A June 2026 United Nations University report puts two country-sized numbers on the physical infrastructure behind cloud computing. Its authors project that data centers could use about 945 terawatt-hours of electricity in 2030, with an associated water footprint of 9.3 trillion liters.
The comparisons are arresting, but their definitions matter. The water figure is a modeled footprint attached to data-center electricity demand, not a meter reading of water piped into server buildings. The electricity figure is a scenario for 2030, not a measured outcome.
This is one report, not settled consensus.
The electricity projection comes from the IEA
The report, Environmental Cost of Artificial Intelligence: Carbon, Water and Land Footprints, was published by the United Nations University Institute for Water, Environment and Health. It is a research report rather than a peer-reviewed journal paper. Miriam Aczel is the lead author, working with colleagues including institute director Kaveh Madani.
Its starting point is a global electricity projection from the International Energy Agency. In the IEA’s 2025 base case, data-center electricity consumption rises from roughly 415 TWh in 2024 to 945 TWh in 2030, just under 3 percent of projected global electricity use. AI is the most important driver of that growth, but the total covers all data-center activity. It should not be described as electricity used only by AI.
The IEA built sensitivity cases around uncertain rates of hardware and software efficiency, AI adoption and energy-system bottlenecks. Its 2026 update puts the central 2030 estimate at about 950 TWh. The near match supports the scale used by the UN University team, while the scenario range remains important.
The comparison with Pakistan, Bangladesh and Nigeria is therefore a translation of the IEA base case. It is not an independent forecast produced by adding up announced data-center projects.
The 9.3-trillion-liter figure is not direct cooling water
Servers turn nearly all the electricity they use into heat. Many data centers remove that heat with cooling towers, where evaporation consumes water at the facility. There is also an upstream demand: power stations and fuel supply chains can withdraw or consume water while producing the electricity that reaches the site.
The UN University team describes its 9.3-trillion-liter number as the water footprint associated with projected 2030 electricity consumption. In its June 3 summary of the report, the university says the footprint includes consequences from cooling and power generation. It is broader than the volume a local utility delivers to a data center and depends on the mix of energy sources supplying the load.
That distinction is essential. Water withdrawal is the amount taken from a river, reservoir or aquifer, some of which may be returned. Water consumption is the portion not returned to the same local system, often because it evaporates. A water footprint can extend further still, counting water embodied elsewhere in the energy supply chain. Those measures answer different questions and should not be used interchangeably.
How 9.3 trillion liters becomes 1.3 billion people
The domestic-needs comparison uses a basic minimum rather than average household consumption. Divide 9.3 trillion liters by 1.3 billion people and the result is about 7,150 liters per person per year, or 19.6 liters a day. Rounded, that is a 20-liter daily benchmark.
That arithmetic does not mean data centers will remove a year of household water from 1.3 billion people. A liter consumed by electricity generation in one watershed is not interchangeable with a liter of drinking water on another continent. The comparison communicates scale. It does not locate the harm or establish that domestic users and data centers will compete for the same supply.
It also uses a deliberately basic allowance. Actual domestic water use varies enormously among countries and households and can be many times higher where piped water is abundant. Changing the benchmark changes the equivalent population without changing the underlying footprint.
A global total cannot show where water risk lands
Water impacts depend on geography, season and water quality. A facility using reclaimed wastewater in a cool, wet region presents a different problem from one evaporating potable water during a drought. Even identical annual volumes can have sharply different consequences in different watersheds.
A World Bank review of data centers and water notes that indirect use associated with electricity can make up 80 percent or more of the overall footprint. It also says precise accounting remains difficult because company reporting is incomplete and both AI adoption and technical efficiency are changing quickly.
Globally, data-center water demand remains small beside agriculture. Locally, a cluster of large facilities can still compete with homes, farms and ecosystems, particularly when its peak cooling demand coincides with hot, dry weather. A worldwide number smooths away that concentration.
Cooling systems move costs rather than erasing them
Evaporative cooling is common because water can carry heat away efficiently. Dry cooling can sharply reduce on-site water consumption, but it may use more electricity or lose performance during high temperatures. More power demand can shift part of the water burden upstream to electricity generation.
The US Department of Energy’s cooling-water guidance defines water usage effectiveness as annual site water use divided by the energy consumed by IT equipment, measured in liters per kilowatt-hour. That is useful for comparing facility cooling, but it does not by itself capture water consumed in producing electricity.
Energy choices create another trade-off. Wind and solar photovoltaic generation generally require little operational water. Thermal power plants often use water for cooling, while hydropower and bioenergy can carry substantial water or land footprints depending on how boundaries are drawn. The UN University report’s broader point is that low-carbon power is not automatically low-water or low-land.
What the headline numbers can and cannot tell us
The projection does not make data centers the dominant source of new electricity demand worldwide. The IEA estimates that they contribute less than 10 percent of global electricity-demand growth through 2030. The difficulty is that data centers are geographically concentrated and can be built in two or three years, faster than many grids, power plants and water systems can expand.
Nor is 945 TWh inevitable. Better chips, more efficient software, slower construction, grid delays or changes in demand could push the total down. Rapid adoption, larger models and rebound effects could push it up. A projection is useful precisely because it allows planners to test what infrastructure would be required under stated assumptions.
The strongest conclusion is about measurement. Regulators need direct withdrawals, direct consumption, water source, seasonal demand and the electricity-related footprint reported separately. Operators should state whether water is potable, reclaimed or returned, and which parts of the supply chain sit inside the boundary of a published number.
The comparison with 1.3 billion people is memorable. The more useful question is where each liter is consumed, what kind of water it is and what else depends on the same source. Those details will decide whether the data-center buildout becomes a manageable engineering demand or a concentrated burden on communities sharing the same grids and watersheds.