Our use of cookies Water consumption of AI data centres in the UK and their impact on water resources
Artificial intelligence (AI) is driving a rapid expansion of data centres across the UK, with tens of new facilities planned or under construction and hundreds of megawatts of new computing capacity being added to the grid.
Around the world, data centres have attracted attention for the large volumes of water they consume to keep high-powered servers cool. This is raising concerns about their impact in the South East, where water resources are already stretched.
This guide explains how AI data centres use water in the UK. Here’s what we cover:
- Why AI data centres consume so much water
- How AI data centres consume water
- How much water is used in AI data centres?
- Water sources for AI data centres in the UK
- AI data centre impact on UK water resources
Why AI data centres consume so much water
The processing demand of data centres dedicated to training and running AI models is vastly higher than in a traditional data centre or the server room of an average business.
To handle this level of processing, AI data centres deploy servers densely packed with Graphics Processing Units (GPUs), resulting in individual server racks drawing six to eight times more power than a traditional enterprise server rack.
The vast majority of the electrical energy consumed by AI servers is converted into heat as processing takes place. To maintain the reliability and service life of GPUs, data centres must continuously remove this heat to prevent the chips from overheating.
Keeping GPUs cool requires a high-capacity cooling system that absorbs the heat produced by the servers. As explained in the next section, the cooling systems used by AI data centres typically rely on a water supply.
How AI data centres consume water
Modern AI data centres typically use direct-to-chip, water-based cooling systems to maintain the temperature of server racks.
Here’s how it works:
Absorption of heat generated from the server
Heat in an AI server rack is generated by the silicon chips used in GPUs. The chips are surrounded by a copper or aluminium heat spreader designed to rapidly absorb the heat they produce.
Inside the heat spreader are microchannels that carry a water supply which is continuously pumped through the metal block.
Water absorbs heat
De-ionised water is typically used as the cooling liquid within the micro-channels.
Water has four times the heat absorption capacity of air and liquid water is much more dense. This means a very small supply of water through the heat spreader can absorb vast amounts of heat from the GPUs.
Water supplied into the microchannels enters at 20–30 °C, and leaves at 30–45 °C efficiently transporting the heat energy away from the servers.
Heat is passed through a heat exchanger
Warm water exits the server rack via a single return line that consolidates all individual microchannels and is pumped towards a facility-level cooling system.
The water-based cooling system of an AI data centre typically operates on its own separate cooling loop to protect servers from contamination or water pressure issues.
A liquid-to-liquid heat exchanger transfers heat from the server water into a secondary water supply used by the cooling system.
The water-based cooling systems then dissipate the heat into the outside air using one of the following two cooling methods:
i. Normal operation: Closed loop heat rejection
Outside most AI data centres is a series of air-cooled heat exchangers that operate similarly to a car radiator.
Fans blow ambient air across metal-finned coils, transferring heat from the circulating water into the fins and releasing it into the surrounding air.
In a closed-loop heat rejection system, cooled water returns to the data centre. Crucially, a closed-loop rejection system does not rely on continuous water consumption. Once cooled, the water can restart the process and absorb more heat from the facility.
ii. Hot weather: Open-loop heat rejection
On a hot summer day, air-cooled heat exchangers may be unable to absorb sufficient heat from the water.
To improve heat exchanger efficiency, water is sprayed onto the finned coils.
Using the same principle as sweating, the water sprayed onto the heat exchanger quickly evaporates, removing significantly more heat than air cooling alone.
The drawback is that the evaporated water is lost and must be replaced by adding more water to the cooling system.
How much water is used in AI data centres?
AI data centres use vast amounts of water when they need to rely on open-loop cooling. This is a particular issue for data centres in California, Arizona and Texas, where daytime temperatures frequently exceed 35 °C.
Large hyperscale AI facilities built in desert regions (where solar energy is cheap) can consume millions of litres of water each day.
