SoftBank Group announced on May 31, 2026 that it will develop and operate 5 GW of AI data center capacity in France, investing up to €75 billion across multiple sites including Dunkirk, Bosquel and Bouchain. The first phase commits €45 billion to deliver 3.1 GW of AI data center capacity in the Hauts‑de‑France region by 2031, in partnership with EDF and Schneider Electric.
This article aggregates reporting from 6 news sources. The TL;DR is AI-generated from original reporting. Race to AGI's analysis provides editorial context on implications for AGI development.
SoftBank’s €75 billion commitment to build 5 GW of AI data center capacity in France is one of the clearest signs yet that compute, not algorithms, is becoming the main bottleneck in the race to AGI. By effectively underwriting an entire layer of European AI infrastructure, SoftBank is shifting from its Vision Fund-era strategy of backing software startups to owning the physical substrate that all frontier models will run on. This is a power move: whoever controls multi‑gigawatt campuses in stable, well‑connected regions controls a scarce input that every major lab and hyperscaler needs. ([group.softbank](https://group.softbank/en/news/press/20260531_0))
For Europe, the deal partially closes the gap with US and Gulf-backed mega‑campuses, and aligns tightly with Paris and Brussels’ push for ‘technological sovereignty’ in AI. Locating the first 3.1 GW in Hauts‑de‑France and anchoring a manufacturing cluster with Schneider Electric in Dunkirk also localizes parts of the AI hardware supply chain that have been overwhelmingly US- and Asia-centric. ([group.softbank](https://group.softbank/en/news/press/20260531_0))
Strategically, this increases SoftBank’s leverage with frontier labs like OpenAI and Anthropic, which are hungry for cheap, reliable compute but wary of over‑reliance on US cloud incumbents. If SoftBank succeeds in stitching together French power, Japanese capital and global GPU supply into a cohesive platform, it could become a kingmaker in who gets to scale to the next generation of AGI‑class models.


