Business & policy

SoftBank pledges €75 billion for 5 GW of AI data centres in France

At a glance:

  • SoftBank will invest up to €75 billion ($87 billion) to build 5 GW of AI data centre capacity in France, with CEO Masayoshi Son and President Emmanuel Macron expected to formally announce the deal at the Choose France Summit this weekend.
  • The first phase commits €45 billion to deliver 3.1 GW by 2031 across three sites in Hauts-de-France: Dunkirk, Bosquel, and Bouchain, with Schneider Electric partnering at the Dunkirk hub.
  • The pledge marks SoftBank’s largest European AI infrastructure bet and relies on France’s nuclear-powered grid and political stability amid rising energy costs and Gulf turmoil that are stalling rival projects.

SoftBank commits €75 billion to French AI infrastructure

SoftBank is preparing to invest as much as €75 billion ($87 billion) to construct 5 gigawatts of AI data centre capacity in France, making it the Japanese conglomerate’s largest AI infrastructure commitment in Europe. Chief Executive Masayoshi Son and French President Emmanuel Macron are expected to unveil the agreement during the Choose France Summit this weekend. The summit serves as Macron’s annual platform for pitching France to global investors, and this year’s headline act is an order of magnitude larger than the digital commitments typically on display.

The first phase alone will channel €45 billion into 3.1 gigawatts across three sites in the Hauts-de-France region by 2031. The identified locations are:

  • Dunkirk
  • Bosquel
  • Bouchain At the Dunkirk site, Schneider Electric will serve as a strategic partner, anchoring a hub for AI infrastructure and robotics manufacturing that is geographically positioned to serve London, Brussels, and Amsterdam. That northern France location gives the campus strategic value beyond raw compute capacity, tying it directly into some of Europe’s densest digital corridors.

The investment is as much a product of personal diplomacy as economic strategy. Son told La Tribune he was "very impressed by the fact that Emmanuel Macron is so personally committed to ensuring France's economic success," noting that the French president approached him directly during a visit to Japan earlier this year. Son, who is accustomed to fielding inquiries from corporate leaders, was struck by an outreach from a head of state, signaling the high-level political capital both sides are willing to spend on the project.

Bloomberg had previously reported that Son floated the idea of investing as much as $100 billion in France. The confirmed figure of €75 billion is smaller but still represents one of the largest single infrastructure commitments any country has received for AI. That gap between the indicative and the confirmed number will invite scrutiny, yet even at the lower bound the deal rewrites expectations for European digital infrastructure.

France’s nuclear edge attracts AI infrastructure

The energy economics of European AI data centres have emerged as a major hurdle for hyperscalers. OpenAI recently paused its Stargate UK project after discovering that industrial electricity costs on the island run at more than four times United States rates, a differential that can erase the margin on power-hungry AI clusters. France offers a structural counterweight: its electricity grid is approximately 70% nuclear-powered, providing low-carbon baseload capacity that most European countries simply cannot match.

Geopolitical risk is also redirecting investment flows. With oil at roughly $100 a barrel, the Strait of Hormuz closed, and Gulf AI data centre ambitions disrupted by regional conflict, boardrooms are actively looking for lower-risk alternatives. France’s political stability and its nuclear energy base position it as a credible refuge for capital that might otherwise have flowed to the Middle East. For Macron, the pitch is twofold: nuclear power supplies both a climate argument and a cost-competitive advantage over markets forced to rely on imported natural gas or volatile renewable grids.

SoftBank’s mounting AI commitments test its balance sheet

SoftBank’s French pledge is only the latest in an extraordinary run of AI infrastructure commitments that now span multiple continents. In March, the company unveiled a large-scale data centre project in Ohio that could ultimately direct $500 billion toward installing 10 gigawatts of capacity powered by roughly $33 billion worth of natural gas-fired electricity. SoftBank is also a partner in the $500 billion Stargate initiative alongside OpenAI, Oracle, and Abu Dhabi’s MGX to build data centres across the United States. Separately, it has committed more than $60 billion to OpenAI in exchange for a roughly 13% stake.

The cumulative scale of these outlays is prompting intense scrutiny of SoftBank’s financing capacity. The company recently scaled back plans for a $10 billion margin loan backed by its OpenAI stake after facing hesitation from some creditors, with discussions now targeting as little as $6 billion, according to Bloomberg. Whether Son can summon enough capital to deliver on all of his AI ambitions simultaneously is the question investors are watching most closely. The divergence between SoftBank’s visionary announcements and its day-to-day funding reality is becoming a recurring tension on earnings calls and in credit markets.

