The Cloud Needs Water: The Missing Resource in Britain’s AI Growth Plan

Artificial intelligence may appear to exist somewhere in “the cloud”, but the infrastructure behind it is physical, power-hungry and, in many cases, surprisingly thirsty.

AI may live in the cloud, but the cloud is not weightless.

It consists of enormous buildings filled with densely packed processors, storage systems, cables, cooling equipment, pumps and backup generators. Every AI request, cloud backup, streamed film and online transaction ultimately depends on physical machinery producing considerable quantities of heat.

That heat must be removed.

In some data centres, air and sealed cooling circuits do most of the work. In others, water is evaporated through cooling systems. Even where relatively little water is used inside the building, additional water may be associated with generating the electricity that powers it.

Water UK has now issued a stark warning: Britain’s ambitions to expand artificial intelligence infrastructure could be restricted because water demand has not been adequately included in national planning.

In written evidence submitted to Parliament in April 2026, the water-industry trade body described official forecasts as “fatally flawed”, arguing that they “explicitly exclude” future data-centre demand. It also criticised several major government AI policy documents for making no meaningful provision for water alongside electricity, land and planning.

This is not simply an argument about computers.

It is about whether Britain can construct a modern digital economy without placing additional pressure on homes, businesses, rivers, wildlife and already stretched water networks.

AI cannot be described as sustainable if its water requirements are invisible.

An Important Qualification: Most of the Forecasts Concern England

Although this is frequently presented as a UK-wide warning, water policy is largely devolved.

The detailed five-billion-litre supply figure, the Environment Agency framework and most of the evidence about planning relate specifically to England. Scotland, Wales and Northern Ireland have their own systems, although many of the underlying challenges—including climate change, infrastructure investment and industrial demand—are shared.

This distinction matters because broad claims that “the UK is running out of water” can conceal where the most serious shortages are likely to occur.

The immediate problem is not necessarily that Britain receives too little rain overall. It is that rainfall arrives in the wrong places, at the wrong times and is not always captured, transferred, treated and stored effectively.

What Water UK Is Actually Warning About

The headline suggests that Britain simply does not possess enough water to operate future data centres.

The full argument is more complicated.

Water UK says the government wants data-centre capacity to treble by 2030, but existing water-resource forecasts do not include that expansion properly. Water companies use those forecasts when deciding how much infrastructure will be needed, including reservoirs, treatment works, pipelines and water-transfer schemes.

When a large source of future demand is missing, companies may be instructed to plan for considerably less water than communities and businesses will eventually require.

The Environment Agency’s National Framework does discuss data centres. It acknowledges that they can require large quantities of water, that companies have struggled to obtain reliable consumption information and that this lack of data prevents accurate forecasting. It also warns that many new facilities could be built before major strategic water projects become operational.

The disagreement is therefore not over whether data centres use water.

It is over whether their future requirements have been measured, included and funded adequately.

That is a far more serious planning problem than a single dramatic headline might suggest.

England Already Faces a Large Water Challenge

The data-centre debate is taking place against an uncomfortable background.

The Environment Agency estimates that, under a “do nothing” scenario, England could require almost five billion additional litres of public water supply every day by 2055. A further one billion litres a day may be needed by sectors outside normal public supply.

That does not mean taps will suddenly run dry in 2055.

It means the gap must be closed through a combination of:

• New reservoirs and water-transfer systems
• Water recycling and desalination where appropriate
• Reduced household and business consumption
• Lower leakage
• Smarter metering
• Environmental restoration
• Better drought preparation

Many of these projects take years or even decades to plan, approve and construct. Water UK says a major reservoir can take approximately 15 years to move from planning to operation. Meanwhile, the government hopes to accelerate AI infrastructure during the remainder of this decade.

We therefore have two very different timetables.

Data centres can be proposed and approved relatively quickly. New water supplies cannot.

Why Do Data Centres Need Water?

The amount of water used depends heavily on the cooling system, the building’s size, the processors installed, the local climate and how intensively the facility operates.

Evaporative cooling

Evaporative systems pass warm air through or alongside water. As some of the water evaporates, heat is removed.

This can be energy-efficient, but the evaporated water is consumed rather than simply borrowed and returned. Usage may rise during hot weather—exactly when households, agriculture and the wider environment are also experiencing maximum demand.

Closed-loop cooling

In a closed-loop system, water or another coolant circulates through sealed pipes.

It may require an initial fill and occasional topping up, but considerably less water is lost during normal operation. However, pumps, fans and chillers may use more electricity.

Air cooling

Air cooling can reduce direct water use substantially. Its effectiveness, however, depends on outside temperatures, processor density and energy availability.

As AI chips become more powerful and generate more heat in a smaller space, relying entirely on conventional air cooling becomes more difficult.

Direct-to-chip liquid cooling

Liquid can be circulated close to the hottest parts of a processor, moving heat away more efficiently than cooling an entire room.

This can form part of a sealed system and may be suitable for increasingly dense AI computing equipment.

Hybrid systems

Many modern facilities combine several methods. They may use outside air during cooler weather, sealed liquid cooling during normal operation and evaporation only during periods of intense heat.

