Water UK has warned that Britain’s plans for rapid datacentre expansion are moving ahead without sufficiently detailed planning for the water needed to cool and operate new computing infrastructure.

In evidence submitted to Parliament, the industry body estimated that existing British datacentres consume about 6.6 million litres of water each day, broadly equivalent to the use of 20,000 average homes. If capacity trebles, average demand could rise to 19.8 million litres, although the local effect will vary according to location, cooling design, weather, and the availability of non-potable supplies.

Summer peaks create a greater infrastructure challenge than the annual average. Water UK estimates that existing facilities may draw around 14 million litres on hot days, potentially rising to 42 million litres if capacity triples and considerably further if new sites perform as poorly as the least efficient facilities.

Those figures remain estimates because British datacentres do not publish water consumption in a consistent facility-level format. Developers have nevertheless approached water companies about connections reaching as high as 21 million litres a day for individual projects, showing that some proposed sites could place substantial pressure on a local network.

National supply figures conceal local shortages

Even the highest estimates represent a small share of the roughly 14 billion litres entering public supply each day, but water cannot be moved freely between every region and network. More than three-quarters of British datacentre capacity is concentrated in the east and south-east of England, where existing water stress overlaps with demand for low-latency access to London’s financial and communications infrastructure.

Cooling requirements can also rise during hot weather, when households, agriculture, and other businesses place additional pressure on the same networks. A facility using evaporative cooling may therefore create its highest demand when spare local capacity is at its lowest.

Reducing water use can increase electricity consumption, which makes the infrastructure trade-off more complicated than selecting a single efficient cooling technology. Air cooling and some closed-loop systems may require less fresh water while drawing more power during hot conditions, whereas evaporative systems can lower electricity use by consuming more water.

Operators and planners need to assess both resources together. A proposal described as energy efficient may shift part of its burden onto water infrastructure, while a low-water design may add demand to an electricity grid that is already delaying large connections.

Location affects both sides of that equation because cooler regions can reduce cooling demand while areas with available grid capacity may sit further from established network routes and customers. Developers will weigh latency, land, electricity, water, planning, and construction cost rather than optimise one resource independently.

Planning requires evidence at facility level

Britain has designated datacentres as critical national infrastructure and is promoting additional capacity through AI growth zones, planning reform, and public computing programmes. Strategic support can accelerate development, but it cannot supply water where regional networks lack capacity.

Water companies plan reservoirs, treatment works, pipes, and abstraction over long periods, while a datacentre can move quickly once land, finance, power, and a major customer are secured. Where expected demand enters resource planning too late, the utility faces expensive reinforcement or the developer discovers that a commercially attractive site cannot obtain its intended connection.

Water UK’s parliamentary evidence calls for better data and earlier planning rather than a blanket halt to development. Reporting water withdrawal, actual consumption, peak demand, cooling technology, and the use of potable or recycled supplies would allow local authorities and utilities to assess proposals against regional conditions.

Developers can reduce pressure through reclaimed water, rainwater collection, closed-loop cooling, and locations where climate and infrastructure are more suitable. Each option carries a cost or operating constraint, leaving planning rules and connection charges influential in whether the more resilient design is selected.

The issue should not be simplified into an automatic contest between household supply and AI. Domestic customers have legal protections, while industrial users may face restrictions, delayed connections, or higher infrastructure costs where supply is insufficient.

Commercial consequences can therefore emerge through projects moving elsewhere, additional investment in water systems, or lower computing capacity at a chosen site. Regions capable of supplying electricity and water together may gain an advantage as the physical requirements of AI become clearer.

Datacentre water demand has been recognised internationally for years, but the density and scale of AI infrastructure have increased the urgency of British planning. Larger facilities are being proposed while several regions are already balancing drought risk, housing growth, agriculture, and ageing networks.

Consistent site-level measurement would not remove those competing demands, although it would replace broad estimates with evidence. Until that information is available, national ambitions, utility plans, and local decisions will continue to rely on an incomplete account of AI’s physical footprint.