Our Take
France wins a bet SoftBank is willing to place; the real test is whether 3.1 GW by 2031 keeps pace with actual demand or sits half-full by then.
Why it matters
Data center capacity is now explicitly a national industrial policy prize, not just a utility decision. France's regulatory environment and energy infrastructure are proven attractive to the world's largest infrastructure investor.
Do this week
Infrastructure leads: map SoftBank's Hauts-de-France sites against your latency requirements and power contracts before Q4 2026 to lock reserved capacity early.
SoftBank commits €75B to French data center expansion
SoftBank Group announced on May 30 that it will invest up to €75 billion (approximately $87 billion, per TechCrunch) to build and operate data center capacity in France. The target is 5 gigawatts of additional capacity total.
The first phase focuses on three sites in the Hauts-de-France region: Dunkirk (Loon-Plage), Bosquel, and Bouchain. These facilities are scheduled to deliver 3.1 gigawatts of capacity by 2031. SoftBank describes this as its largest AI infrastructure investment in Europe.
The company is simultaneously both an investor in and customer of OpenAI, making this commitment partly a bet on sustained demand for large-scale model training and inference.
French economic minister Roland Lescure framed the announcement in terms of national AI leadership: "a testament to President Emmanuel Macron's ambition to position France as a leading destination all along the AI value chain" (per TechCrunch).
Capacity commitments now anchor national AI strategy
Data center expansion has become a visible proxy for AI industrial policy. The U.S. debate over data center construction has intensified around environmental concerns and electrical grid strain, creating friction in states like Ohio despite SoftBank's parallel announcement of a 9.2 gigawatt natural gas facility there.
France's regulatory environment, energy profile, and industrial incentives proved attractive enough for a multi-decade, nine-figure commitment. This is not a favor to France; it is a business decision that reflects real constraints elsewhere.
The 2031 delivery window matters. Demand for GPU capacity in 2028-2030 is likely to exceed current supply commitments globally. SoftBank is signaling confidence in sustained high-margin demand, but also locking in geography and price before bottlenecks tighten further.
Timing also signals SoftBank's view of model scaling costs. A 3.1 GW phased build assumes either stable power economics or improving ones; if training efficiency stalls and power demand continues climbing, this bet could overshoot actual workload.
Plan infrastructure allocation early
If your training or inference workloads require 2029-2031 capacity, SoftBank's Hauts-de-France facilities will likely be oversubscribed. Document your current power and compute requirements now and contact SoftBank's infrastructure sales team (or its upstream partners) before public reservation windows close in Q4 2026 or Q1 2027.
Verify power SLA terms separately from capacity commitments. A 3.1 GW facility means little if the regional grid cannot sustain consistent draw during peak training runs. Request historical grid stability data for Hauts-de-France and model your workload against published utility constraints.
For on-premises alternatives, this announcement raises the floor for negotiating multi-year cloud capacity agreements. Cloud providers will now compete against a known, large supply addition. Use that leverage to lock discounts or priority allocation before SoftBank's facilities reach market.