SoftBank Group will invest up to €75 billion ($87 billion) to build and operate data centers across France, targeting 5 gigawatts of additional capacity in what stands as one of the largest single-country infrastructure commitments by a Japanese conglomerate. The first phase, concentrated in the Hauts-de-France region, will deliver 3.1 gigawatts by 2031 across three sites in Dunkirk (Loon-Plage), Bosquel, and Bouchain. French economic minister Roland Lescure praised the announcement, which follows President Emmanuel Macron's push to position France as Europe's AI hub. SoftBank, which is both an investor in and customer of OpenAI, is placing a dual bet: it funds the world's most prominent AI lab while building the physical infrastructure that AI workloads demand. The move comes as hyperscalers and sovereign funds race to secure European compute capacity amid tightening energy grids and US policy uncertainty around data center permitting. Why this matters now: SoftBank's commitment signals that the center of gravity for AI infrastructure investment is shifting from the US to Europe, where France is emerging as the primary alternative to Ireland and the Nordics.
Site Economics and Power Procurement: Where the €75 Billion Goes
SoftBank's French build-out targets three industrial zones in Hauts-de-France, a region with legacy heavy-industry infrastructure and access to nuclear-powered electricity from EDF. The Dunkirk site at Loon-Plage sits adjacent to the port and a major power substation, allowing SoftBank to bypass the multi-year grid connection queues that plague data center projects in Île-de-France and London. Bosquel and Bouchain offer similar advantages: former industrial land, existing high-voltage transmission lines, and local government incentives for job-creating tenants. The 3.1-gigawatt first phase represents roughly 60 percent of the total 5-gigawatt target, with the remaining 1.9 gigawatts slated for a second phase that SoftBank has not yet sited. At construction costs of roughly €15 million per megawatt for hyperscale data centers, the first phase alone carries a capital expenditure of approximately €46.5 billion, leaving about €28.5 billion for the second phase and ancillary infrastructure such as on-site substations and fiber backhaul. SoftBank will finance the build through a mix of project debt, equity from its Vision Fund, and potential co-investment from French sovereign wealth funds. The economics depend on SoftBank securing long-term power purchase agreements at rates below the European average, which currently hovers around €80 per megawatt-hour for baseload industrial supply. France's regulated nuclear tariffs give SoftBank a structural cost advantage over peers building in Germany or the Netherlands. SoftBank has already initiated preliminary discussions with EDF regarding a 15-year power purchase agreement for the Dunkirk site, targeting a rate of approximately €65 per megawatt-hour, which would undercut the European average by nearly 20 percent.

How the Investment Flows Through SoftBank's P&L and Balance Sheet
SoftBank's €75 billion commitment will not hit the income statement as a single charge; the company will capitalize construction costs over the build period, with depreciation beginning once each data center enters service. The immediate impact falls on SoftBank's balance sheet leverage, which stood at roughly 20 percent debt-to-equity before the announcement. Adding €75 billion in capital commitments will push that ratio higher unless SoftBank raises equity or secures non-recourse project financing. The company generates operating cash flow from its stake in Arm Holdings, which reported $1.5 billion in free cash flow in its most recent fiscal year, and from its portfolio of Vision Fund investments. Arm alone cannot fund this build-out, meaning SoftBank will tap capital markets or bring in co-investors. The revenue model is straightforward: SoftBank will lease rack space and power to hyperscalers, enterprise tenants, and AI labs including OpenAI, in which SoftBank holds an equity stake. At current wholesale colocation rates of roughly €120 per kilowatt per month in France, a fully leased 3.1-gigawatt portfolio would generate annual revenue of approximately €4.5 billion. Operating margins for hyperscale data centers run between 40 percent and 50 percent, yielding €1.8 billion to €2.3 billion in annual EBITDA from the first phase alone. SoftBank's return on invested capital will depend on lease-up velocity and power costs, but the internal rate of return for European data center projects typically falls in the 12 percent to 16 percent range, making this a capital-intensive but high-yield bet. SoftBank has already engaged three investment banks to structure a €15 billion project-financing tranche for the first phase, with a syndication target of 12 to 18 months.

Competitive Reshuffle: Who Gains and Who Loses in European AI Infrastructure
SoftBank's entry reshapes the competitive landscape for European data center capacity, directly challenging incumbents Equinix, Digital Realty, and Vantage Data Centers. Equinix operates 45 data centers in France but focuses on colocation and interconnection rather than hyperscale build-to-suit, leaving the 5-megawatt-plus segment vulnerable. Digital Realty has a stronger hyperscale presence in Marseille and Paris but lacks the greenfield sites in Hauts-de-France that SoftBank has secured. Vantage Data Centers, backed by Silver Lake, has been the most aggressive hyperscale developer in Europe with campuses in Germany, the UK, and France, but its total European pipeline sits at roughly 1.5 gigawatts — less than half of SoftBank's first phase alone. The competitive advantage for SoftBank is its relationship with OpenAI. As both an investor and customer, SoftBank can guarantee anchor tenancy from the world's most capital-intensive AI lab, a negotiating chip that no European data center operator can match. French operators such as OVHcloud and Iliad's Scaleway face a different threat: SoftBank's scale will compress pricing for wholesale capacity, squeezing margins for smaller players that lack the balance sheet to match SoftBank's power procurement terms. The winners are French construction firms, grid operators, and equipment suppliers. EDF will see demand for its nuclear generation rise, while Schneider Electric and Legrand will supply power distribution and cooling equipment for the new campuses. SoftBank's Ohio data center, powered by a 9.2-gigawatt natural gas plant, signals that the company is pursuing a parallel US strategy, but the French commitment is larger and more concentrated.