Government evidence citing an industry survey says 51% of the surveyed UK facilities used closed-loop or nominally “waterless” systems, 44% used hybrid systems and 5% were fully water-cooled. That is an important reminder that not every data centre consumes vast amounts of water.

The real question is not simply, “Do data centres use water?”

It is:

How much potable water will each facility consume, where will it be built and what happens during a hot, dry summer?

Water Withdrawal Is Not the Same as Water Consumption

Water figures can easily become misleading.

A facility may withdraw water, circulate it through a system and return most of it. Another may withdraw a smaller volume but lose almost all of it through evaporation.

Those two operations could appear similar in a simple supply figure but have very different effects on the local water system.

Developers should therefore publish several separate figures:

• Total water withdrawn
• Water actually consumed
• Water returned to the environment or sewer
• Potable water used
• Recycled or non-potable water used
• Peak demand on the hottest day
• Expected demand during drought restrictions

Without this information, the public cannot compare one development fairly with another.

The Numbers Do Not Yet Agree

One of the most revealing parts of this debate is that official and industry estimates differ.

Water UK estimates that English data centres currently use approximately 6.6 million litres of water a day and argues that this could rise to 19.8 million litres if capacity trebles without other changes.

The government’s written evidence gives a lower figure. Its analysis of approximately 200 English data centres estimated consumption of 1,125 megalitres a year—about 3.1 million litres a day. It says this is less than 1% of national non-household water consumption.

The sources do not provide a simple explanation for the difference. They may be using different definitions, datasets, facilities or measurements.

But the disagreement makes the central point rather well.

We are planning a rapid expansion of critical digital infrastructure without possessing a complete, consistent and publicly accessible record of its water use.

Mandatory reporting should have come before accelerated expansion, not after it.

A Small National Percentage Can Be a Large Local Problem

It would be easy to look at the national total and conclude that data centres are insignificant compared with household demand or leakage.

That would miss the most important issue: concentration.

Large facilities are not distributed evenly across the country. Many are clustered in London, Slough and the wider South East—areas that already experience serious pressure on water resources.

Government evidence says 38 of England’s 50 highest-consuming data centres are in the South East. It also found that data-centre water consumption in that region increased substantially between 2021 and 2025, with demand tending to peak during the summer.

Affinity Water supplies the Slough–West London corridor, where approximately 125 data centres are proposed or under construction. It says some individual facilities have requested supplies of up to 35 litres a second—approximately three million litres a day and comparable with the peak demand of 3,500 homes.

One exceptional proposal reportedly sought 21 million litres a day. These were requested capacities rather than proof of actual daily consumption, but they illustrate the scale that planners may be asked to accommodate.

A data centre using three million litres a day may appear manageable within a national total of billions.

Place several of them within the same water-resource zone during a drought, however, and the calculation changes completely.

The Hottest Days Create the Greatest Conflict

Data-cententre water demand can rise when temperatures increase.

Unfortunately, those are also the days when:

• Gardens need more water
• Agricultural demand is high
• People drink and wash more
• Rivers may be at their lowest
• Reservoir evaporation increases
• Hosepipe restrictions are most likely
• Wildlife is most vulnerable

Average annual consumption can therefore hide the pressure placed on the network during a short period of extreme heat.

A facility might use relatively little water for most of the year but request the ability to draw millions of litres during the hottest days.

Water companies must maintain sufficient treatment, pumping and storage capacity to meet that potential peak, even when the full allowance is rarely used.

That is why planning permission should consider peak-day demand as well as annual averages.

My Concern Is Fairness as Much as Supply

I am not opposed to artificial intelligence.

I use AI, recognise its value and believe it can improve education, scientific research, engineering, medicine and public services. Properly directed, it may also help us manage energy, identify leaks, forecast demand and understand environmental change.

But support for technology cannot mean ignoring its physical consequences.

I have previously written about hosepipe bans and the way restrictions affect ordinary people. My own mother cannot easily carry heavy watering cans and has needed permission to use a hose sparingly to keep a few plants alive.

It is difficult to ask elderly people, gardeners and families to save every possible litre while the water requirements of major commercial developments remain uncertain or commercially confidential.

That does not mean households should stop conserving water.

It means large users should face at least the same expectation of transparency, efficiency and responsibility.

Public support will quickly disappear if people believe they are being asked to make sacrifices while powerful companies are given priority access to limited resources.

Water UK’s Diagnosis Deserves Attention—but Not Every Prescription

Water UK is an industry trade body. Its members have an understandable interest in securing investment for new infrastructure and increasing the amount of water they are allowed to supply.

Its evidence recommends useful measures, including efficiency standards, non-potable water, rainwater harvesting, better developer contributions and improved forecasting.

However, it also suggests temporarily relaxing some restrictions on abstraction from the environment.

That proposal deserves much greater caution.

A shortage cannot be solved sustainably by simply taking more water from rivers, wetlands, chalk streams and aquifers that may already be under pressure.

The purpose of environmental abstraction limits is not to obstruct growth. It is to stop economic activity from damaging the natural systems upon which our water supply ultimately depends.

