Global AI Infrastructure Upgrade: SoftBank's French Data Hub and Hohhot's Green Computing Platform Go Live

Global AI Infrastructure Race Intensifies: SoftBank’s €75-Billion French AI Data Center and Hohhot’s Green, Full-Stack Computing Platform Launch Simultaneously—Geopolitically Anchored Compute Deployment Enters a New “Energy–Network–Hardware” Three-Dimensional Sprint

In June 2024, two major infrastructure developments—apparently parallel yet deeply resonant—drew intense attention from global technology strategy circles. First, Japan’s SoftBank Group announced plans to invest €75 billion in constructing a hyperscale AI data center cluster in France’s Hauts-de-France region (formerly Picardy), with a planned total power capacity of 5 GW, scheduled to enter phased operation starting in 2027. Almost simultaneously, Hohhot City in China’s Inner Mongolia Autonomous Region launched the nation’s first “green, full-stack AI computing platform.” Supported by a national computing hub node, this platform integrates domestically developed AI chips (Cambricon MLU370 and Huawei Ascend 910B), direct wind-and-solar-to-storage power supply systems, immersion liquid-cooling infrastructure, and an independently controllable AI training orchestration platform—achieving a Power Usage Effectiveness (PUE) of ≤1.12 and a renewable energy consumption rate of ≥93%. One project spans East and West; one is overseas, the other domestic. The SoftBank investment sets a new record for its largest overseas AI infrastructure commitment, while the Hohhot deployment pioneers the commercialization of “full-stack green AI” in China. These are not isolated events—but rather a pivotal signal that strategic competition among China, the U.S., and the EU over foundational AI infrastructure has evolved structurally: from contesting the “chip high ground” toward a deep, end-to-end race across the entire “energy–compute–network” value chain.

Geopolitical Drivers Reshape Compute Infrastructure Paradigms: From “Chip Dependence” to “Resource Sovereignty”

Historically, AI competition centered on acquiring or substituting high-end GPUs (e.g., NVIDIA H100/B100). Yet the synchronized rollout of SoftBank’s French project and the Hohhot platform signals a structural shift in competitive dimensions. Of SoftBank’s €75-billion investment, approximately 42% (€31.5 billion) is explicitly allocated to renewable energy infrastructure, including a newly built 1.8-GW offshore wind farm, co-located energy storage facilities, and a dedicated high-voltage direct-current (HVDC) transmission line. Meanwhile, the Hohhot platform leverages Inner Mongolia’s abundant renewable resources—averaging over 3,000 annual hours of wind and solar generation—and employs an integrated “source-grid-load-storage” design to suppress curtailment rates for wind and solar power to under 3%. This reveals a deeper logic: Compute is sovereignty—and sovereignty rests fundamentally on energy autonomy and territorial control. The EU’s Artificial Intelligence Act mandates that high-risk AI systems undergo training and validation within EU territory; the U.S. CHIPS and Science Act’s subsidy provisions implicitly require localized data center operations; and China’s “East Data, West Computing” initiative strategically reconfigures the national spatial layout of energy and computing resources around designated hub nodes. As AI model parameters surge past the trillion-parameter threshold—and single training runs consume electricity equivalent to tens of thousands of households’ annual usage—the question of where computation occurs has transcended technical choice to become a core pillar of geopolitical security.

End-to-End Value Reassessment: Structural Opportunities Emerge for Green Power, IDCs, Optical Modules, and Domestic AI Servers

This paradigm shift is rapidly reshaping industry valuation frameworks:

• Green power generators and ultra-high-voltage (UHV) grid networks have become the “invisible bedrock” of compute infrastructure. The French project necessitates new cross-regional transmission corridors—directly benefiting grid equipment suppliers such as Alstom and Siemens Energy. In China, UHV transmission corridors supporting Inner Mongolia’s large-scale wind-solar bases (e.g., the Mengxi–Tianjin South line) are seeing significantly accelerated construction timelines, enhancing order visibility for firms like NARI Electric and XJ Electric.