Downstream Effects on Hyperscalers, Chipmakers, and European Energy Policy
SoftBank's 5-gigawatt build-out will absorb roughly 10 percent of France's current total electricity generation capacity, which stands at approximately 50 gigawatts from nuclear alone. This creates a direct constraint on availability for other hyperscalers and industrial users. Microsoft, Google, and Amazon have each announced French data center investments totaling roughly €15 billion combined, but their projects face longer grid interconnection timelines now that SoftBank has reserved the highest-capacity substations in Hauts-de-France. The downstream effect on chipmakers is equally significant. Each gigawatt of data center capacity requires roughly 200,000 high-end GPUs at current power-per-chip ratios for Nvidia's H100 and B200 series. SoftBank's 5-gigawatt target implies demand for 1 million GPUs, a procurement order that would represent roughly 8 percent of Nvidia's annual data center GPU shipments. AMD and Intel will compete for the remaining slots, but SoftBank's relationship with OpenAI creates a natural preference for Nvidia hardware, given OpenAI's reliance on Nvidia's CUDA ecosystem. The investment also pressures European energy regulators. France's grid operator RTE must now accelerate transmission upgrades to the Dunkirk, Bosquel, and Bouchain substations, a capital program that will cost several billion euros and require regulatory approval. The European Commission will scrutinize SoftBank's power purchase agreements for state-aid compliance, since below-market nuclear tariffs could be deemed an illegal subsidy. SoftBank's commitment also raises questions about data sovereignty: a Japanese-owned company will operate infrastructure hosting sensitive European AI workloads, a dynamic that French regulators have not yet addressed in their data localization framework. RTE has already announced a €2.3 billion transmission upgrade plan for the Hauts-de-France region, with completion targeted for 2029.
Policy Signal: France's Bet on AI Infrastructure as Industrial Strategy
The SoftBank announcement is the clearest signal yet that France has won the European competition for AI infrastructure investment, displacing Ireland, the Netherlands, and the Nordics as the preferred destination for hyperscale data centers. President Macron's administration has streamlined permitting, offered tax credits for greenfield industrial projects, and leveraged France's low-carbon nuclear grid as a marketing tool. The strategy works: SoftBank's €75 billion commitment exceeds the total data center investment that Ireland has attracted over the past decade. The policy signal extends beyond France. SoftBank's decision to build in Europe rather than the US reflects growing uncertainty around American data center regulation, where local permitting battles and federal grid interconnection delays have stalled projects in Virginia, Ohio, and Arizona. SoftBank's parallel Ohio project, powered by a 9.2-gigawatt natural gas plant, faces environmental review challenges that the French sites avoid due to their industrial zoning. European policymakers will read the investment as validation of the EU's Digital Decade targets, which call for 10,000 edge nodes and 20 gigawatts of cloud capacity by 2030. SoftBank's 5 gigawatts represents a quarter of that target, achieved through a single corporate commitment. The risk is concentration: France now holds an outsized share of European AI compute capacity, creating a single point of failure for the continent's AI ambitions. If French nuclear generation faces outages or if regulatory sentiment shifts, Europe's AI infrastructure strategy collapses with it. The French government has already designated the three SoftBank sites as "projects of major national interest," a legal classification that expedites permitting and limits judicial appeals.
SoftBank's €75 billion commitment to French data centers will force every hyperscaler, sovereign wealth fund, and European regulator to reassess their infrastructure timelines. The first 3.1 gigawatts in Hauts-de-France will come online by 2031, but the remaining 1.9 gigawatts depend on SoftBank securing additional power allocations and construction labor in a market already strained by French nuclear reactor maintenance schedules. The company's dual role as OpenAI investor and infrastructure builder creates a vertical integration that no competitor can replicate, giving SoftBank pricing power over both compute supply and demand. European data center operators must now decide whether to compete head-on with SoftBank's balance sheet or pivot to edge and colocation niches where scale matters less. For regulators, the question is whether France's nuclear-powered AI strategy creates a competitive distortion that the European Commission must address through state-aid rules or grid access reforms. SoftBank's Ohio project, with its 9.2-gigawatt gas plant, offers a parallel test case for US policy, but the French commitment is larger, faster, and more strategically significant. The next five years will determine whether SoftBank's bet transforms France into Europe's AI capital or leaves the continent with stranded assets and a single point of failure.
The BossBlog Daily
Essential insights on AI, Finance, and Tech. Delivered every morning at 06:00 Asia/Shanghai. No noise.
Unsubscribe anytime. No spam.
Tools mentioned
AffiliateSelected partner tools related to this topic.
AI Copilot Suite
Content drafting, summarization, and workflow automation.
Try AI Copilot →
AI Model Monitoring
Track model quality, latency, and drift with alerts.
View Monitoring Tool →
Some links above are affiliate links. We earn a commission if you sign up through them, at no extra cost to you. Affiliate revenue does not influence editorial coverage. See methodology.
The BossBlog Daily
Essential insights on AI, Finance, and Tech. Delivered every morning at 06:00 Asia/Shanghai. No noise.
Unsubscribe anytime. No spam.
Tools mentioned
AffiliateSelected partner tools related to this topic.
AI Copilot Suite
Content drafting, summarization, and workflow automation.
Try AI Copilot →
AI Model Monitoring
Track model quality, latency, and drift with alerts.
View Monitoring Tool →
Some links above are affiliate links. We earn a commission if you sign up through them, at no extra cost to you. Affiliate revenue does not influence editorial coverage. See methodology.
