The single most important variable in AI infrastructure site selection is no longer chip supply or tax incentives. It’s electricity. SoftBank just proved it by committing €75 billion ($87 billion) to build 5 gigawatts of AI data center capacity in France, the largest single AI infrastructure deal in European history. The reason Masayoshi Son chose France over expanding his U.S. footprint: France draws 70% of its electricity from nuclear reactors, posts industrial power prices under half the UK’s, and can deliver baseload capacity to hyperscale facilities without requiring SoftBank to build its own power plants.
That last point matters more than the headline number. In Ohio, where SoftBank is also planning a 10 GW campus, the company must independently finance a $33 billion natural gas plant to generate 9.2 GW of power before a single GPU spins up. In France, state-owned EDF already has the grid.
The Deal: €75 Billion Across Three Sites by 2031
SoftBank’s plan breaks into two phases. The first phase, at €45 billion, will deliver 3.1 gigawatts of capacity across three locations in the Hauts-de-France region: Dunkirk (Loon-Plage), Bosquel, and Bouchain. Target completion for Phase 1 is 2031. The second phase expands total capacity to 5 gigawatts and pushes total investment to €75 billion.
The Bouchain site is particularly telling. EDF is converting a decommissioned power plant into a data center campus, handing SoftBank a facility that was already connected to France’s nuclear grid. Schneider Electric, the French power and cooling giant, will build a robotized manufacturing cluster at the Port of Dunkirk to produce data center enclosures and integrate power modules on site. The Dunkirk hub is positioned to serve London, Brussels, and Amsterdam, making it a distribution node for AI compute across northwestern Europe.
Son called the investment a reflection of where the AI industry is headed. “AI is entering a new era, and the countries that build the infrastructure for this transformation will shape the future of technology, industry, and society,” he said in SoftBank’s May 31 announcement.
France’s Nuclear Grid Changes the Math
The economics behind this deal are straightforward. France’s industrial electricity runs at roughly £69 per megawatt-hour. The UK’s equivalent rate: £168 per MWh, more than double. Gas sets the benchmark electricity price in France only 7% of the time. In the UK, that figure is 97%.
This gap has real consequences for AI infrastructure economics. A single hyperscale AI data center drawing 1 GW of power consumes electricity 24 hours a day, 365 days a year. At French industrial rates, the annual power cost for that facility is roughly 40% to 50% lower than the same facility in the UK. Across 5 GW of planned capacity, that cost differential compounds into billions of dollars over the operational lifetime of these centers.
France also carries a structural advantage that no amount of investment can replicate quickly: it is the world’s largest net electricity exporter. EDF operates 56 nuclear reactors. The grid doesn’t just have capacity; it has surplus. When OpenAI reportedly paused its Stargate expansion plans in the UK, the cited reason was electricity costs running four times higher than in the United States. France undercuts both.
The nuclear advantage extends beyond cost. AI data centers need baseload power, meaning consistent, uninterrupted supply. Solar and wind are intermittent. Natural gas requires building new generation capacity (exactly the problem SoftBank faces in Ohio). Nuclear is both low-carbon and dispatchable, which satisfies two requirements at once: reliable power for uptime and clean energy credentials that hyperscalers increasingly need for investor and regulatory compliance.
The NextEra/Dominion $67 billion merger signaled that AI is reshaping utility economics in the U.S. SoftBank’s France deal signals the same dynamic at a global scale: energy assets built decades ago are now the most valuable cards on the table.
Macron Pitched Son Directly
The deal didn’t come together through normal diplomatic channels. French President Emmanuel Macron approached Son personally during a visit to Japan earlier in 2026, pitching France’s structural advantages: abundant nuclear energy, a skilled workforce, and fast-tracked administrative approvals for AI installations.
Son, who is more accustomed to corporate recruitment pitches than head-of-state outreach, responded publicly. “I was very impressed by the fact that Emmanuel Macron is so personally committed to ensuring France’s economic success,” he told La Tribune.
The personal diplomacy marks a shift in how nations compete for AI infrastructure. A decade ago, governments courted manufacturing plants with tax breaks. Now, heads of state fly to Tokyo to pitch nuclear grid capacity. The asset that matters most isn’t land or labor; it’s the power infrastructure that took 40 years to build.
Choose France: €93 Billion in AI Commitments
SoftBank’s deal anchored the annual Choose France summit, held at the Palace of Versailles on June 1. Total foreign investment pledges at the event reached €93 billion across 71 projects, creating an estimated 15,600 jobs. For context, the 2025 summit set its own record at €20 billion. The 2026 figure is more than four times that, and AI infrastructure accounts for the majority.
Other notable commitments at the summit included Brookfield’s $10 billion (€8.5 billion) data center in Escaudain, a $5 billion Ardian and Verne data center in the Paris region, and Foxconn’s €120 million AI motherboard production line in Angers (partnering with French supercomputer specialist Bull).
The concentration of AI infrastructure investment tells its own story. France isn’t attracting this capital because of AI talent or research output (the U.S. still leads both categories by wide margins). It’s attracting it because of a physical asset that cannot be replicated in less than a decade: a fully built, nuclear-powered electrical grid with surplus capacity.
The Power Bottleneck Is the New GPU Bottleneck
SoftBank’s France commitment sits alongside two other massive AI buildout projects: the $500 billion Stargate initiative (with OpenAI, Oracle, and Abu Dhabi’s MGX) for U.S. data centers, and a potential $500 billion campus in Ohio requiring its own natural gas power plant. This is the same SoftBank that filed for a $100 billion IPO for its Roze robotics subsidiary to build robots that construct data centers. Including France, Son has now committed to over $1 trillion in AI infrastructure across three continents.
The pattern reflects a broader shift. Global data center capital expenditures are approaching $1 trillion in 2026, according to Dell’Oro Group, with cumulative spending projected to hit $1.7 trillion by 2030. The top four U.S. hyperscalers (Amazon, Google, Meta, Microsoft) entered 2026 with combined data center capex budgets of nearly $600 billion. Google alone committed $190 billion in annual AI capex at I/O 2026.
Two years ago, GPU supply constrained AI buildouts. Nvidia’s allocation lists determined who could train frontier models. Today, Nvidia is shipping at scale; even after losing its entire China business, it still grew revenue 85% last quarter. The constraint now is power. Who can deliver gigawatts of reliable, affordable, clean electricity determines where the next generation of AI infrastructure gets built.
France bet on nuclear power in the 1970s, when the oil crisis forced the country to rethink its energy strategy. Fifty years later, that bet is paying dividends no one predicted. The 56 reactors that power 70% of French electricity were built to guarantee energy independence. They are now guaranteeing that France captures a disproportionate share of the most capital-intensive technology buildout in history.
For the rest of Europe, and for the U.S., the lesson is uncomfortable. You cannot build 56 nuclear reactors in five years. The countries that invested in baseload power decades ago now hold a structural advantage in the AI race that no amount of venture capital or tax policy can quickly offset.