SoftBank to Invest €75B in France to Build AI Compute Hub, Cementing Europe’s Role as the World’s Third Compute Power Pole

SoftBank’s €75-Billion Bet on French AI Infrastructure: Escalating the European Compute Race and Global Reallocation of AI Capital

Japanese conglomerate SoftBank Group has announced plans to invest up to €75 billion in artificial intelligence infrastructure in France—delivering a total computing capacity of 5 GW. This figure not only sets a new record for the largest single AI infrastructure investment outside the United States but also serves as a strategic fulcrum reshaping the global geography of compute power. Amid intensifying U.S.–China tech rivalry and unprecedented global awareness of energy security and digital sovereignty, this massive commitment—personally championed by Masayoshi Son and finalized within 48 hours of President Macron’s dinner with Son in Tokyo—is far more than a routine commercial move. It represents a systemic reallocation of capital across data centers, optical communications, power equipment, green energy, and semiconductor supply chains.

Geopolitical Logic: Forging a “Third-Pole” Compute Hub Amid U.S.–China Decoupling

Global AI compute supply today exhibits extreme concentration: The U.S. dominates via NVIDIA’s H100/B100 chips and hyperscale cloud clusters (AWS, Azure); China has built an autonomous closed loop centered on Huawei’s Ascend chips and its national intelligent computing centers under the “East Data, West Computing” initiative. Europe, despite deep expertise in industrial software and scientific research, has long been constrained by three interlocking bottlenecks—underdeveloped compute infrastructure, hollowed-out chip manufacturing capabilities, and unstable green electricity supply. SoftBank’s choice of France—over Germany or the Netherlands—carries clear geopolitical significance. France is the EU’s sole core nation possessing: (1) a nuclear-dominated low-carbon grid (nuclear power accounts for ~65% of generation), (2) full-spectrum defense-grade energy engineering capability, and (3) an unwavering commitment to “strategic autonomy.” Its National AI Strategy 2030 explicitly identifies compute sovereignty as a top priority, while the Macron administration is fast-tracking the Digital Sovereignty Act, mandating local storage and processing of critical AI service data. SoftBank’s capital injection effectively offers Europe a “third path”—bypassing dependence on both U.S. technology standards and supply chains (and thus evading U.S. export controls on advanced AI chips) and avoiding the political risks of deep integration into China’s compute ecosystem.

Industrial Ripple Effects: How 5 GW of Compute Load Ignites Demand Across Entire Value Chains

A 5-GW peak-compute cluster is no simple aggregation of server racks. Industry benchmarks indicate that each 1 GW of AI compute requires:

• ~1.8–2.2 GW of stable green power;
• 300,000–400,000 high-speed optical modules (including 800G/1.6T); and
• over 100,000 liquid-cooled servers plus intelligent power distribution systems.

SoftBank’s project will therefore directly catalyze demand across three major industrial sectors:

First, green energy infrastructure enters an explosive growth window. France currently operates ~61 GW of nuclear capacity—but AI clusters demand dedicated, millisecond-responsive microgrids deployed on-site. The project will accelerate France’s approval process for small modular reactors (SMRs) and drive large-scale expansion of hybrid wind/solar + storage plants. According to RTE (Réseau de Transport d’Électricité), the added load equals annual electricity consumption for ~2 million households—significantly boosting order volumes for domestic power equipment manufacturers such as Schneider Electric and Alstom’s grid business. It also opens opportunities for Chinese firms: Sungrow and CATL have already signed integrated wind–solar–storage agreements with French utility EDF; SoftBank’s investment may expedite deployment of these projects.

Second, optical communications and liquid cooling emerge as pivotal bottleneck technologies. East–west traffic within AI training clusters is surging, pushing traditional copper-based architectures to their physical limits. Penetration of 800G optical modules will jump from today’s ~35% to over 80%, while commercialization of 1.6T silicon photonics modules will accelerate by at least 18 months. Leading A-share firms—including Innolight, Accelink, and TFC—have already passed qualification audits by European cloud providers; SoftBank’s project could become their largest single overseas customer. Concurrently, immersion liquid cooling is set to boom: France’s warm climate demands higher thermal efficiency, and firms such as Sugon and Inspur Information—already with proven liquid-cooled data center deployments in Paris—will see strengthened localization capabilities in Europe.

