A note from our political editor, Zoë Grünewald: This summer, while Londoners were told to take shorter showers and let their gardens go brown, the AI industry was planning a building boom.

Data centres in the pipeline across London and the South East would add large scale new demand to the grid – and a substantial water bill.

Today, we’re launching a look at Britain’s data centre sector, led by Anne Alexander, a journalist who specialises in reporting on the AI industry.

Over the coming weeks and months, we’ll look at who is building these facilities, what they cost the places that host them, and why the conversation about “controlling AI” keeps missing the most concrete harm.

Anne starts today with an argument for a kill-switch: not for rogue robots, but for Shrimp Jesus memes during a heatwave.

cable network — Photo by Taylor Vick on Unsplash

There has been much discussion in the media, in the boardrooms of tech companies and in the corridors of power about whether it is either desirable or even possible to “control AI”. Opinion is sharply divided. Dario Amodei, the CEO of Anthropic, published a wordy essay on the topic. Trump doubled down on cheerleading an accelerationist approach to the technology arms race while claiming to have Xi Jinping’s approval for renaming AI “super intelligence.”

For some, any prospect of opening a conversation about regulating or holding the AI industry to public account for the vast resources being consumed in its quest to build “intelligent machines” is welcome.

But there is also danger of distraction from the AI industry’s role in exacerbating the genuine “existential threat” we face: climate and ecological breakdown.

Don’t get me wrong: I’m deeply concerned about what the summer’s spree of so-called “rogue agents” breaking things online - including exploiting a vulnerability to access private health statistics from an Australian government website.

That isn’t because I think that there is any evidence of artificial consciousness. The “agents” concerned are programmes executing code based on text generated by Large Language Models (LLMs). These are better thought of as a kind of statistical typewriter than an intelligent lifeform. Of course there is some “magic” in there: but only the kind achieved by trillions of calculations to generate predictive text and random numbers. In other words, the shiny aura of money, machines and power.

And yes, having an “off-switch” for machines is generally a good idea. Suggesting otherwise would be a bit like 19th century engineers arguing that you shouldn’t add brakes to a steam locomotive in case its feelings were hurt. The Victorians went through decades of catastrophic railway accidents without pretending that the engines “went rogue” and eventually legislation start to hold railway companies to account for their atrocious disregard of public safety.

The conversation we need to start having however, is about how embedding AI products into everyday tasks inevitably results in more power and water use.

Travelling around London this summer, it has been hard to avoid adverts urging adults to “save water” for the kids. I’m completely in favour of fixing leaky cisterns and taking short showers. But while ordinary people have watched their gardens shrivel, the data centre and AI industry is galloping ahead with a construction pipeline which if it is all built out will add up to 5 GW of new electricity demand to the grid across London and the South East.

Producing electricity is a highly water-intensive industry in the UK, according to experts at UN’s University Institute for Water, Environment and Health. In a report published earlier this year, they calculate that running AI models on servers in the UK will use on average 20 litres per kilowatt hour just at the power generation source, before even a drop comes anywhere near a data centre cooling system.

The same report highlights how the routine use of AI models to generate images and video or carry out internet searches risks raising the power and water footprint of everyday information tasks exponentially. Generating a typical AI image requires roughly 1450 times more energy than an old-style internet search using text classification methods.

During the heatwave in 2025 the Environment Agency recognised that old emails taking up storage space in data centres add stress to overstretched water and power systems, with an appeal for water-conscious users to delete them.

Share

But the spiralling energy and water costs of AI tools could be more easily controlled from the top, by the companies which make them, rather just nudging consumer behaviour in the right direction.

Luckily there are some quite simple methods to achieve this. Turning off access to image and video generation based on the users’ IP address ought to work. Google already claims to take into account local water stress when deciding what kind of cooling system to install in its new data centres, so why not add in settings which are responsive to heatwaves and droughts?

Subscribe now

In reality it will take a mixture of approaches to rein in an entire industry which has “gone rogue.”

A “kill-switch” for Shrimp Jesus memes in the middle of heatwave would be a start, but it needs to be connected to much broader campaigns for accountability. A moratorium on new data centre construction in the UK would be another vital step.

In a world already reeling from the impact of extreme heat we cannot afford to wait. We need a fundamental rethink of the relationship between Big Tech companies and their host societies here on planet earth. ■

Leave a comment

About the author: Anne Alexander is a journalist and author specialising in reporting on the AI industry. She is currently working on a project investigating the UK’s data centre sector with Watershed Investigations. You can read more of her work on Substack here.

👫Agree with Anne? Share this story with your friends, family and colleagues to help us reach more people with our independent journalism, always with a focus on people, policy and place.

Share

Make sure you’re signed up to get the next instalments of our reporting on AI data centres and we’ll be back in your inbox on Tuesday with the latest from our political editor Zoë Grünewald

Subscribe now

Why we need a "kill-switch" for Shrimp Jesus