UN report highlights escalating water and energy demands of future data centres

A new United Nations University report has put an eye-catching scale on the resource demands behind the data centre build-out, estimating that global facilities could consume about 945 terawatt-hours of electricity in 2030 and carry a water footprint of 9.3 trillion litres.

The report, Environmental Cost of Artificial Intelligence: Carbon, Water and Land Footprints, was produced by the United Nations University Institute for Water, Environment and Health. It is a research report rather than a peer-reviewed journal paper, and its authors, led by Miriam Aczel, base their headline projections on International Energy Agency scenarios rather than on a separate bottom-up count of planned sites or contracted loads.

That distinction matters. The electricity figure is a forecast for 2030, not a measured reading of current use, and the water number is a footprint attached to that projected power demand rather than simply the volume of water delivered to cooling systems inside server halls. According to the UN University summary of the work, the water estimate includes both direct cooling needs and water associated with electricity generation.

The International Energy Agency’s modelling has become the backbone of much of the recent discussion around AI’s energy appetite. In its 2025 base case, the agency said data-centre electricity use could rise from roughly 415 TWh in 2024 to 945 TWh by 2030, with AI accounting for much of the increase but not all of it. IEEE Spectrum has reported that the IEA later updated its central 2030 estimate to about 950 TWh, reinforcing the same broad picture: demand is rising fast, even if the exact end point remains scenario-dependent.

The UN University report has also drawn attention because of the comparison it makes with national electricity systems. Yet that kind of framing should be read as a scale comparison, not as a claim that data centres will literally displace the annual power consumption of a particular country. It is a way of translating a global number into something legible, not a forecast of direct competition between data centres and households.

The water figure is even easier to misread. Data centres often rely on evaporative cooling, which can be efficient but uses water at the site. There is also indirect water use upstream in power generation, depending on the fuel and grid mix. The result is that a “water footprint” can be much larger than the volume a facility physically draws from a local supply, and the two should not be treated as interchangeable measures.

That distinction matters for industrial decarbonisation because water stress is highly local. A facility using reclaimed water in a cool, water-rich region poses a very different challenge from one drawing potable water during a heatwave in an already stressed basin. A global total can show scale, but it cannot show concentration, timing or competition with other users.

The World Bank has previously warned that indirect water use linked to electricity can dominate a data centre’s broader footprint, while also noting that the available accounting is patchy and fast-moving because operators do not all report in the same way. That makes absolute precision difficult at a time when AI adoption, chip efficiency and cooling design are all changing quickly.

Cooling choices also shift burdens rather than removing them. Dry cooling can reduce on-site water use, but it may require more electricity or suffer performance losses in hot conditions. Conversely, evaporative systems can be more efficient on power but increase water consumption. The US Department of Energy uses water usage effectiveness as one way to compare sites, but that metric does not capture the water consumed in producing the electricity that powers the facility.

Energy supply adds another layer of trade-off. Wind and solar PV generally have low operational water use, while thermal power plants usually require water for cooling. Hydropower and bioenergy can also carry significant water or land footprints depending on how boundaries are drawn. The broader conclusion in the UN University report is that lower-carbon electricity is not automatically lower-water electricity.

Even so, the headline numbers do not mean data centres will become the dominant source of new electricity demand worldwide. The IEA has said they will account for less than 10 per cent of global electricity-demand growth through 2030. The difficulty is that the sector is geographically concentrated and can be built far faster than many power grids, generation assets and water systems can expand to accommodate it.

That is why direct measurement will matter more than broad comparisons. Regulators and investors will need clearer reporting on withdrawals, consumption, water source and seasonal loading, alongside separate disclosure of electricity-related footprints. Operators, meanwhile, will increasingly be expected to specify whether water is potable, recycled or returned, and where the boundaries of their accounting begin and end.

For industrial users, the message is less about shock-value comparisons and more about siting, system design and transparency. The most useful question is not whether 9.3 trillion litres sounds large, but where that water is used, under what conditions, and what else depends on the same local supply.