Summary
- SoftBank plans up to 5GW of AI data centre capacity in France, with investment of up to €75bn.
- The first phase targets 3.1GW in Hauts-de-France by 2031 across Dunkirk, Bosquel, and Bouchain.
- The programme ties AI infrastructure to French power availability, industrial land, manufacturing, and sovereignty policy.
SoftBank Group has committed to develop and operate up to 5GW of AI data centre capacity in France, representing a potential investment of up to €75bn.
The first phase would deliver 3.1GW of AI data centre capacity in the Hauts-de-France region by 2031, backed by an initial €45bn investment. The planned sites are Dunkirk, Bosquel, and Bouchain. SoftBank said the programme forms part of the 2026 Choose France summit and is its largest AI infrastructure investment in Europe.
At Bosquel, SoftBank and Sesterce have formed a joint venture to develop and operate a 1GW AI data centre campus. At Bouchain, EDF has selected SoftBank as preferred bidder for a 400MW data centre at the former thermal power-plant site. At Dunkirk, SoftBank plans to work with Schneider Electric on a large industrial production cluster at the port.
The Dunkirk cluster is expected to include one SoftBank-operated facility to manufacture enclosures and one Schneider Electric facility to integrate data centre power modules. SoftBank said the arrangement is intended to localise and strengthen the supply chain for AI data centre infrastructure in France and Europe.
France sells power, land, and industrial policy
The scale of the SoftBank commitment makes France one of Europe’s most visible AI infrastructure battlegrounds. The country is leaning on low-carbon electricity, industrial land, state support, and manufacturing capacity to attract data centre investment that might otherwise concentrate in the United States, the Gulf, or established European hubs.
Power is the centre of the proposition. AI campuses at gigawatt scale cannot be treated as ordinary real estate projects. They require transmission access, substations, power-procurement strategy, backup arrangements, and confidence that the grid can support both initial build-out and later phases. France’s nuclear-heavy electricity system gives it a stronger story than many markets, but each project still needs connection agreements, local studies, and delivery programmes.
The site choices are telling. Dunkirk is already a major industrial and port location. Bouchain is a former power-station site with favourable grid conditions. Bosquel places capacity in Hauts-de-France, within reach of major European economic corridors. The common thread is industrial infrastructure rather than metropolitan prestige.
The manufacturing element also strengthens the programme. AI data centres need repeatable supply of enclosures, power modules, electrical systems, cooling plant, networking, and high-density compute integration. By pairing data centre campuses with Schneider Electric’s power module work, SoftBank is trying to address both capacity demand and the bottlenecks that slow delivery.
The plan carries substantial execution risk. A 5GW programme requires sustained capital, permitting, grid co-ordination, equipment supply, construction labour, and customer demand across multiple sites and partners. Large AI infrastructure announcements can outpace the practical work needed to energise capacity.
Even so, the French programme marks a significant shift in Europe’s data centre geography. AI demand is pulling capacity towards regions that can offer power and industrial sites, not just connectivity and cloud market depth. Countries are no longer only competing for cloud regions; they are competing for the energy-intensive industrial base of AI.
