From the archive
31 Aug 2026
The word does a lot of quiet work. Cloud, network, algorithm: none of them suggest a building, and yet every prompt sent to an AI tool is processed inside one - a data centre running around the clock, drawing electricity and water in quantities that are only now being counted. Imogen Cripps, a researcher at Julie's Bicycle, has written the second in a series of essays asking what this infrastructure costs, and her starting point is that gap between the language and the concrete.
The figures she assembles are worth sitting with. Data centres consumed 1.5 per cent of global electricity in 2024, a share expected to double by 2030, with emissions projected to pass those of all global aviation. One research estimate puts AI's annual water use on a par with the entire world's bottled water consumption. The water goes to cooling, and when the buildings sit in drought-stressed regions - parts of Europe, this summer - it comes from the same sources as everyone else's, and often cannot be returned because of contamination.
The costs do not fall evenly, and the essay is careful to say where they land. Civil rights groups are taking Elon Musk's company xAI to court over a planned power station intended to run its data centres, acting for a mostly Black community in Mississippi whose members live near an existing site where 27 gas turbines are said to have been installed illegally, with serious health effects for residents. In Ireland, data centres built by American companies on peatlands have damaged a fragile ecosystem and cut local people off from a landscape that matters to them.
Then there is the number that is easy to miss. A recent study cited in the piece suggests that oil companies' use of AI to find new reserves could produce emissions 3.3 to 13.3 times higher than the pollution from powering the data centres themselves. In Scotland, proposed data centres threaten to displace planned renewable energy projects - the machines competing directly with the transition.
Cripps's argument to the arts sector is modest: that the sector increasingly adopting these tools should count all of it, the servers and what the servers are used for, before deciding what it can honestly justify. It is an essay rather than a project, and it offers no one to visit and nothing repaired - but the accounting it asks for is itself a kind of work.
If your organisation uses AI tools, put the digital estate in the same environmental policy as flights and freight: ask your provider where its data centres sit, whether they publish water and energy figures, and switch if they won't say. One afternoon, no cost, and the questions matter more than the answers.
Julie's Bicycle · Read the source