Third, the semiconductor supply chain experiences structural loosening. Though SoftBank has not disclosed chip procurement details, insiders confirm it will adopt a heterogeneous hybrid architecture: AMD MI300X GPUs will handle general-purpose inference on select nodes, while custom ASICs (e.g., Graphcore’s IPU from the UK) will shoulder the majority of compute workloads. This opens a window for European chip design houses—and incentivizes TSMC to accelerate 2nm capacity ramp-up at its Dresden fab in Germany, scheduled for volume production in 2026—precisely overlapping with Phase II construction of SoftBank’s project.

Capital Paradigm Shift: Japan’s “Third-Pole” Strategy Triggers Chain Reactions

SoftBank’s move signals Japan’s formal abandonment of its historical “fence-sitting” posture between the U.S. and China—and its pivot toward building independent technological influence anchored in infrastructure. The underlying logic is twofold: U.S. chip restrictions against China have cost Japanese equipment makers (e.g., Tokyo Electron’s etching tools) access to the mature-node market; simultaneously, China’s explosive growth in AI applications has generated strong, non-negotiable demand for high-bandwidth, low-latency compute. The French project strikes a perfect balance—enabling GDPR-compliant data hosting services for Chinese clients while offering Japanese automakers and robotics firms a dual-continent AI training platform spanning Europe and Asia.

This paradigm is already triggering resonance across markets. SK Hynix is in talks with France’s CEA (Atomic Energy and Alternative Energies Commission) on co-developing in-memory computing chips; Abu Dhabi’s Mubadala Investment Company has dispatched a delegation to assess data center sites in Normandy; and notably, Hohhot’s newly launched “Green Compute Full-Stack AI Platform” in China emphasizes integrated wind/solar/hydrogen power + intelligent computing + “Model-as-a-Service”—a transcontinental echo of SoftBank’s blueprint. As compute becomes a new type of critical infrastructure, capital is no longer chasing algorithms or applications alone—it is racing to secure end-to-end control across energy, chips, networks, and algorithms.

Risk Disclosure: Geopolitical Volatility and Technological Generational Gaps Cannot Be Overlooked

We must remain sober about the project’s real-world constraints:

1. Geopolitical security spillover: Frequent attacks on Black Sea and Baltic Sea shipping, along with Ukrainian drone strikes on Russian southern oil depots, underscore that energy infrastructure has become a prime target in hybrid warfare. As NATO’s nuclear deterrence frontline, France’s data centers must meet STELLA-4 cybersecurity standards (equivalent to the U.S. military’s JWCC)—substantially increasing construction costs.

2. Technological generational gap: NVIDIA has already unveiled its Blackwell Ultra architecture, whereas Europe’s indigenous GPUs remain at the prototype stage. Should SoftBank rely excessively on U.S.-origin chips, its “third-pole” narrative would be significantly undermined.

3. Green-power stability challenges: France’s aging nuclear fleet faces acute corrosion issues—40% of reactors were offline in 2023 due to pipe degradation—posing severe resilience tests for the grid under a 5-GW load.

SoftBank’s €75-billion investment is not an endpoint—but rather the starting pistol for a multipolar era in global AI infrastructure. As compute power emerges as a new frontier of great-power competition, capital flows now transcend commercial rationality and ascend into a three-dimensional contest over energy sovereignty, technology sovereignty, and digital sovereignty. For China’s industrial chain, this represents both a historic opportunity—to export optical modules, liquid-cooling systems, and green-energy equipment—and an urgent call to action: to accelerate breakthroughs in domestic AI chips. True compute autonomy will always rest on the bedrock of hard-tech self-reliance.