Unlock the Editor’s Digest for free
Roula Khalaf, Editor of the FT, selects her favourite stories in this weekly newsletter.
Europe talks a lot about tech sovereignty, but has so far done little. Now France is taking the bull by the horns, with plans to build the continent’s biggest data centre. It is understandable to fear being beholden to the US for all things AI, but it’s not clear jumping on the infrastructure bandwagon is the best way forward.
The French plans are undoubtedly grand. At 5.1 gigawatts of power, the new facility would be a whisker above Meta Platforms’ proposed Hyperion campus in Louisiana and half the size of another upcoming mega campus in Ohio.
But real sovereignty is a pipe dream. The project is backed by Japanese money; tech group SoftBank is pledging €75bn. Given the total spending required for every 1GW of AI power is reckoned to be about $50bn-$60bn, that will need to be supplemented with another €170bn — at least. It is unlikely European investors will spring for the whole caboodle.
And there is no escaping Uncle Sam. The French cluster will be powered with US-designed chips. Given the size, cloud providers are likely to include US hyperscalers. Amazon, Microsoft and Google account for nearly three-quarters of that market in Europe. France’s OVHcloud, one of the biggest European operators, doesn’t disclose capacity but it’s certainly measured in mega rather than gigawatts.
This still beats being in hock to offshore data centres, and in that sense France has pulled off something of a coup. The UK, where OpenAI’s data centre plans were mothballed earlier this year, would have loved the SoftBank funds and kudos. But its energy is expensive, and space is scarce. The UK spreads roughly the same population over a landmass half the size of France, and revels in planning permits and other such red tape.
Still, it is too soon to claim victory. The project’s first phase is due in 2031, by which time SpaceX boss Elon Musk plans to have solar-powered data centres in orbit. Even with AI progressing at warp speed, it’s impossible to tell what capacity will be required when the 2030s kick off.
There is even a case to be made that the world may need fewer rather than more data centres than are currently planned, as the architecture of AI changes. China’s DeepSeek has proved AI needn’t be aggressively compute-intensive. More functions, such as cooling, are moving directly on to chips. And more AI “inference” is likely to move to devices like smartphones and laptops, as Nvidia’s latest laptop superchips indicate.
SoftBank is in good company when it comes to pouring money into land-based data centres — and France is shrewd to stake its claim — but that does not guarantee that what results won’t be un éléphant blanc.
[email protected]
