Not a drop to drink: How Britain's AI data centre surge threatens water security

Not a drop to drink: How Britain's AI data centre surge threatens water security is a report published by Global Action Plan in April 2026 and co-authored by Oliver Hayes, GAP's Head of Big Tech and Campaigns, with Nicole Sugerman, Campaign Manager at the U.S. climate-and-Big-Tech organisation Kairos. It is the water-side evidence base of the GAP / Foxglove campaign on UK hyperscale data centres — the Foxglove / Global Action Plan challenge to UK hyperscale data centres — and was released into a UK policy moment in which the government's "AI Growth Zones" agenda was actively accelerating planning approvals for hyperscale facilities.

The report's central claim is that the water footprint of AI data centres in the UK is much larger than commonly acknowledged and that the build-out is being concentrated in precisely the parts of England that can least absorb it. Its headline finding — derived from cross-referencing publicly known proposed data-centre developments with the Environment Agency's 2021 classification of water-stressed areas — is that 84% of proposed water-intensive UK data-centre developments are planned for areas that are already water-stressed or are projected to be water-stressed by 2040. The report walks through the supporting numbers: a single hyperscale data centre requires daily water equivalent to the needs of around 10,000 people; training a single cutting-edge AI model can evaporate hundreds of thousands of litres of clean water; global AI-related water withdrawals could reach 4.2–6.6 billion cubic metres a year by 2027, exceeding annual UK household consumption; Slough alone hosts 32 operational data centres in a "seriously water-stressed" area; and under high-growth scenarios data centres could account for around 30% of new water demand in the Thames Water region, equivalent to 270 million litres a day. The argument is built on the contrast between the conservation expected of UK households and farmers — three of the past five years having experienced excessive dry periods — and the comparative absence of water-side disclosure, regulation, or environmental impact assessment of hyperscale facilities whose indirect water consumption from electricity generation is itself poorly tracked.

Within the corpus, Not a drop to drink is the water-side counterpart to the carbon framing that the GAP / Foxglove campaign had previously led with — Foxglove's October 2025 estimate that ten UK hyperscale data centres would generate roughly 2.7 million tonnes of CO2 a year — and it broadens the campaign's evidence base from climate-and-grid arguments into water-security arguments aimed at a sector (UK water utilities, environmental regulators, Thames Water region MPs) the carbon framing did not reach directly. It is therefore an example of a movement-strategic Publication that exists not on its own terms but as the analytical companion to an active strategic-litigation campaign: the January 2026 Government Legal Department concession on the Woodlands Park data centre had already established that ministerial approvals could be quashed for failing to require Environmental Impact Assessment screening, and the report supplies the empirical case for what such EIA scrutiny would actually need to consider once water is on the table. The Hayes-Sugerman co-authorship also marks a transatlantic strand the corpus had not previously recorded: a UK environmental charity and a U.S. campaigning organisation jointly authoring the evidence base for a UK campaign, on a shared analysis that AI infrastructure is being built into water-stressed catchments without disclosure or accountability commensurate with the scale of the demand it imposes.