India wants to accelerate its position in the global AI race, but the data centers behind that expansion need more than GPUs. They need electricity to power the servers and water for certain cooling systems — two resources already under significant strain in several parts of the country. Available estimates put Indian data centers’ water consumption at around 150.3 billion liters in 2025, projected to reach roughly 358.66 billion liters by 2030, an increase of nearly 139%.
India’s AI water and energy consumption in 20 seconds
- India surpassed 1,500 MW of data center capacity in 2025.
- Its water use is estimated at 150.3 billion liters a year, potentially reaching 358.66 billion by 2030.
- 75% of data centers are concentrated in five states.
- More than half sit in water-stressed regions.
- The cooling technology used largely determines each facility’s direct water footprint.
India’s situation helps illustrate a problem also emerging in the United States, Europe and other markets where AI infrastructure is growing rapidly. An AI model can look like a purely digital product, but every response ultimately depends on physical infrastructure made of semiconductors, servers, networks, electrical systems and cooling equipment.
The impact can’t be reduced to the claim that “AI uses water,” either. Not every data center uses the same amount, and not all the water tied to a query is consumed inside the building. Location, climate, electricity source, cooling type and server density can all substantially change the outcome.
India adds another complication: much of its new digital infrastructure is concentrating in territories that already face water availability problems.
Why a GPU ends up tied to water consumption
The chain starts with electricity.
Training and running large models requires enormous amounts of math. That relies on GPUs and other specialized accelerators installed in servers that can be grouped by the thousands or tens of thousands.
Today’s AI accelerators can draw several hundred watts per chip and, in the highest-performance configurations, approach or exceed a kilowatt depending on the hardware and its operating conditions.
But the GPU is only part of the picture.
Memory, CPUs, storage, network switches, optical systems, power supplies and the rest of a server’s components also draw electricity.
Then comes an unavoidable consequence: almost all of that electrical energy ends up as heat.
A data center has to continuously remove that heat to keep chips within their operating temperatures.
Traditional systems combine air, fans, HVAC units, chillers, cooling towers, pumps and heat exchangers.
AI racks are also accelerating the adoption of direct-to-chip liquid cooling, because moving enough air to cool racks drawing hundreds of kilowatts is becoming increasingly difficult.
That’s where water enters the picture.
Some facilities use evaporative systems that remove heat by evaporating water. Part of that water stops being locally available in liquid form and needs to be replenished.
The physical chain can be simplified like this:
| Step | What happens |
|---|---|
| AI | Runs mathematical operations |
| GPU/TPU/accelerator | Consumes electricity |
| Server | Generates heat |
| Cooling | Removes that heat |
| Evaporative system | Can consume water |
| Power grid | May also carry a water footprint |
| More AI infrastructure | Increases aggregate demand |
There’s also an indirect water footprint that’s often left out of the conversation.
Generating electricity can also require water. Thermal and nuclear power plants, for example, can use significant amounts for their cooling processes, though water withdrawal and net consumption are different metrics.
That’s why calculating an AI application’s full water footprint requires looking at both the data center and the power grid feeding it.
India could go from 150 billion to nearly 359 billion liters
The scale of the projected growth explains why the issue is entering India’s public debate.
Mordor Intelligence estimates 150.3 billion liters of water consumption in 2025 for the country’s data centers, projecting 358.66 billion by 2030. That’s roughly 139% more over five years.
| Year | Estimated consumption |
|---|---|
| 2025 | 150.3 billion liters |
| 2030 | 358.66 billion liters |
| Increase | ~139% |
An important correction is worth making here regarding the chart circulated by Outlook Business. Although the infographic attributes the data to CEEW, the outlet’s own article states the figures are based on reporting drawn from Mordor Intelligence. CEEW has published its own research on Indian infrastructure and its relationship with water and energy, but the 150.3 billion and 358.66 billion figures come directly from Mordor Intelligence’s estimates.
Nor should they be read as an exact measurement of water consumed by “AI.”
They cover the broader data center market, where artificial intelligence coexists with cloud services, storage, video, enterprise applications, e-commerce and many other workloads.
AI is one of the factors accelerating that expansion, but it isn’t responsible for all of that consumption.
The problem is also where these facilities get built
Location turns the water question into something far more complex than a single national figure.
WRI India counts 278 data centers and notes that 75% are concentrated in five states: Maharashtra, Tamil Nadu, Karnataka, Telangana and Uttar Pradesh. More than half of the country’s facilities sit in regions under water stress.
Water stress measures the ratio between existing demand and available renewable resources. The higher that ratio, the more competition there is between households, agriculture, industry and other users.
That means consuming a million liters of water doesn’t necessarily carry the same impact everywhere.
A facility located in a region with abundant water resources, moderate temperatures and access to reclaimed water sits in very different conditions from one built where households and farmers are already competing for limited supplies.
That concentration shows up in some of the major cities captured in WRI’s map:
| Area | Data centers identified |
|---|---|
| Mumbai | 47 |
| Chennai | 34 |
| Bengaluru | 31 |
| Hyderabad | 29 |
| Navi Mumbai | 22 |
| Pune | 16 |
| Noida | 15 |
| Ahmedabad | 11 |
| Kolkata | 10 |
WRI’s visualization also shows that many facilities coincide with areas classified as having high or extremely high water stress.
That overlap matters more than the aggregate national figure.
Liquid cooling doesn’t automatically mean less water use
There’s also a common mix-up between liquid cooling and water consumption.
