6,000 People Liked My Snarky Comment About AI Water Use and the Drought. But Were We Right?

It’s an image Brits have gotten used to this summer: our once “green and pleasant land” reduced to a crisp, arid vista. The once-verdant grass of England’s parks and commons scorched yellow and brown after weeks of heat and barely any rain.

The Evening Standard recently posted a carousel of photographs showing the impact of England’s exceptionally dry summer on London. I left a fairly flippant comment underneath: “But by all means, build a massive data centre in east London.”

More than 6,000 people liked it, so apparently I wasn’t the only person thinking about the curious clash between the UK’s hottest summer on record and the AI boom.

Since then, the situation has become even starker. England and Wales officially recorded their driest July since records began in 1836, while those of us in southern England experienced the driest month of any kind on record. As of 10th August, 71.3% of England was officially in drought, affecting 45 million people, while 27 million were living under water-use restrictions.

Temporary-use water bans have meant paddling pools being relegated back to the garage, cars going uncleaned and gardens unwatered. And don’t even think about a BBQ: the dehydrated earth is primed for wildfire. I’d hazard a guess that most of us are willing to forgo these favoured summer pastimes for the sake of saving water and staying safe.

But it does seem curious that, as bans are introduced across parts of the country and water companies implement drought plans and ask customers to curb demand, Britain is racing to build the infrastructure required to turn England into an AI powerhouse.

It made me wonder whether my sarcy Instagram comment was actually fair. Just how thirsty is AI? And should we be thinking more seriously about what our increasingly ubiquitous use of it is costing? I decided to find out.

AI Water Use: Wait, Just How Thirsty Is AI?

I think part of the problem is that our relationship with AI feels almost entirely intangible. I can type a question into ChatGPT, ask it to summarise a document or generate an image and, seconds later, an answer appears on my screen, as if by magic. What I don't see, and you don't either, is the enormous physical infrastructure needed to make that exchange possible.

Every prompt has to be processed somewhere. Data centres house thousands of servers which generate heat as they work, requiring extensive cooling systems to keep them operating safely. Depending on the location and cooling system used, that can require water — sometimes vast quantities of it. A House of Commons Library briefing notes that there aren't even official figures for how much water UK data centres collectively consume, because they aren't required to report it.

"Water is mostly used to cool the data centres. That is, to remove the heat from the computers (mostly GPU's) used to compute the responses to queries for ChatGPT," explains Andrew A. Chien, William Eckhardt Distinguished Service Professor in Computer Science at the University of Chicago.

He explains that generative AI is also more resource-intensive than many of the digital activities we're used to. "The main reason to worry about AI is that each request consumes much more power (vs traditional email, Snapchat, or Facebook feed). This means it consumes correspondingly more water."

What then, about all those viral claims translating ChatGPT prompts into bottles of water? Chien says it's complicated. "This is difficult because it depends on exactly which datacenter is used, and a lot of technical detail about the exact AI models." However, he adds: "I've seen good estimates as high as 50 ml of water per prompt."

That doesn't mean every inane ChatGPT query is draining a reservoir somewhere. Water consumption varies enormously according to where a data centre is located and how it is cooled. Yet despite AI increasingly infiltrating our daily lives, most of us have precious little idea what it takes to keep it running. Chien notes that AI is now "being embedded in nearly every computing activity" and warns that, while efficiency is improving, "we're still likely to see increasing total water consumption by AI."

And we're building more of them?

At the same time, Britain is going all-in on AI. The government has made becoming an “AI maker, not an AI taker” a central ambition and has identified data centres as critical national infrastructure. But the AI boom needs somewhere physical to happen, which means building more — and bigger — data centres.

Including in London. Earlier this year, plans were approved to transform part of the Old Truman Brewery site on Brick Lane — where, just a few weeks ago, I met emerging designers as part of Graduate Fashion Week — into a large data centre, despite objections from some residents and campaigners. The proposed development would demolish and redevelop buildings on the site to create around 5,200 square metres of data-centre floorspace, alongside offices.

A large data centre doesn't automatically mean enormous water consumption — how thirsty a facility is depends heavily on where it is built and how it is cooled — but the concentration of new developments in and around London is still significant.

“The concern isn't that data centres as a whole are big water users today; most aren't. It's that we're approving a wave of new capacity in exactly the wrong places without properly planning for it,” says Stephen Slessor, Chair of British Water. “Around 84% of new UK data centre developments are earmarked for areas already water stressed or projected to be by 2040, and the majority of existing and planned sites cluster around London and the South East, the most water stressed part of the country.”

And the potential scale of that demand is not to be scoffed at. According to Slessor, “Thames Water estimates data centres could drive around 30% of all new water demand in its region. A single hyperscale facility can use as much water in a day as around 10,000 people.” The timing matters too: “the hybrid cooling systems many data centres rely on draw the most water on hot, dry days, which is exactly when household demand also peaks and rivers and reservoirs are under the most strain.”

