Will artificial intelligence use all our electricity?

France 24

02-05-2025

The city of Espoo lies 200 km west of Helsinki. The panorama around the city is what you might expect of Finland – evergreen forests, crisp lakes and ice thawing at the end of winter. But the skyline is about to be occupied by another kind of building: a data centre.
Excavators and all sorts of machines rumble on in the background under the watchful eye of Alistair Speirs, Senior Director of Azure Global Infrastructure at Microsoft.
'A data centre is the home of the cloud. It's an industrial facility that holds thousands and thousands of servers. It's powering all the cloud services that we use for work, for home, gaming, streaming, financial services, education and even medicine,' he says, pointing at the construction site behind him.
This future data centre campus aims to be one of the most innovative yet for the American tech giant. 'This data centre will be one of the most sustainable data centres in Europe,' the engineer explains. ' Excess heat that's generated from the data centre will go on to heat around 250,000 houses in the area.'
It was this promise that made the project so attractive to local authorities, who have welcomed data centres on their soil. ' This data centre will be our main source of clean heating,' says Kai Mykkänen, mayor of Espoo and Finland's former Minister of Climate and the Environment.
The vast, sparsely populated areas and long, cold winters make Finland an ideal location for these behemoths, which have to install energy- and water-hungry cooling mechanisms in warmer climates. What's more, the local electricity grid is well equipped. 'Finland has a huge amount of clean electricity. We've been able to scale up electricity production during the past ten years from non-fossil sources, better than in most areas in Europe. And we have a very strong national grid connection. So it was possible to find a location where there is a strong grid, enough electricity, and then also a need for this excess heat,' says the mayor.
But these data centres are not so easy to host all over Europe. In Ireland, for example, in 2023, the country was forced to halt the construction of new centres in the Greater Dublin area. In one year, the existing data centres had consumed as much electricity as the whole country's urban housing.
European startups join the race for energy
Not all energy networks are adapted to such a high electricity demand. And it's not going to get any easier. In its latest report, dated 10 April 2025, the International Energy Agency (IEA) estimates that by 2030, data centre global electricity consumption will have doubled, and even quadrupled for AI-optimised data centres. So how can we cope with such a surge in demand, especially at a time when Europe has pledged to reduce its carbon emissions by half compared to 1990, to achieve its goal of being climate-neutral by 2050?
For many tech companies, a lot of the energetic cost comes from developing and training AI models. Some European players are trying to optimise the resources they have to decrease the electricity bill. That's the case of Kyutai, a French AI research lab. In their Parisian office, facing Les Halles and the Bourse de Commerce, Kyutai developers work in open source: they make their research available so that others wanting to replicate their models can access it, use it, and avoid repeating the same processes of trial and error they went through. And that saves energy.
'Our aim with open source is to ensure that the cost of training is worth it, by reducing the cost of experimentation for other structures, whether they're start-ups or other competing labs,' says Alexandre Défossez, Chief Exploration Officer at the startup, tapping out lines of code on his computer.
'Some big American companies are focused on performance at all costs. Us European players, we have access to and use far fewer computing resources than the Americans. So that also forces us to be a bit clever about what we do,' he adds.
Thirty-five nuclear power plants to cover our needs
Engineer Marlène de Bank is focused on finding solutions to reconciling AI energy consumption with the climate transition. De Bank researches digital tech at The Shift Project, a French think tank working on decarbonising the economy.
'The demand for AI at European level is 35 gigawatts. 1 gigawatt is one nuclear power plant. So we're going to have to add the equivalent of 35 nuclear power stations. If it's something else, that means we'll have to add 35 times that equivalent. We also have to find it,' she explains.
The think tank is currently drawing up a report assessing the impact of AI on the climate transition. At the moment, they are undecided as to whether AI is entirely good or bad for the environment. AI can help us do things like prevent natural disasters or manage waste, but it can also help optimise oil extraction.
De Bank believes that a lot depends on what AI is used for. 'It's like a car. To travel one kilometre, you can choose to bike instead. So your carbon emissions at the end of the year depend on how many times you've driven and how many times you've cycled. For AI, it's going to depend on how often you use AI and how often you use your brain instead.'