DATA CENTRES & WATER · INDIA · 16 JUNE 2026
⚠ A single 100 MW facility can consume around 2 million litres a day, comparable to the daily needs of thousands of households.
⚠ In Hyderabad, groundwater in the Gachibowli IT belt fell about one metre in just three months in early 2026.
We talk endlessly about the power that data centres consume. We talk far less about what they drink.
01 · HOW MUCH WATER ARE WE TALKING ABOUT?
A data centre uses water mainly to stay cool.
The cheapest method in a hot climate is evaporative cooling — spraying water that evaporates to carry heat away. It works, but it is thirsty: a single 100 MW facility can consume around 2 million litres a day, comparable to the daily needs of thousands of households.
Scale that up. India’s data centres used roughly 150 billion litres of water in 2025, projected to more than double to about 358 billion litres by 2030. That headline figure includes the water embedded in generating the electricity they consume; the on-site cooling draw is smaller, but it is concentrated — large, steady withdrawals in specific places.
And here is the catch: most of that demand sits in cities that are already short of water — Mumbai, Hyderabad, Chennai, Bengaluru and the Noida belt. India still treats Water Usage Effectiveness (WUE) as a voluntary metric. Most state data-centre policies set no binding water-performance standard at all.
02 · THE SOCIAL COST IS LOCAL — AND IT IS ALREADY HERE
Averages hide the pain.
A sector that is a small share of national water use can still hollow out a specific aquifer, because its draw is large, constant and clustered. The early warning signs are visible:
In Hyderabad, groundwater in the Gachibowli IT belt fell about one metre in just three months in early 2026.
In Noida’s Gautam Buddh Nagar, groundwater extraction already runs at roughly 105% of natural recharge.
Data centres are classified as ‘essential services’ in Maharashtra and Telangana — so in a shortage they can be prioritised for water ahead of drinking water and sanitation.
Set that against the backdrop of an estimated 600 million Indians already facing high water stress, and the equity question writes itself.
This is why hyperscale projects in stressed regions are starting to draw protests and legal challenges.
03 · THE GOOD NEWS — THE FIXES ALREADY EXIST
This is not a problem waiting on a breakthrough.
The toolkit is mature; what is missing is the will and the standards to use it.
Air and closed-loop / liquid cooling recirculate or avoid water almost entirely — and liquid cooling is anyway becoming necessary for dense AI racks.
Treated-wastewater reuse — cooling with recycled effluent instead of fresh drinking water.
Reusing the cooling ‘blowdown’ for landscaping, dust suppression or flushing, rather than sending it to drain.
And, for the right sites, desalination.
The leading hyperscalers have also pledged to be ‘water-positive’ by 2030. That is welcome — but worth a caveat: replenishing a watershed somewhere else does not necessarily relieve the specific aquifer a facility is draining. For communities, local sourcing matters more than distant offsets.
04 · IS DESALINATION THE ANSWER? YES — AND NO
For a coastal data centre, desalinated seawater is genuinely attractive.
It draws on the ocean instead of community freshwater. India already runs large seawater reverse-osmosis (SWRO) plants in Chennai, and a coastal AI hub such as Visakhapatnam is a natural candidate. But it is not a silver bullet, and three engineering realities are usually glossed over.
First, energy. SWRO needs roughly 3–4 kWh to make a cubic metre of fresh water. Modest next to a data centre’s IT load, but real — and it should be powered by the campus’s own renewable supply, not by adding coal to solve a water problem.
Second, the water must be improved after desalination. RO permeate is almost pure, slightly acidic and corrosive, and stripped of minerals — it cannot be used un-conditioned. It must be remineralised (lime or limestone with CO₂), pH-corrected, and dosed with corrosion inhibitor and biocide before it ever touches a cooling tower.
Third, the by-products. About half of the intake leaves as concentrated brine, which must be diluted through a proper outfall or processed (for salt, bromine, magnesium) — done badly, it harms marine life. At the cooling tower, ~75–80% of the make-up water is consumed by evaporation; the remaining ~20–25% leaves as ‘blowdown’ — not waste, but a reusable resource.
And it does nothing for inland hubs — Hyderabad, Noida, Bengaluru — where the pressure on groundwater is greatest. There, air and closed-loop cooling and treated-wastewater reuse are the levers that matter.
05 · WHAT INDIA SHOULD DO NOW
None of this requires us to slow the digital build-out.
It requires us to build it water-smart:
Make WUE standards and water-use disclosure mandatory for data centres — measure it, cap it, publish it.
Steer siting away from over-exploited groundwater blocks, and toward treated-wastewater and coastal options.
Prefer air / closed-loop cooling and recycled water as the default, with desalination for coastal sites — renewable-powered, with responsible brine management.
Revisit the ‘essential services’ classification so that community drinking water is never deprioritised in a crisis.
THE VERDICT
The cloud feels weightless. It is not. Every AI query has a water cost, and in India that cost is increasingly paid by the communities next door. We can build a digital superpower and protect our water — but only if we decide to measure and manage both, starting now.
ABOUT THIS PIECE
M. V. Radhakrishnan. Four decades across India’s power and renewable-energy value chain; writing on the energy, water and infrastructure choices behind the AI build-out. Figures drawn from public reports and trade press, 2025–26; several are estimates.