AI Data Centres Are Draining Britain’s Water Supply — And You’re Paying the Price

Every time you type a question into ChatGPT, water evaporates somewhere in Britain.

Not a drip. Not a splash. According to researchers at the University of California Riverside, a single 100-word AI response consumes roughly 519ml of water — close to the volume of a standard drinks bottle. Do that 50 times in a working day and you’ve used more water than a quick shower. The difference is you never see it leave the tap.

This is the hidden cost of the AI boom. Data centres — the physical buildings packed with servers that power every AI tool, every cloud service, every streaming platform — use enormous quantities of water to stay cool. And as AI demand explodes across the UK, these buildings are being built at a pace that has water experts genuinely alarmed.

How Data Centres Use Water in the First Place

Servers generate intense heat. Left unchecked, that heat would destroy the hardware. To prevent this, data centres run cooling systems — and many of those systems use evaporative cooling, which works exactly like sweating. Water absorbs heat and evaporates, carrying the thermal energy away from the building.

The water that evaporates doesn’t go back into the supply. It’s gone. A single hyperscale data centre — the kind operated by Amazon, Microsoft or Google — can consume enough water in a single day to meet the needs of around 10,000 people. Not for drinking alone. For everything: showers, washing up, laundry, garden sprinklers, all of it.

There’s also an indirect water cost. Generating electricity requires water — for cooling power stations, for hydroelectric output, for steam turbines. So even data centres using air-cooled or liquid-cooled systems pull water from rivers, reservoirs and aquifers via the national grid.

How Much Water Does AI Really Use?

When Sam Altman, OpenAI’s chief executive, mentioned water consumption in 2025, he gave a figure of around 0.32ml per query. That’s for direct on-site cooling only.

The University of California Riverside study looked at the full picture — including the water consumed by the electricity generation that powers the servers. Their figure: 519ml per 100-word response. That’s a 1,600-fold difference, and it illustrates why corporate disclosures on environmental impact rarely tell the whole story.

Across all AI usage globally in 2025, data centres consumed an estimated 264 billion gallons of water. AI-related water withdrawals are projected to reach between 4.2 and 6.6 billion cubic metres per year by 2027. To put that in context: it’s more than the total annual water withdrawal of every UK household combined.

When I looked into this, I found the numbers surprising even after expecting them to be large. Training a single cutting-edge AI model — not running it, just training the initial version — can evaporate hundreds of thousands of litres of clean water. GPT-4 alone is estimated to have required millions of litres during training.

The UK’s Water Supply Is Already Under Strain

Britain’s relationship with water is getting worse, not better. Three of the past five years have included excessively dry periods. The Environment Agency warned in early 2026 that the year is on track to be the UK’s driest since 1976 — the year that brought standpipes in the streets and hosepipe bans across the country.

As of spring 2026, 14 water companies were still in drought or prolonged dry weather status. Anglian Water and South East Water have already imposed restrictions on certain customer uses. In Suffolk and Norfolk, housing developments have been blocked specifically because water supply infrastructure can’t support the growth.

The UK population has grown by around 10 million people since the 1990s, but water infrastructure investment hasn’t kept pace. Pipe leakage remains high — Thames Water alone loses roughly 630 million litres per day through leaking mains. Rivers in parts of England are running at record lows.

This is the backdrop against which the UK is now planning a massive expansion in data centre infrastructure.

84% of New Data Centres Are Being Built in the Wrong Places

This is the figure that should concern everyone. According to a 2026 report by Global Action Plan titled Not a Drop to Drink, 84% of proposed UK data centre developments are planned for areas that are already water stressed — or are projected to become water stressed by 2040.

The Thames Water service area holds around 80% of the UK’s existing data centres, with approximately 100 more currently proposed for the same region. The south-east of England — which regularly experiences the driest conditions in the UK — is becoming the country’s primary data centre hub, largely because of its proximity to London’s business and financial infrastructure.

The irony is stark. The areas most likely to face hosepipe bans and restrictions for ordinary households are precisely the areas being targeted for the most water-intensive industrial facilities in modern history.

Big Tech’s Numbers Don’t Add Up

Corporate disclosures on water use have been incomplete at best. Microsoft reported its global water consumption rose 34% in a single financial year — from around 4.7 billion litres to 6.4 billion litres — as AI infrastructure ramped up. Google disclosed 5.6 billion gallons globally in its 2023 environmental report.

What both companies don’t consistently break out is which specific facilities are in which specific catchment areas — meaning it’s nearly impossible for regulators or communities to understand which local water systems are under the most pressure.

The UK Government’s own report on water use in AI and data centres, published in June 2026, acknowledged a critical gap: data centres are not currently required to disclose their water usage at all. Unlike energy consumption, which has some disclosure requirements baked into planning permissions, water use is largely invisible.

What the Government Is (and Isn’t) Doing

The UK Government announced in 2025 and 2026 that it views data centre expansion as a national strategic priority. Tax incentives, planning fast-tracks, and infrastructure investments have all been directed at attracting hyperscale operators to Britain.

The European Commission issued guidance in early 2026 suggesting that data centre waste heat could be repurposed for water purification and carbon capture — pointing to emerging technologies that could change the equation. Closed-loop liquid cooling reduces direct water consumption by 70–90% compared to evaporative systems, though it costs significantly more to install.

Environmental groups including Global Action Plan are calling for mandatory water use disclosures as part of planning applications, along with water efficiency standards similar to those applied to other industrial uses. The UK Water Industry Research body has flagged the need for data centre operators to appear in drought contingency planning — something that currently doesn’t happen. As of June 2026, none of these changes have been adopted into law.

What AI Water Use Means for Your Water Bill

Water companies in the UK have already won approval for significant bill increases. Ofwat approved a 36% average bill rise for the period 2025–2030, partly to fund infrastructure upgrades and leakage reduction. Thames Water alone has proposed further increases beyond that baseline.

When water systems come under additional strain from industrial users — without any parallel requirement on those users to pay proportionally into infrastructure investment — the cost tends to be borne by households. It’s the same dynamic seen with energy network costs: industrial-scale demand, but diffuse pricing that ends up on residential bills.

Data centre operators typically pay commercial water tariffs, but these don’t necessarily reflect the true cost of water abstraction from stressed catchments. Campaigners argue that the current pricing structure means ordinary households are effectively subsidising the water footprint of AI.

What This Means for You

The core message here isn’t that you should stop using AI tools. That would be impractical, and the individual impact of your ChatGPT queries is not the real problem.

The problem is infrastructure planning. Big decisions are being made right now about where to build data centres, how they’re cooled, and what disclosures they’re required to make. Those decisions will determine how much strain is added to already-stressed water systems across the UK over the next decade.

What UK residents can do:

• Write to your MP about mandatory water disclosure requirements for data centres — planning reform bills currently in parliament are a live opportunity
• Pay attention to Ofwat consultations on water pricing — the next review sets tariff structures that determine how costs are distributed
• Support environmental groups pushing for water efficiency standards in data centre planning permission
• Ask your employer or local council whether any data centre developments near you have gone through water impact assessment

AI is not going away. But the version of AI expansion that demands no accountability for water use, and builds in the most water-stressed corners of England, doesn’t have to be the one we get.

This article is for educational purposes only and does not constitute financial advice.