In the village of Broad Oak in Kent, the locals call the neighbouring landfill "The Mountain," and a faint smell drifts across the countryside from it. A developer named Era4 wants to turn that smell into a selling point. Its planning application proposes a data centre powered by methane captured from the decomposing rubbish next door, using the gas both to generate electricity and to help with cooling. It is an odd marriage of the lowest-tech industry imaginable with the highest, and it exists for a simple reason: Britain has decided it wants to be an AI superpower, and the electricity to do that is not where the servers need to be.
The scale of the ambition is real. The UK already has an estimated 450 to 500 data centres, and Barbour ABI counts around 120 more either under construction or working through planning, from the 71-hectare, one-gigawatt Elsham Wolds to a proposed billion-pound facility at Pinewood Studios, home of James Bond. Add up the capacity of just 68 of these and you get an estimated 63 terawatt-hours of demand a year. Britain's entire current consumption is 322 terawatt-hours. Not all of these projects will be built, but the direction is unmistakable.
The bottleneck is the grid. The National Energy System Operator says 140 data centres are now queued for a connection, representing roughly 50 gigawatts of capacity. To put that in perspective, Britain's peak electricity demand this year, on a cold February evening, was 45 gigawatts. The waiting list has run as long as fifteen years, though reforms may cut it to three to six. Slough, west of London, hosts 40 data centres, the largest cluster outside the United States, and has essentially run out of room; the overflow into neighbouring Hillingdon has collided with something voters care about even more than AI, which is housing. The council there says its local network hit capacity in 2022, delaying thousands of new homes, and it is now blunt that it will not trade houses for server halls.
Water is the other flashpoint. A petition against the Kent scheme has gathered thousands of signatures, borrowing protest templates from American campaigns, and the county has already been designated an area of water stress. The industry insists the fear outpaces the facts. Era4's design uses closed-loop liquid cooling, which recirculates sealed fluid rather than evaporating it away, so there is no discharge. Tech UK says 51 per cent of sites use no water beyond ordinary building needs and that more water is lost to leaks in a single day than the entire sector uses in a year. Both things can be true: the sector's thirst may be overstated, and a community running short of drinking water can still reasonably object to a newcomer that wants any at all.
What makes the story more than a planning row is the quiet argument breaking out over whether Britain even needs the giant boxes. The vast monolithic halls being thrown up in the US are built for training enormous models. But as the industry shifts from training to inference, from building models to running them, the useful hardware gets smaller and can sit closer to users. Amanda Brock of OpenUK suggests converting Britain's glut of empty high-street shops into small local data centres. A startup called Deep Green already pipes the waste heat from its servers into public swimming pools, warming the water in Exmouth and near Manchester for free. One sustainability researcher notes that you can now run a model as capable as GPT-4, state of the art only two years ago, on the phone in your pocket.
That is the tension worth watching. The demand is genuine, and it is now large enough to reshape energy markets on its own; in the US, the same AI boom has tech giants signing billion-dollar wind deals to feed their data centres. But the assumption that the answer must be ever-bigger buildings is starting to look less like a law of nature than a habit. A data centre that runs on rubbish gas and heats the local pool is a gimmick today. It may also be a preview of the only version of this build-out that communities are willing to live next to.