The AI Debate Has Reached The Water Supply

For years, artificial intelligence has largely been discussed as though it exists entirely in software.

The conversation has centred on models, algorithms and capabilities. We compare ChatGPT with Claude, debate which image generator is most convincing and speculate about what the next generation of AI might be able to do. What often disappears from view is the infrastructure that makes any of this possible.

Every AI response depends on servers running inside data centres. Those servers consume electricity, generate heat and require cooling. As AI adoption accelerates, that cooling has become an increasingly important part of the conversation because, in many cases, it relies on water.

That reality has started to surface in unexpected places.

During this summer’s hosepipe bans, social media filled with comparisons between restrictions on household water use and the growing demands of AI infrastructure. Some posts pointed to estimates suggesting AI conversations indirectly consume fresh water through data centre cooling. Others highlighted projections that UK data centres currently use around 6.6 million litres of water every day, with some forecasts suggesting this could rise to 42 million litres per day by 2030 as demand for AI infrastructure grows.

One widely shared graphic overlaid the locations of planned UK data centres with areas experiencing drought or prolonged dry weather, asking whether the two trends revealed a growing conflict between AI infrastructure and water resources. It was a striking comparison and helped fuel the debate, even if the relationship is more complex than a single map can show.

The comparison resonated because it felt intuitive. If households are being asked not to water their gardens, why should AI continue consuming increasing amounts of water?

Like many debates around AI, however, the comparison is both understandable and incomplete.

The growing demand from AI infrastructure is real; Modern data centres require enormous amounts of computing power, and computing power inevitably produces heat. Some facilities rely on evaporative cooling systems that consume fresh water, while others use air cooling or closed-loop systems designed to reduce consumption. The amount of water associated with an individual AI interaction depends on where that workload is processed, the local climate and the design of the facility itself.

The precise number attached to a single prompt is almost beside the point. The more important observation is that AI has environmental costs that are largely invisible to the people using it.

More Than A Water Debate

The discussion around hosepipe bans also highlights a second issue.

Restrictions on household water use are rarely introduced because there is simply no water left. More often, they reflect regional pressures on reservoirs, rivers and treatment systems following prolonged periods of dry weather. At the same time, the UK continues to lose vast quantities of treated drinking water through ageing infrastructure and leaks before it ever reaches consumers. These are different challenges. One reflects the growing resource requirements of digital infrastructure, while the other reflects decades of investment decisions surrounding public utilities.

On social media, however, the two often become intertwined because they create a powerful narrative. If one group is being asked to use less while another appears to be using more, the comparison feels like an obvious contradiction. Reality is rarely that straightforward.

Removing every AI data centre would not eliminate the need for hosepipe bans during prolonged droughts. Equally, replacing every leaking water main would not remove the environmental questions raised by the rapid expansion of AI infrastructure. Both conversations deserve attention, but they deserve to be understood on their own terms. Perhaps the more interesting development is what this debate reveals about AI itself.

Only a few years ago, AI was largely discussed as software. Increasingly, it’s becoming a question of national infrastructure. Governments are investing in semiconductor manufacturing, energy generation, data centres and sovereign computing capability because they recognise that AI depends on far more than models alone. Water now belongs in that conversation too.

That does not necessarily mean AI should consume less water than it does today, nor does it mean households are directly competing with language models every time they turn on a hosepipe. It does mean that decisions about AI are becoming inseparable from decisions about physical resources.

As AI becomes woven into everyday life, questions about energy, water, land and planning permission will become just as important as questions about model performance or regulation. The technology itself may continue to evolve at extraordinary speed, but it will always remain dependent on the physical systems that support it.

The debate over hosepipe bans is unlikely to determine the future of AI, but it may change the questions we ask about the infrastructure that makes AI possible.