Water UK may be right about the planning failure without being right about every proposed solution.

What a Responsible Data-Centre Water Policy Should Include

1. A water assessment before planning permission

Every major proposal should include a detailed water statement before approval.

It should explain normal demand, peak demand, drought-year demand, cooling technology, supply source, wastewater output and contingency arrangements.

“Water-efficient design” should not be accepted as a vague planning phrase. Developers should provide measurable figures.

2. Mandatory public reporting

Operational data centres should report their annual and seasonal water use.

A standard Water Usage Effectiveness measure could show the litres consumed for each unit of computing energy, but it should be accompanied by the source and local context.

A low figure in a water-stressed catchment can still matter more than a higher figure in an area with resilient supplies.

3. Potable water should be the last option

Highly treated drinking water should not automatically be used for industrial cooling.

Depending on local circumstances, alternatives might include:

• Recycled wastewater
• Treated effluent
• Captured rainwater
• Closed-loop cooling
• Direct-to-chip systems
• Alternative liquid coolants
• Seasonal hybrid cooling

Not every alternative will be appropriate everywhere, but developers should have to show that they have examined them.

4. Water availability must influence location

Planning policy already considers power connections, land, telecommunications and economic benefits.

Water should be treated as an equally important constraint.

Facilities without strict latency requirements could be directed towards cooler regions with stronger water resources and better access to renewable electricity, rather than continuing to cluster every development around London and the South East.

5. Developers should pay the full infrastructure cost

A large development should not consume the spare capacity intended to protect local homes and businesses from drought or equipment failure.

Where a facility requires new pipelines, treatment capacity, recycling infrastructure or reservoirs, the developer should make a proportionate contribution.

The environmental and financial costs should not simply be transferred to existing customers.

6. Britain needs a clear drought hierarchy

During an exceptional drought, who receives priority?

Hospitals would clearly rank above ornamental landscaping. But where do food production, manufacturing, data centres, schools, small businesses and household gardens fit?

The House of Lords Environment and Climate Change Committee has called for a clear emergency prioritisation plan. Without one, decisions may be made during a crisis, when political pressure is at its greatest.

Data centres have been designated as critical national infrastructure. That may be justified for facilities supporting banking, communications, healthcare and essential public services.

However, not every computing workload is equally critical.

A system supporting emergency communications should not necessarily be treated in the same way as one generating entertainment content or processing non-urgent AI training.

7. Water, electricity and heat must be planned together

Saving water by switching entirely to mechanical chilling may increase electricity consumption.

Reducing electricity demand through evaporation may increase water consumption.

The correct solution must consider both resources rather than moving the environmental burden from one system to another.

There are also opportunities to capture waste heat from servers and supply nearby homes, leisure centres, greenhouses or industrial processes. The government has identified heat networks as one potential benefit of well-located data centres.

That opportunity should be designed into projects from the beginning rather than added as an afterthought.

Could Data Centres Become Part of a Greener System?

A data centre does not have to be an environmental liability.

A well-designed facility could:

• Operate primarily on low-carbon electricity
• Use sealed liquid-cooling systems
• Avoid drinking water for routine cooling
• Capture rainwater
• Reuse treated wastewater
• Recover heat for surrounding buildings
• Reduce or move non-essential workloads during droughts
• Publish transparent environmental performance data
• Fund improvements to local energy and water infrastructure

The problem is not the existence of data centres.

The problem is constructing them in unsuitable places, with inappropriate cooling systems, based on incomplete forecasts and an assumption that unlimited water will somehow be available.

We Should Not Have to Choose Between Water and Technology

This debate should not become another false choice.

Britain does not have to decide between abandoning artificial intelligence and allowing unrestrained data-centre construction.

It can support digital growth while imposing clear environmental boundaries.

But that requires honesty.

Faster planning permission cannot manufacture water. Describing data centres as critical infrastructure does not create reservoir capacity. Economic ambition cannot repeal the physical limits of a catchment.

The government has now acknowledged the need for better data, improved transparency and closer coordination between departments, regulators and the water industry. It says a Water Delivery Taskforce is building the evidence needed for future planning and that a new national policy statement will address environmental requirements for nationally significant data centres.

That work must move quickly.

Applications are being considered now. Water-stressed regions are under pressure now. The infrastructure decisions made during this decade will affect communities and rivers for generations.

Conclusion: Green AI Must Begin With Honest Accounting

Artificial intelligence may help us solve some of society’s greatest challenges.

It may improve medical diagnosis, accelerate scientific discovery, optimise renewable-energy systems, detect leaks and help businesses work more efficiently.

But AI cannot be labelled green merely because its environmental costs take place inside a distant, windowless building.

The water is real.

The electricity is real.

The land, concrete, processors and cooling systems are real.

Water UK’s warning should not be used to create panic or hostility towards technology. It should force government and industry to replace assumptions with measurements and promises with practical infrastructure plans.

Britain can expand its AI capability without placing unacceptable pressure on homes and the environment—but only if water is considered from the first planning discussion, not after the servers have been ordered.

The future of artificial intelligence cannot be built on the assumption that the tap will always turn on.