They aren’t the same thing.
Direct-to-chip cooling can use a closed loop where the liquid circulates repeatedly between servers and heat exchangers. That doesn’t mean all of that water is continuously consumed.
Consumption depends on how the heat is ultimately rejected.
A system using evaporative cooling towers may need continuous replenishment. A dry-cooling architecture can cut direct water use significantly, though it usually brings other trade-offs related to energy, efficiency, outdoor temperature and cost.
That’s why two data centers with the same IT capacity can have very different water footprints.
India’s own government notes that the sector is turning to technologies such as direct-to-chip liquid cooling, adiabatic cooling and immersion cooling to improve both energy and water efficiency. The Ministry of Electronics and Information Technology also said national capacity grew from 375 MW in 2020 to more than 1,500 MW in 2025.
That figure illustrates the pace of the expansion.
Electricity could become an even bigger problem
Water is only part of the equation.
The International Energy Agency (IEA) projects India’s total electricity demand will grow an average of 6.4% a year between 2026 and 2030. The country would add more than 570 TWh of annual consumption during that period.
Not all of that growth comes from data centers. Industry, air conditioning, transport, agriculture and broader electrification carry far more weight.
But digital infrastructure adds a new, especially concentrated load.
Outlook Business reported in August estimates currently putting Indian data center capacity around 1.6-1.8 GW, with the potential to reach 10.5-11.5 GW by 2035. India’s Ministry of Power reportedly estimated electricity demand from the sector could reach 13.56 GW by 2032.
Another projection cited by the same outlet estimates capacity growing from 1,668 MW in 2025 to 8,120 MW in 2030. These are projections from different sources and shouldn’t be treated as if they described exactly the same metric or scenario.
Globally, the trend is similar. The IEA calculates that data centers could go from consuming around 415 TWh in 2024 to roughly 945 TWh in 2030, with artificial intelligence as the main driver of that additional growth, alongside other digital services.
Google, Microsoft and Amazon are speeding up their investments
India has economic reasons to want that infrastructure.
The country combines a population of more than 1.4 billion people, a huge digital economy, growing cloud adoption and a strategic position for building local compute capacity, one of the threads running through India’s broader deep-tech push across chips, space and capital.
Big tech companies are responding with multi-billion-dollar investments.
Outlook Business points to Google’s $15 billion project in Andhra Pradesh, Microsoft’s plans tied to Hyderabad within a $17.5 billion investment, and Amazon’s program to invest $35 billion in its Indian businesses through 2030, of which a significant share is expected to go toward data centers.
That generates investment, construction-phase activity, demand for local suppliers and digital capacity.
But the relationship between a data center and jobs differs from that of a large traditional factory.
Building a campus requires many workers, contractors, engineers and suppliers. Once complete, operating a highly automated facility can require proportionally far fewer permanent employees than other industrial activities with comparable investment.
The local debate emerges precisely when communities weigh those local benefits against the resources needed to keep the facility running.
The conflict isn’t simply for or against AI
India’s case shows why reducing this debate to “AI uses too much water” falls short.
The real problem combines several questions.
How much water does each facility actually use? Is it potable, reclaimed or from another source? How stressed is the local watershed? What cooling technology is used? How much electricity does it need? Who pays for new grid infrastructure? Where does that power come from? What jobs and investment does it generate?
The answers can turn two seemingly similar projects into infrastructure with completely different impacts.
It’s also important not to automatically pin a region’s existing water problems on data centers.
Outlook Business has examined declining groundwater levels in major data center hubs such as Noida, Bengaluru, Mumbai and Pune. However, the outlet itself notes that groundwater trends also reflect population growth, urbanization, agriculture and other industrial activity, so that data doesn’t prove data centers are the cause of the decline.
This distinction matters especially in India, where agriculture accounts for a huge share of water use and the monsoon shapes its availability.
Data centers add an incremental problem: they introduce a fast-growing industrial consumer precisely in some of those same regions.
The next data center metric could be water
For years, a data center’s best-known efficiency metric has been Power Usage Effectiveness (PUE), which compares a facility’s total energy use with what its IT equipment consumes directly.
Growing water pressure is giving more weight to another metric: Water Usage Effectiveness (WUE).
It isn’t enough, though, to simply chase the lowest possible WUE in isolation.
Drastically cutting water use can raise electricity consumption in certain cooling systems. And a facility that’s extremely energy-efficient can use more water if it relies heavily on evaporation.
The optimal design therefore depends on location.
In water-stressed regions, it can make sense to accept some increase in electricity use to cut water consumption, especially where abundant renewable generation is available. In other locations, the opposite may hold true.
Reusing treated wastewater offers another way to reduce competition with the potable supply.
CEEW specifically considers that India has significant potential to expand the use of treated wastewater in industrial applications.
AI’s expansion is turning these decisions, once mainly left to facility engineers, into economic and political ones as well.
A data center doesn’t exist only within the internet. It connects to a substation, occupies land, expels heat and needs cooling. Depending on how it’s designed, it can also consume considerable amounts of water.
India shows that reality at scale: the country wants to multiply its compute capacity at the same time that many of the regions hosting that infrastructure already face water stress and rising electricity demand.
The question isn’t whether India should develop artificial intelligence. The more concrete debate is where to build its data centers and what efficiency, power-supply and water-use conditions they should meet, so that growing compute capacity doesn’t shift a disproportionate share of its costs onto the communities around it.