The International Energy Agency estimates that the data centre sector consumes over 650 billion litres of water annually. This is equivalent to around six weeks’ worth of water supplied by the entire British water industry.
In the UK, where temperatures are lower, AI data centre water consumption is generally much lower.
AI data centres opening in the UK are not required to report how much water they will consume, and business water suppliers have not disclosed data on the consumption of their customers.
As a result, precise figures for the water consumption of AI data centres are not available. However, consumption levels can be inferred from the water abstraction licence obtained by the Equinix Slough Campus, one of the UK’s most important AI-ready infrastructure facilities.
Water consumption at the Equinix Slough Campus
The Equinix Slough Campus is not a single building but a multi-facility hub comprising several colocation data centres near the Slough Trading Estate in Berkshire.
Equinix UK Limited has obtained a groundwater abstraction licence for up to 72,000 litres of water per hour. This rate is equivalent to the hourly water footprint of approximately 3,000 average UK households.
However, this licence does not imply that water is used at this rate continuously; rather, it likely represents the maximum water consumption when open-loop cooling is used during hot weather.
Water sources for AI data centres in the UK
The significant water consumption of data centres during hot weather means they must rely on high-capacity water sources. Below are the typical water sources used by AI data centres in the UK.
Mains water supply
A mains water supply delivers potable water through the same water infrastructure used by homes and businesses.
This is a common solution for smaller data centres, as it provides a quick and reliable source of water but requires payment of business water rates to a licensed supplier.
Larger data centres often use a mains business water connection as a reliable back-up to other sources.
Groundwater abstraction
Groundwater abstraction uses a borehole water system to extract water from an underground aquifer.
A borehole water supply offers a cheaper solution at scale, particularly as treatment is not required for water used in data centre cooling systems.
Borehole water supplies are considered a sustainable water source for businesses, as the water industry does not need to use energy to pump water through the mains system.
Surface water abstraction
Surface water abstraction draws water directly from canals, rivers or reservoirs.
Due to growing water scarcity, the Environment Agency frequently limits the issuance of new abstraction licences.
The high visibility of the environmental impact makes this a less popular option for data centres.
Supplementary rainwater harvesting
AI data centres are well positioned to benefit from a commercial rainwater harvesting system to supplement their other water sources.
The water used in data centre cooling systems does not need to be potable, meaning that rainwater can be stored and used directly by the cooling system.
AI data centre impact on UK water resources
At present, AI data centres are not a dominant driver of UK water abstraction, as temperatures allow these facilities to rely on closed-loop cooling systems for most of the year.
However, their future impact may increase, particularly in water-stressed regions such as the South East, as AI workloads increase cooling intensity, peak demand, and the likelihood of evaporative cooling being used during hot periods.
The main risk is not annual volume but concentrated peak abstraction from potable or groundwater sources during drought conditions, which complicates water resource planning.
Are there alternatives to water-cooling at data centres?
A survey by TechUK on English data centre water consumption in August 2025, found that 51% of the sites surveyed use waterless cooling systems.
In the sections below, we explain two alternative cooling solutions for data centres and why they are not typically used in AI data centres.
Air-cooled data centres
Common in older data centres, a system of fans allows cool outside air to absorb the heat produced by server racks.
Air from outside is delivered either through ceiling ducts or through perforated floor tiles beneath server racks.
The hot air produced is then exhausted outside using a series of fans.
Since air has a much lower heat capacity than water, air-cooled systems are only practical for older servers with much lower power consumption and cannot support the power and heat output of AI GPUs.
Refrigerant-based cooling
A refrigerant-based cooling system similarly uses cold air to absorb heat from server racks. However, unlike air cooling, an air-conditioning-style system is used to artificially chill the air, making it more effective.
The main drawback of this method is that air-conditioning systems consume large amounts of electricity. Given the high business electricity prices in the UK, refrigerant-based cooling is an uneconomic solution for AI data centres.
Back to blog- Share:
