SoftBank to Invest Up to 75 Billion Euros in AI Data Centers in France
On 30 May 2026, SoftBank announced plans to invest up to 75 billion euros (around 87 billion US dollars) (SoftBank press release, 31 May 2026) to build AI data centers in France. The plan targets up to 5 gigawatts of additional data center capacity. The announcement came at President Emmanuel Macron’s “Choose France” summit — and is, according to SoftBank, the company’s largest AI infrastructure investment in Europe. For an industry where compute has become the scarce resource, this redraws the map: a central part of Europe’s AI infrastructure will now take shape in northern France.
What was announced
- Up to 75 billion euros (around 87 billion US dollars) in total investment for AI data centers in France (SoftBank, 31 May 2026).
- Target: up to 5 gigawatts of additional data center capacity.
- Phase 1: 45 billion euros for 3.1 gigawatts in the Hauts-de-France region by 2031 (CNBC, 31 May 2026).
- Phase 1 sites: Dunkirk (Loon-Plage), Bosquel and Bouchain.
- Partners: EDF (former power-plant site in Bouchain) and Schneider Electric (industrial cluster at the Port of Dunkirk).
What applied before
The largest AI data center projects of recent years were clearly concentrated in the US — driven by hyperscalers and compute deals such as the one between Anthropic and SpaceX around the Colossus cluster. By comparison, Europe was seen as a laggard: less capital, fragmented power markets, slower permitting. Anyone wanting to train or operate large models at scale fell back on US capacity.
France had movement of its own — Mistral’s 722-million-euro data center near Paris, for instance — but at a different order of magnitude. Such projects covered individual providers’ own demand, but did not define a continental infrastructure layer that multiple players could draw on.
What applies now
1. A new order of magnitude for Europe. At up to 75 billion euros, the SoftBank project plays in a league previously reserved mostly for US projects. The planned 5 GW are a multiple of what individual European providers have built so far. This is not a self-supply data center but an infrastructure block that noticeably raises the AI capacity available on the continent.
2. Phase 1 is concrete and dated. The first stage covers 45 billion euros for 3.1 GW in the Hauts-de-France region, finished by 2031 (CNBC). The sites are fixed: Dunkirk (Loon-Plage), Bosquel and Bouchain. Unlike vague letters of intent, this names amounts, capacity, region and time horizon.
3. Power is the real argument. SoftBank is working with EDF, which provides a former power-plant site in Bouchain, and with Schneider Electric for an industrial cluster at the Port of Dunkirk. The choice of France is no accident: the country draws roughly 70% of its electricity from nuclear power, is the world’s largest net electricity exporter, and posts industrial power prices well below those of many neighbours. It is exactly this stable, cheap base load that is the limiting factor for AI data centers.
Context
The news is less a single event than another point on a line: Europe is building AI compute capacity, and France is deliberately positioning itself as the hub. Mistral’s data center near Paris, SoftBank’s block in Hauts-de-France, and the international compute deals of the model providers all show the same pattern — the bottleneck of the AI economy is no longer the model, but power and floor space.
In practice, this means two things above all. First: the question “where does AI capacity emerge in Europe” gets a concrete answer — northern France, carried by cheap nuclear power. Second, caution is warranted. This is a letter of intent over a horizon stretching to 2031; part of the figure is framed as “up to”, and large projects of this kind tend to slip in both timeline and final capacity. Anyone deriving concrete availability for themselves today should distinguish between announced and built capacity.
Regardless of the exact final figure, the strategic point stands: power is becoming the key resource in the AI race, and location decisions increasingly follow the electricity price, not proximity to developer hubs.
What you can do now
If you are planning AI capacity in Europe: treat the announcement as a trend signal, not as available supply. Phase 1 is dated to 2031 — it does not create short-term capacity.
If you assess infrastructure risk: watch the power angle. Choosing France shows that cheap, stable base load is becoming the decisive factor — relevant for any cost calculation around AI operations.
If you follow the European AI landscape: read this in the context of the other infrastructure news (see the linked stories on Mistral and the Anthropic-SpaceX deal) to place the “compute as bottleneck” pattern.
Entdecke mehr
Three Days with Claude Fable 5 — and Why Opus Stays My Daily Driver
Fable 5 fixed a bug in one shot that Opus failed at twice. I'm still not switching. My honest impressions from the trenches after three days.
GlossarMiniMax
MiniMax is an AI lab founded in Shanghai in 2021 whose open-weight M model family (LLMs) targets very long context windows and low running costs via sparse attention.
LexikonComparing AI models — who builds what and how to choose
The major model families in 2026 at a glance. Who builds Claude, GPT, Gemini, Llama, Mistral, DeepSeek, Qwen — and which model to pick when.