Macron’s sovereign AI vision gains a heavyweight backer

Macron has championed sovereign AI as a policy priority, arguing that nations outside the United States and China must build their own infrastructure to retain control over data and frontier technology. France has already backed Mistral AI, the Paris-based foundation model company that reached $300 million in annual recurring revenue earlier this year, as a domestic counterweight to American labs. SoftBank’s commitment validates the sovereign AI thesis at a scale that no European government initiative has yet matched, effectively anointing France as the continent’s leading destination for private AI compute investment.

The Choose France Summit is Macron’s annual showcase for attracting foreign investment, and previous editions have drawn headline pledges from Microsoft, Google, and Amazon. SoftBank’s €75 billion promise dwarfs them all. The energy source also gives Macron a diplomatic talking point: while SoftBank’s Ohio project runs on natural gas and xAI is spending $2.8 billion on gas turbines to power its own clusters, the French grid means SoftBank’s northern campuses could operate on substantially lower-carbon power than their American equivalents. For Macron, that is both a climate narrative and a competitive sales pitch to regulators and investors who are increasingly weighing carbon intensity in site-selection decisions.

Execution risks shadow Europe’s largest AI infrastructure pledge

Whether the promised €75 billion actually arrives at the scale and pace announced is the test that every infrastructure megaproject eventually faces, and SoftBank’s record offers plenty of reasons for caution. Son’s track record on vision is extraordinary, spanning early bets on Alibaba and the creation of the Vision Fund, yet his track record on execution is more complicated, with several high-profile portfolio companies requiring dramatic rescues or write-downs. France is now betting that this time the cheque clears, but the timeline runs through 2031, leaving ample room for shifting interest rates, power-grid bottlenecks, or changes in SoftBank’s own liquidity to alter the roadmap.

The announced 3.1 gigawatt first phase is already a massive undertaking by European standards, and industry observers will be watching for concrete milestones such as land acquisition, grid-interconnection agreements, and the breaking of ground at Dunkirk, Bosquel, and Bouchain. If SoftBank can move from letter of intent to steel in the ground faster than it has with some past projects, the French sites could reshape the continent’s AI supply chain. Until then, the pledge remains exactly that—a pledge—and one whose full €75 billion price tag will require a balance sheet still healing from earlier tech sector bets.

Editorial SiliconFeed is an automated feed: facts are checked against sources; copy is normalized and lightly edited for readers.

FAQ

How much is SoftBank investing and where will the data centres be located?
SoftBank has pledged up to €75 billion ($87 billion) to build 5 GW of AI data centre capacity in France. The first phase will spend €45 billion to deliver 3.1 GW by 2031 across three sites in the Hauts-de-France region: Dunkirk, Bosquel, and Bouchain. Schneider Electric will be a strategic partner at the Dunkirk location, which will serve as a hub for AI infrastructure and robotics manufacturing positioned to supply London, Brussels, and Amsterdam.
Why did SoftBank choose France over other European or Middle Eastern locations?
France benefits from a structural energy advantage because its electricity grid is roughly 70% nuclear-powered, offering low-carbon baseload capacity that few neighbours can match. By contrast, OpenAI paused its Stargate UK project after finding industrial electricity rates there at more than four times US levels. Geopolitical disruptions in the Gulf, where oil is at $100 and the Strait of Hormuz is closed, have also paused rival investment flows, making France's political stability comparatively attractive. Macron's personal outreach to Son during a Japan trip this year further sealed the preference.
How does this fit with SoftBank's other AI investments and what are the financing risks?
The French commitment sits alongside SoftBank's Ohio data centre plan, which could reach $500 billion for 10 GW, and its role in the $500 billion Stargate US initiative with OpenAI, Oracle, and Abu Dhabi's MGX. SoftBank has also committed more than $60 billion to OpenAI for roughly a 13% stake. Financing concerns are acute: SoftBank scaled back a planned $10 billion margin loan backed by its OpenAI stake to as little as $6 billion after creditors hesitated. Investors are now questioning whether Son can fund all of these simultaneous AI megaprojects without stretching the balance sheet beyond comfort.

More in the feed

Prepared by the editorial stack from public data and external sources.

Original article