Data centers in France: the mirage of €75 billion.
France does indeed possess a considerable strategic advantage: its nuclear electricity, relatively abundant, controllable and low-carbon. But it risks turning that asset into a mere political communication argument rather than a real industrial advantage.
Recent announcements around SoftBank and French data centers are spectacular: up to €75 billion in investment, including a first phase of €45 billion to create 3.1 GW of AI data-center capacity in Hauts-de-France by 2031. SoftBank even announces a total ambition of 5 GW in France.
But this is precisely where the problem begins. In artificial intelligence, competition is not won with press releases. It is won with real time, installed power, connected GPUs, immediately available energy, fast permits, fast grid connections and a controlled industrial chain.
1. The €75 billion figure is impressive, but it hides a very slow timetable
The SoftBank announcement speaks of 5 GW of capacity in the long term, but the first phase targets 3.1 GW by 2031. That means France is celebrating today a power capacity that will not be fully operational for several years.
A capacity of 3.1 GW, if operated continuously all year, would represent
3.1 GW × 8,760 hours = 27,156 GWh per year, or about 27.2 TWh per year.
A capacity of 5 GW, also operating continuously, would represent
5 GW × 8,760 hours = 43,800 GWh per year, or about 43.8 TWh per year.
For comparison, RTE estimates that data centers already present in France consume several terawatt-hours per year, and that demand will rise sharply with artificial intelligence. The issue is therefore not only to announce capacity. It is to know whether France can authorize, connect, power, cool, staff and secure it in time.
2. France has electricity, but it has procedures that slow everything down
On paper, France’s nuclear base is an immense asset. In practice, every project must face land-use constraints, environmental procedures, appeals, connection delays, local negotiations, administrative uncertainty and a regulatory culture that often transforms industrial urgency into paperwork.
Meanwhile, other countries move faster. The United States, the Gulf states and parts of Asia are building compute capacity with a speed that France struggles to match. AI infrastructure is not a symbolic contest. The first to connect power, chips and talent gains a strategic lead.
3. The real danger: becoming a host country instead of a sovereign AI power
If France merely provides land and electricity to foreign giants while losing control over models, chips, cloud architecture, cybersecurity and data governance, it will not become an AI power. It will become a hosting platform.
A sovereign strategy would require much more: French and European compute capacity, secure cloud architecture, industrial partnerships under conditions, training of engineers, GPU procurement strategy, sovereign inference capacity, and public-sector AI infrastructure that does not depend entirely on foreign platforms.
4. A nuclear advantage must be transformed into an industrial doctrine
France’s nuclear electricity can become a decisive advantage only if it is connected to a clear doctrine: build quickly, authorize quickly, connect quickly, and keep strategic control. Otherwise, the country will once again possess the right asset and fail to turn it into power.
The question is therefore simple: will France use its energy advantage to build sovereignty, or will it merely rent it out to those who already dominate the digital world?