SoftBank's 75 Billion Euro France Pledge Is an Energy Story, Not a Compute Story

When SoftBank Group announced at the May 30 Choose France summit that it would build and operate 5 gigawatts of AI data-center capacity in France for up to 75 billion euros, the coverage fixated on the dollar figure. That is the least interesting number in the announcement. The interesting numbers are gigawatts, grid-connection years, and the price of a megawatt-hour of nuclear electricity. SoftBank did not place an 87.5-billion-dollar bet on French real estate. It placed a bet on the French grid, and on the proposition that the binding constraint in AI is no longer chips or capital but energy that can actually be delivered to a site on a usable timeline.

What was committed

The firm commitment is the first phase: 45 billion euros for 3.1 GW, delivered by 2031, across three sites in the Hauts-de-France region: Dunkirk's Loon-Plage, Bosquel, and Bouchain. The path to the full 5 GW and 75 billion euros is conditional on an extension. The partner list signals the real strategy: Schneider Electric integrating power modules and building a power-module facility at the Port of Dunkirk, EDF supplying electricity to Bouchain, SB Energy on development, and local engineering schools on R&D. SoftBank is not renting a data center. It is building a vertically integrated cluster that manufactures its own enclosures and power blocks beside a deep-water port, located where the electrons are clean.

Why France: the grid is the asset

Roughly 70 percent of French electricity comes from nuclear power, ranging between about 60 and 75 percent on a given day. That gives France three things American sites cannot match: dispatchable low-carbon baseload for 24/7 training clusters, genuine headroom (France exported a record 92.3 TWh in 2025, equivalent to just over 10 GW of sustained capacity), and price. CRE, the French regulator, anticipated wholesale prices near 58 euros per MWh for France in 2026 versus roughly 88 euros per MWh in Germany, a 30-euro gap that compounds into billions over a 3.1 GW campus's life.

Expected 2026 wholesale electricity price (EUR per MWh)

Expected 2026 wholesale electricity price (EUR per MWh)
country	price
France	58
Germany	88

The caveat the French pitch skips: European industrial electricity still runs about double US levels for energy-intensive users. France is the cheapest seat in an expensive theater. It wins Europe decisively; it does not automatically beat Texas or Louisiana, where cheap gas and behind-the-meter generation undercut even nuclear. That is why the French and EU case leans on a second argument price cannot supply: sovereignty.

The energy wall and the announced-versus-energized gap

In 2026 AI scaling is an energy problem. The constraint is megawatts deliverable to a specific location at a specific time, not GPUs or money. Capital scales on a quarterly clock; silicon on a monthly one; transmission infrastructure on a clock of seven to ten years in standard Western European interconnection queues. That mismatch produces the single most underweighted metric in the industry: the gap between gigawatts announced and gigawatts energized.

Announcing 5 GW is a press release issued in an afternoon. Energizing 5 GW is a decade-long civil-engineering and regulatory campaign gated by 400 kV connections, substations, permits, cooling-water authorizations, and the physics of commissioning power blocks one at a time. The market rewards the announcement; the grid enforces the reality. Every announced figure deserves a discount for the probability and timeline of actual energization, and that discount is large.

Illustrative: SoftBank France announced vs energized GW

Illustrative: SoftBank France announced vs energized GW
year	announced	energized
2026	5	0.2
2028	5	1.2
2029	5	2
2031	5	3.1

France's answer: coordinate, do not deregulate

France understands the wall and has responded characteristically. In May 2025, CRE introduced a fast-track procedure (deliberation n° 2025-120) for connecting very large consumers, specifically hyperscale data centers, to the 400 kV grid in three to four years instead of seven to ten. The catch: it applies only to sites the state designates as favorable, with feasibility confirmed by RTE, the transmission operator. The state picks the sites, the state operator confirms the grid, the state utility supplies the power, and a national champion supplies the gear. That centralized model is the only reason the 2031 timeline is even arguable. It is slower to start and faster to finish, which is the right trade for twenty-year infrastructure.

This is also a deliberate reallocation of a strategic resource. RTE projects data-center consumption reaching 15 to 20 TWh by 2030 (around 3 percent of French electricity) and roughly 35 TWh by 2035, up from under 1 TWh on dedicated sites in 2025. Sovereign compute, in energy terms, means converting electricity France used to export into compute France sells to the world.

Sovereignty as the second argument

The EU has made sovereign compute concrete and funded. In January 2026 the Council amended the EuroHPC regulation to add AI gigafactories, facilities of roughly 100,000 processors each, about four times the current generation. France's AI Factory France federates national compute with the forthcoming Alice Recoque exascale system and is in talks to link with Germany's JAIF toward a virtual 50,000-GPU system. Mistral AI plans 200 MW of self-built European capacity by end of 2027 after raising 830 million euros in debt. SoftBank's commitment sits atop this: foreign capital, European soil, French power, French-made power modules. Sovereign in everything but the balance sheet, the pragmatic compromise Europe has settled on.

The contrast with US hyperscalers

The American model optimizes for time-to-energize at any cost: cheap gas, deregulated siting, strained local grids, and a deferred environmental bill. US AI data centers are projected to need 20 to 30 GW by late 2027, with Meta's Hyperion at 5 GW, OpenAI's Stargate at 10 GW across sites, and xAI's Colossus 2 near 2 GW. France cannot out-build that this cycle and is not trying to. Its bet is that a cleaner, state-coordinated model that actually energizes what it announces is more durable, and sufficient to host Europe's own frontier models and regulated industries. Sovereignty does not require parity; it requires a credible floor.

What to watch

Watch grid-connection milestones, not capex announcements: RTE confirmations and substation commissioning at Loon-Plage, Bosquel, and Bouchain; the EDF supply contract for Bouchain as the first real test of converting export headroom into dedicated compute; the Schneider power-module facility ramp at Dunkirk; whether the conditional 1.9 GW extension is triggered or quietly deferred; and local consent on water permits, land conversion, and new transmission lines. If those line up, France will have proven that coordination can beat the grid clock. If they slip, SoftBank's 5 GW joins the growing pile of gigawatts announced and never energized, and the energy wall claims another headline.