Six consumers alone accounted for 65 per cent of total water use, with the largest data centre in the sample taking up more than a quarter of the 1,066 million litres consumed by the entire 200 facilities.
Thus there is an entirely different story which needs to be told about the future of the sector to the one which TechUK wants us to hear. According to MOSL’s calculations, if the two largest users were representative of the sector as a whole, this would push up “non-household demand by two per cent and the sector usage by 10-fold.”
Blaming
The data centre industry’s claims that this is still expansion from a low base miss the point. Water companies have already made detailed estimates for how to manage competing demands for water from households, industry and public services between 2024 and 2029, and have no idea how many million litres are missing from their plans as data centres mushroom.
And as people in the industry know very well, simply counting the number of facilities tells you little of value, although where they are concentrated in the landscape is important.
The data which really matters is the amount of power required to run servers and other computing equipment (IT Load). Combined with information about the operator’s choice of cooling system configuration, this can be used to estimate water demand.
Precise estimates are difficult, and water companies, data centre operators or public authorities are blaming each for the lack of usable data instead of publishing what they know.
But it is very clear that the choice of cooling system makes a huge difference to the amount of water a data centre uses.
Water
The Environment Agency’s report concludes that hybrid cooling systems are driving these seasonal spikes in demand.
These combine the use of air-cooled and water-cooled systems, largely relying on air cooling when the temperature is low, and only turning on water-intensive chillers when temperatures rise.
Hybrid systems were in use in 44 percent of data centres in TechUK’s survey, with a further 5 percent using water-based systems alone, while the remaining 51 percent claimed not to use any water at all, except for domestic use such as flushing the toilets.
What the Environment Agency analysis confirms is that hybrid cooling systems are still “water intensive.” MOSL’s analysis also highlights how the increasing scale of data centres exacerbates the risks posed by spikes in demand for water during heatwaves.
When challenged on their water use, data centre operators often claim that new facilities entering the planning system won’t require large amounts of water on site.
Gigafactory
The problem is that reducing water use increases the power required for fans and mechanical cooling systems, meaning that “waterless” data centres will have a greater indirect water use at the electricity generation source.
Data centres now being built or planned across the UK consume electricity on the same scale as many small power plants produce it. Other data centre hotspots globally, such as the major clusters in northern Virginia and Ireland, have reshaped the entire electricity grid around their demands.
Ireland banned new connections for data centres in the Dublin area between 2021 and late 2025 as the grid struggled to cope. Add in the impact of spikes in power demand caused by heatwaves as fans and cooling systems work overtime to reduce temperatures in heat trap homes, offices and hospitals and the choice between less water and more power becomes even more problematic.
The Environment Agency cautions that their analysis has not even considered the impact of AI hyperscale data centres. Nor is the issue of AI production simply confined to “gigafactory” style sites away from urban centres.
Efficiency
Some AI workloads need to run in physical proximity to the end-user: for example in autonomous vehicles’ driver assistance systems. City-centre data centres will be also more likely to opt for water-based cooling because they don’t have the floor space for alternatives, major data centre operator Global Switch told MOSL.
The authors of a UN report published last month confirm there is no easy trade-off between power, water and land use for data centre operators and the AI industry in a world where climate breakdown is accelerating.
“A lot of people think that the environmental footprint of AI reduces, as technology improves and processes become more efficient. But that is only a partial picture of the overall problem,” said Professor Kaveh Madani, a co-author of the report.
“More efficient and affordable AI and energy mean more consumption of AI, making the overall footprint far bigger than what we save through efficiency gains.”
This Author
Anne Alexander is a journalist and researcher. She is working currently working on a project with Watershed Investigations looking at the impact of data centres and the AI industry on the living planet.
