to 358 billion litres.
More than half of India's data centres are located in water-stressed regions. An S&P Global study forecasts that 60 to 80 percent of India's data centres could face high water stress within this decade.
In Chennai, the facility is being built in a city that had no drinking water five years ago.
THE ELECTRICITY PROBLEM HAS A NAME
India's power grid is not built for the kind of demand that data centres impose.
Unlike factories or commercial buildings, which have gradual load curves and predictable patterns, large data centre campuses can draw and release gigawatts of power rapidly. The IEEFA has flagged that sudden withdrawal of 1 to 2 GW from the grid — during maintenance windows, emergency shutdowns, or load-shedding events — could destabilise grid frequency and cascade into regional outages.
Tamil Nadu has invested significantly in renewable energy — it has one of the largest solar and wind portfolios in India. But renewables are intermittent. A data centre cannot run on intermittent power. The gap between renewable availability and data centre demand is filled by the grid — which means, in practice, by coal.
The state's generous electricity duty waivers for data centres mean that this infrastructure is being subsidised by Tamil Nadu's electricity consumers — including farmers on PM-KUSUM, small businesses on commercial tariffs, and households already paying some of the highest domestic electricity rates in South India.
The subsidy is real. The beneficiary is a global technology corporation. The cost is distributed across ten crore people in Tamil Nadu.
WHAT INDIA GAINS
This is not a simple story of harm. The data centre boom brings genuine and significant benefits.
Jobs. The sector is expected to create one lakh engineering jobs by 2030. A 100 MW campus can employ 1,000 to 2,000 people during construction, and hundreds more in operations. These are high-skilled, well-paying jobs — the kind that Tamil Nadu's engineering colleges produce graduates for but has historically struggled to retain.
Sovereign AI capability. India currently routes a significant share of its AI inference and data storage through servers physically located outside the country. This is a strategic vulnerability — economic, legal, and in some cases national security. Building domestic data infrastructure reduces dependence on foreign-controlled cloud infrastructure and creates the foundation for an Indian AI ecosystem that actually runs on Indian soil.
Multiplier effects. Data centres attract adjacent industries — chip design, hardware maintenance, cybersecurity, telecom infrastructure. The BFSI sector, e-commerce, health tech, and e-governance all become faster, cheaper, and more reliable when compute infrastructure is close and abundant. The indirect GDP contribution is estimated at 0.5 to 1 percent.
FDI and positioning. India is in direct competition with Singapore, Malaysia, and the UAE for data centre investment. Chennai's submarine cable connectivity — it is the landing point for multiple trans-oceanic cables — gives it a genuine geographic advantage. Losing this window of investment to regional competitors would be costly.
These are real gains. The question is not whether to build data centres. The question is how, and at what hidden cost.
THE POLITICAL CHOICE IN PLAIN SIGHT
When 342 million Indians still lack access to safe drinking water, and when a city of 11 million people spent a summer without piped water seven years ago, the decision to build two-million-litre-per-day facilities in that city is not a technical decision. It is a political one.
The Tamil Nadu government has not published a cumulative water impact assessment of its approved data centre projects. It has not set water usage efficiency standards — the metric the industry calls WUE — as a condition of approval. It has not required Scope 2 carbon disclosure, or mandated that facilities achieve a minimum percentage of renewable energy procurement before drawing from the state grid. It has not modelled what a 10-year data centre build programme means for Chennai's groundwater table in years with deficient monsoons.
These are not exotic regulatory demands. Singapore, which hosts some of Asia's largest data centres, suspended new data centre approvals from 2019 to 2022 specifically to develop water and energy standards. The Netherlands imposed a moratorium on new data centres in its already-congested Amsterdam cluster. Ireland capped data centre electricity connections at 70 MW per site until grid capacity catches up.
India has done none of this. Tamil Nadu has done none of this. The incentives run only in one direction: attract investment, offer subsidies, clear the land, fast-track approvals.
What is not tracked is not managed. And what is not managed tends to be exploited until a crisis makes it visible.
Chennai knows what that crisis looks like. It looked like Malar standing in a queue at six in the morning, holding two plastic containers, waiting for a tanker.
THE UNWRITTEN NUMBER
Here is the number that no Tamil Nadu government document contains:
The combined water draw of all data centres approved for construction in Tamil Nadu — measured against Chennai's safe annual water supply, in a year with average rainfall.
Nobody has calculated it.
Not the Tamil Nadu Data Centre Policy. Not the MoU documents with Adani or Henox. Not the Tamil Nadu Water Supply and Drainage Board's forward planning documents. Not the Central Ground Water Authority's assessments. Not TANGEDCO's grid planning reports.
This is the pattern across India's data infrastructure challenge. The state is extraordinarily precise about the investment value of each new facility — ₹2,500 crore here, ₹50,000 crore there. It is extraordinarily vague about the resource cost.
We know exactly how much money is coming in. We have not tried to measure what is going out.
India is building a twenty-first century AI economy. It is doing so on the same water table, the same transmission lines, and the same planning frameworks that failed Chennai in 2019 and fail farmers across the country every dry summer.
The server needs water. That water has to come from somewhere. Until we know precisely where — and what happens when it runs short — we are not building infrastructure. We are making a bet.
Malar already knows how that bet can end.
If you found this useful, share it with someone who thinks India's AI moment is just about servers and software. The hardware has a hydrology problem.