SoftBank Bets €75 Billion on Nuclear-Powered AI Data Centers in France

The race to build AI infrastructure has entered a new phase — one defined not by which company has the best model, but by which nation has access to enough cheap, reliable, low-carbon electricity to power the data centers that run them. On May 31, 2026, SoftBank Group made the most consequential single bet of that race: a commitment of up to €75 billion ($87 billion) to build 5 gigawatts of AI data center capacity in France.

The announcement came at the annual Choose France economic summit hosted by President Emmanuel Macron, where SoftBank’s pledge was framed as Europe’s response to the AI infrastructure buildout racing ahead in the United States and Gulf states. If the full investment materializes, it will be the largest single foreign direct investment in French history and one of the largest AI infrastructure commitments made by any company anywhere in the world.

The Nuclear Thesis

The core logic behind SoftBank’s choice of France is electricity.

France generates approximately 70 percent of its electricity from nuclear power, giving it among the most carbon-intensive-free and price-stable grids in Europe. For AI data centers, this matters enormously. A large-scale AI training cluster can consume as much electricity as a small city. The economics of running that cluster depend heavily on whether power costs $0.04 per kilowatt-hour or $0.12 per kilowatt-hour — and whether the grid can reliably deliver 500 megawatts to a single campus without brownout risk.

In regions without surplus low-carbon capacity — much of the United States, the UK, and large parts of Central Europe — the proliferation of AI data centers creates a real dilemma: either strain existing residential grids and push up household bills, or constrain AI buildout by limiting new data center connections. France, with its combination of surplus nuclear capacity, existing high-voltage transmission infrastructure, and a government actively seeking major AI investment, neatly sidesteps that dilemma.

SoftBank’s partnership with state-owned utility EDF reinforces this thesis. EDF has made available a decommissioned power plant site in Bouchain, in the Hauts-de-France region near the Belgian border, that comes with existing high-voltage grid connections already in place. Repurposing a decommissioned plant site eliminates the years of permitting, environmental review, and grid connection work that typically delays new data center construction — one of the primary bottlenecks in AI infrastructure deployment globally.

Three Sites in Northern France

The first phase of the project allocates approximately €45 billion to deliver 3.1 gigawatts of data center capacity across three locations in the Hauts-de-France region by 2031. The sites are:

Dunkirk (Loon-Plage): SoftBank will collaborate with Schneider Electric to develop an industrial cluster at the Port of Dunkirk. The port location provides access to subsea cable landing infrastructure that connects Europe to North America and the Middle East — critical for the low-latency connectivity that hyperscale AI services require.

Bosquel: An inland site in the Somme department, chosen for land availability and grid access.

Bouchain: The EDF partnership site, offering a decommissioned power plant campus with existing 400kV grid connection infrastructure that SoftBank will repurpose for high-density AI compute.

A second phase, bringing total investment to €75 billion and total capacity to 5 gigawatts, extends beyond the initial three sites across additional French regions. The exact locations and timeline for Phase 2 have not been publicly specified.

Geopolitical Significance: Europe Enters the Race

The scale of the SoftBank commitment — €75 billion over what will be the better part of a decade — transforms France from an ambitious aspirant in the AI infrastructure race to a genuine contender.

To put the number in context: Microsoft’s entire 2026 global capital expenditure budget for AI infrastructure is approximately $80 billion. SoftBank is committing essentially the same amount to a single country. The investment is larger than France’s annual defense budget and represents a significant fraction of the country’s GDP growth in any single year.

French economic minister Roland Lescure described the investment as supporting “President Macron’s ambition to position France as a leading destination all along the AI value chain.” That ambition has been explicit since France hosted the AI Action Summit in Paris in early 2025, where major nations signed commitments to AI governance and investment. The SoftBank deal represents the most concrete and capital-intensive manifestation of those ambitions yet.

The geopolitical subtext is equally significant. As the United States tightens export controls on advanced AI chips, and as the Gulf states deploy sovereign wealth capital to build competing AI infrastructure, Europe has faced a real risk of being structurally dependent on AI compute built elsewhere. A 5 GW data center campus in France — connected to Europe’s grid, operating under European data governance, and serving European hyperscalers and enterprise customers — would give the continent a degree of AI infrastructure sovereignty that presently does not exist.

SoftBank’s Strategic Calculus

For SoftBank, the investment reflects a strategy that CEO Masayoshi Son has articulated consistently over the past two years: that AI is the most important investment opportunity in human history, and that the bottleneck determining who captures that opportunity is not talent, capital, or model quality — it is physical infrastructure.

SoftBank is simultaneously an investor in and a customer of OpenAI, having deployed billions through its Vision Fund and committed to purchasing AI services at scale. The France data center investment fits within a broader bet that demand for AI compute will grow faster than supply for the next decade, making owned infrastructure more valuable than contracted cloud capacity.

The company has also signaled that the France project is not a one-off. SoftBank has indicated plans for similar-scale infrastructure investments in other geographies with favorable power and permitting environments — the same thesis applied to different electricity grids.

Industrial Partners and Employment

Schneider Electric’s involvement in the Dunkirk hub is more than a construction contract. Schneider, which is one of France’s flagship industrial companies and a global leader in electrical infrastructure for data centers, brings both technical expertise and political credibility to the project. The collaboration also creates a supply chain anchor in the region, as Schneider’s involvement tends to cluster adjacent suppliers and service providers.

SoftBank has committed to creating thousands of high-skilled jobs across data center development, engineering, energy systems, robotics, operations, maintenance, and advanced manufacturing. The precise job creation figures have not been publicly quantified, but at 5 GW of capacity, the operational workforce requirements alone would represent one of the larger technology employment investments in France in recent history.

What 5 GW Actually Means

To understand why 5 gigawatts is a transformative number, consider that the entire global stock of installed AI-optimized data center capacity as of 2024 was estimated at roughly 20 to 30 GW. SoftBank’s France commitment represents 15 to 25 percent of that total, concentrated in a single country.

A 5 GW campus, once operational, would be capable of running tens of thousands of Nvidia H100-class GPUs simultaneously — enough compute to train the next several generations of frontier AI models and run inference at a scale that serves tens of millions of users. The Dunkirk-Bosquel-Bouchain triangle, if built to plan, would be one of the largest concentrations of AI computing power in the world.

Whether the project proceeds on time and at stated cost is a separate question from its strategic ambition. Large infrastructure commitments of this scale regularly encounter delays, permitting complications, supply chain constraints, and market changes that stretch timelines. But the signal it sends — that the global AI infrastructure race is now being fought on European soil, with European nuclear power as a competitive weapon — is likely to prove accurate regardless of the specific delivery schedule.