• IDC infrastructure technology requirements have leapt upward. Traditional air-cooled data centers can no longer meet thermal management demands for 5-GW-scale clusters. Liquid-cooling penetration is projected to rise from under 15% today to over 60% by 2027. Sugon’s liquid-cooled servers already cover 80% of nodes in the Hohhot platform, while Inspur’s immersion liquid-cooling solutions have received preliminary certification for SoftBank’s French project.

• Surging market attention on optical modules carries strategic warning implications. Although recent rumors about Innolight being added to the U.S. Entity List were later debunked, the market’s sharp reaction underscores a critical reality: Against the backdrop of intercontinental AI data center interconnectivity, 800G/1.6T optical modules—particularly those based on Co-Packaged Optics (CPO) technology—have become the indispensable “choke point” for data transmission. Technical iteration capabilities at firms such as Eoptolink and TFC Communications are now being assessed within national security evaluation frameworks.

• Domestic AI server vendors are entering a “full-stack adaptation” window. The Hohhot platform adopts a vertically integrated architecture—combining Huawei’s Ascend AI chips, EulerOS, the MindSpore framework, and proprietary storage systems—driving adoption of Zhongke Controllable and Huawei Atlas-series servers in key sectors including government services and finance, where their market penetration has exceeded 35%. This, in turn, is accelerating full-stack validation cycles—from chip to system—for Cambricon and Biren Technology.

Policy Coordination Strengthens: Dual-Driven Momentum from SOE Reform and Future Industry Strategy

This infrastructure upgrade is far more than a set of discrete corporate initiatives—it is the concrete embodiment of national strategic intent. A seminal article titled “Forward-Looking Planning and Development of Future Industries,” published in Qiushi, the flagship journal of the Central Committee of the Communist Party of China, states clearly: “Future industries possess forward-looking, strategic, and disruptive characteristics, requiring scientific planning and holistic coordination,” and emphasizes the principle of “industries posing questions, science and technology providing answers.” The Hohhot green computing platform precisely embodies this philosophy: Its technical roadmap was jointly formulated by the Inner Mongolia Department of Industry and Information Technology and the Institute of Computing Technology, Chinese Academy of Sciences; its energy solution was designed under the leadership of State Grid Inner Mongolia Branch. Concurrently, the intensive implementation of the Implementation Plan for Further Deepening State-Owned Enterprise (SOE) Reform (2026–2029) provides institutional scaffolding for infrastructure development: China Electric Equipment Group has identified “AI-powered dispatch terminals for next-generation power systems” as a top-priority R&D task; state-owned IDC operators—including China Telecom Cloud Computing Company—are accelerating deployment of domestically sourced intelligent computing centers at western hub nodes, per the reform plan’s directives. This dual resonance of policy and market is elevating compute infrastructure from a commercial undertaking to a national security imperative.

Conclusion: The Remapping of “Compute Geography” and the Long-Term Premium on Global Resources

SoftBank’s French project and the Hohhot platform are two sides of the same coin—jointly sketching a new geographical map for the AI era. Compute is no longer evenly distributed across the globe; instead, it is rapidly concentrating along three axes: zones rich in clean energy (Northern Europe, Inner Mongolia, Chile’s Atacama Desert), politically stable regions (core EU countries, Western China), and digital sovereignty strongholds (Singapore, Dubai). This “geography of compute” is fundamentally restructuring global resource valuations—permanently lifting long-term premiums on green electricity, critical minerals (e.g., for liquid-cooling fluids and magnetic materials), and high-quality land (low seismic risk, low water consumption). For investors, this demands moving beyond hardware-centric analysis to adopt a four-dimensional assessment model: energy cost–land constraints–network latency–policy certainty. For industry players, sustainable competitive advantage has shifted decisively—from raw chip performance metrics to system-level integration capability spanning energy, compute, and networking. As the AI race enters the deep waters of infrastructure, the decisive factors may no longer reside inside server racks—but in the rotational rhythm of wind turbine blades, in the surging current of UHV transmission lines, and in the millisecond-precision scheduling of every watt of green electricity.