France taps nuclear power for new AI training cluster
Summary
Macron aims to dedicate a gigawatt of nuclear power to create one of the world’s largest AI-computing facilities.
France is making a bid to catch up in the artificial intelligence race by leaning on one of its strengths: plentiful nuclear power.
The French government plans Monday to pledge a gigawatt of nuclear power for a new artificial-intelligence computing project expected to cost tens of billions of dollars, according to its private-sector backers and the French government.
Combined with another newly announced French AI project funded by Middle Eastern investors that also aims for gigawatt scale, the plans would greatly expand Europe’s AI-computing capabilities to rival a vast expansion in the U.S.
The nuclear project, which aims to have a first tranche of 250 megawatts of power hooked up to AI-computing chips by the end of 2026, rivals the Stargate project in the U.S., backed by SoftBank and OpenAI. Stargate is starting with a campus in Texas initially fed by 200 megawatts of power, with plans to expand to 1.2 gigawatts.
FluidStack, the company spearheading the nuclear-powered AI cluster in France, said it aims to begin construction in the third quarter. Still, there is no guarantee its project will move forward as envisioned, or if it will secure enough money—or AI chips—to build it.
AI computing requires vast amounts of power as big tech companies shell out billions of dollars to build massive clusters of electricity-hungry chips. Those chips, mostly made by Nvidia, are the workhorses of the AI boom, performing the computations that underlie AI models.
Some of today’s most advanced AI models were trained at data centers with about 30 megawatts of electricity, the research group Epoch AI estimates. But by 2030, leading AI models may need more than 5 gigawatts of electricity—a Manhattan-sized amount.
The emergence of DeepSeek, a Chinese-made AI model purportedly built with far fewer chips than its competitors, raised doubt recently about the need for clusters of chips that have ballooned into the hundreds of thousands. But Nvidia said rapid AI advances would require ever more of its chips, and big tech companies are pushing forward with unprecedented spending on them.
To finance the first tranche of construction, FluidStack said it plans to deploy its own cash and secure loans for 10 billion euros, or $10.3 billion. It said talks are continuing with some of the world’s largest AI developers about using the new facility, which could house around 120,000 of Nvidia’s AI chips in its first phase and some 500,000 by 2028 if the site is fully built out. The company said it could further expand to a 10 gigawatt facility, 10 times larger, by 2030.
Much of the financing is necessary to purchase the chips, which have been in short supply in recent years amid the AI boom. AI-computing companies such as CoreWeave have pioneered new financing vehicles secured by Nvidia’s chips and contracts with AI developers to raise billions of dollars for new data centers, something FluidStack said it plans to do as well.
FluidStack said it is in regular contact with Nvidia about the project and has no worries about being able to finance or get access to the chips. “Nvidia has told me that they will send those chips when we need them," César Maklary, the company’s co-founder and president, said in an interview. Nvidia declined to comment.
The AI project furthers a longtime push from French President Emmanuel Macron, who has sought to expand France’s AI computing power. Last year, he told a closed-door roundtable of executives visiting to discuss investments in France that “low carbon and competitive electricity" is “one of our competitive advantages" for AI.
With a fleet of 57 reactors across 18 plants, France currently produces more than two-thirds of its electricity from nuclear power. France last year produced about a fifth more electricity than it consumed, exporting the rest.
If completed as envisioned, the FluidStack project could tip the balance of AI development toward France and Europe. Some European companies, including France’s Mistral AI, are part of AI’s vanguard, and the continent houses AI labs run by American tech giants that contribute to AI research. But Europe has largely been on the sidelines as big American companies—OpenAI, Microsoft, Oracle, Google and Nvidia among them—led the AI boom.
The FluidStack project aligns with that strategy, alongside another deal between France and the United Arab Emirates, announced on Thursday. Those countries agreed to begin investments with the aim of eventually creating an AI campus in France that would also use a gigawatt of electricity, something French officials said would cost tens of billions of dollars.
Write to Sam Schechner at Sam.Schechner@wsj.com and Asa Fitch at asa.fitch@wsj.com