Dr Eduardo Rico Carranza, whose work on the water-system impacts of data centres was submitted to Parliament, is also concerned. Asked whether we should worry about expanding data-centre capacity in water-stressed areas such as London and the South East, his answer is unequivocal: “Very. Our numbers indicate that the future demand cannot be met by the existing water infrastructure in place, especially during hot seasons.”

That doesn't mean AI is about to make water stop coming out of Londoners' taps. Carranza says domestic supply is likely to be prioritised. But he warns that there could be other consequences: future housing developments struggling to secure sufficient water, more frequent hosepipe bans and new infrastructure being required to meet data-centre demand — the cost of which could ultimately find its way onto consumer bills.

Which brings me back to my Instagram comment: as Britain races to build the infrastructure required for these AI aims, who is making sure computing power doesn't put further pressure on an already strained water supply?

Why are we the ones being asked to cut back?

The thing which confuses me isn't just that AI uses water. Almost everything we do has an environmental cost. It's the seemingly odd imbalance in who is expected to think about it.

As citizens, we're constantly reminded to monitor our own consumption. Take shorter showers, turn off the tap when brushing your teeth, don’t ask for fresh towels on holiday; so reads the sign in the hotel bathroom. These aren't unreasonable requests, but they do make the speed at which AI is being rolled into our lives feel jarring.

Sini Matikainen, Economic and Fiscal Policy Director at LSE’s Centre for Economic Transition Expertise, thinks the conversation between private and public costs is overdue. “We should be having a transparent conversation as a society about the costs associated with this infrastructure, instead of letting large private companies dictate the pace of development.”

Matikainen explains that data centres are voracious consumers of electricity, too. To power even a mid-range estimate of Britain's future data-centre demand entirely with onshore wind would require around 600,000 hectares of wind farms — enough land to cover Luxembourg two-and-a-half times over. Meanwhile, in the US, 11 planned data-centre campuses are expected to produce more greenhouse-gas emissions than the entire country of Morocco.

She questions, too, what the AI boom might look like if those costs weren't so easily externalised. “If these companies actually had to bear the full costs — including paying for the cost of emissions and other environmental impacts — I think we’d see AI being rolled out much more selectively.”

Slessor argues that this isn't simply an environmental issue, but one of fairness. “This is the fairness question, and it's the single biggest risk to public support for conservation. Research at the Environment Agency has consistently found that people back hosepipe bans and are willing to do their bit, but that trust collapses if they don't see companies fixing leaks and cutting their own and other big users' non-essential use first.”

There’s an understandable excitement (and apprehension) about what AI might allow us to do, yet its environmental cost doesn’t crop up as much as we might hope. As Slessor puts it: “Asking households to conserve while a fast-growing, largely unregulated industrial sector expands unchecked in the same water-stressed catchments isn't a sustainable message. If we want the public to keep buying into water efficiency, and we need them to, industrial users including data centres need to be seen visibly pulling their weight too.”

Leslie Morris-Iveson, international water expert and Senior Research Fellow at Chatham House, argues that government therefore needs to establish “clear environmental and social guardrails” around its AI plans. As she puts it: “There are going to be tradeoffs, and I think we as a society need to start defining what uses of AI are important to us, and where AI will move us to a more prosperous future for all, based on the evidence and realistic planning and forecasting.”

I'm not suggesting we stop taking shorter showers or start watering our lawns with wild abandon. But if water is becoming scarce enough that ordinary people are being asked to change their behaviour, shouldn't the companies building the technology transforming our lives — and the governments encouraging them — be expected to account for their consumption just as visibly?

So, was my snarky comment right?

Not exactly. AI didn't cause London's drought, and not all data centres are equally thirsty. Where they're built, how they're cooled and the technology they use can massively change their water footprint.

As Leslie Morris-Iveson puts it, “What gets lost in these debates is that water is entirely local. A data centre in an area with abundant water might not be a problem, but placing high-volume cooling in an already water-stressed catchment during a summer drought puts AI use directly in competition with local homes and farmers.”

So, no, my comment didn't tell the whole story; after all, it was an Instagram comment that, as another commenter put it: “You’re posting on a page that’s owned by a company that owns the biggest data centres on the planet!” The irony was not lost on me, but I don't think that makes the 6,000 people who liked it wrong, either.

What I wanted to highlight — and what I think 6,000+ people resonated with — was the contradiction. We're being warned that resources are becoming scarcer and asked to change our individual behaviour accordingly, all the while watching governments race to accommodate the rapid expansion of AI, the technology for which, and its physical footprint, most of us barely understand.

Sini Matikainen puts the hierarchy of priorities plainly: “Access to water is a fundamental right for human beings, not data centres.”

I'm not about to stop turning the tap off while I brush my teeth. Nor am I about to stop using AI altogether; it’s already embedded into so many of the apps I use. But I think before we ask AI another question, we should be asking a few more questions about AI.