The Geopolitics of Compute

While immense private investments in the United States pour into highly capitalized infrastructure and massive models, Western sanctions pressure in China intensified the incentive for maximum hardware efficiency: Alongside monolithic dense architectures, highly efficient sparse Mixture-of-Experts (MoE) systems gained critical momentum, demonstrating how active compute requirements per token can be significantly reduced.

Europe stands at a strategic juncture: Possessing strong foundational research in mathematics and computer science, limited access to localized high-performance compute historically hampered autonomous scaling. The European Tech Sovereignty Package and the proposed Cloud and AI Development Act (CADA) aim to establish a new sovereign operating environment alongside EuroHPC supercomputers.

Facing constrained access to cutting-edge accelerators, Chinese labs deployed sparse and MoE architectures with particular consistency for efficiency gains. However, the foundational techniques reflect an international research history.

Export controls intensified incentives to utilize existing hardware more efficiently, amplifying an existing research trajectory toward leaner architectures. Frontier Chinese models demonstrate that competitive performance is achievable with significantly lower active compute and parameter overhead.

Despite significantly lower measured private AI investment ($285.9B in the US versus $12.4B in China according to the Stanford AI Index), leading Chinese models closed much of the capability gap. However, comparing private venture flows alone captures state-directed investments and alternative capital vehicles in China only incompletely.

Crucially, citation metrics show a pronounced shift: According to tracking by ASPI and Stanford HAI, Chinese institutions account for approximately 36% of global AI publications and up to 48% of high-impact citations. The assertion that Chinese research is purely volume-driven without citation impact is challenged by these benchmarks.

European research institutions have made significant contributions to efficient AI, mathematical optimization, and foundational theory. However, many of today’s pivotal architectures emerged from a global research ecosystem spanning international universities and industrial laboratories.

Limited access to large-scale compute infrastructure made it challenging for European research teams and startups to scale frontier research domestically. Consequently, portions of talent and commercial value creation gravitated toward more heavily capitalized international ecosystems.

The US CLOUD Act (18 U.S.C. § 2713) clarifies that US providers must comply with lawful US disclosure orders regardless of where data is stored, provided the records remain within their possession, custody, or control.

Pure data localization can serve as a relevant baseline for compliance, but does not eliminate extraterritorial jurisdictional risks as long as corporate parent entities remain subject to foreign disclosure orders. Achieving comprehensive autonomy requires corporate governance and operational separation.

⚖️ Left Pillar · Regulatory Framework (CADA): The proposed Cloud and AI Development Act establishes tiered sovereignty criteria for public procurement and critical data to mitigate jurisdictional exposures.

🖥️ Middle Pillar · Hardware (EuroHPC): Targeted capital injection for European exascale supercomputers (such as JUPITER and Jules Verne) and their expansion into dedicated AI Factories powered by clean energy.

🌐 Right Pillar · Software (Open Source First): Strategic support for European open-source and open-weights foundation models (via consortia like ALT-EDIC and European research alliances), aligned with EU standards and transparency.

Level 1 establishes a baseline EU foundation: Providers, infrastructure, and data processing must be largely situated within the Union and fulfill fundamental security and transparency requirements.

Level 2 expands requirements with stronger operational, legal, and supply-chain independence from third countries (including localized operational and support teams and contractual safeguards).

Level 3 tightens requirements specifically around ownership and control. Providers subject to third-country control face substantial structural hurdles; definitive classification depends on the final enacted text, third-country recognition decisions, and service audits.

Level 4 introduces the most stringent requirements for personnel vetting, third-country autonomy, and software supply chains, including comprehensive transparency and verified mitigation pathways for critical dependencies.

European providers (such as Hetzner, OVHcloud, Scaleway, IONOS) possess structural alignment with Level 3 requirements due to EU headquarters and domestic governance, subject to future formal conformity audits.

International hyperscalers (AWS, Microsoft Azure, Google Cloud) face substantial structural hurdles under the proposed criteria regarding corporate separation and extraterritorial obligations under the US CLOUD Act.

Exascale systems such as JUPITER (FZ Jülich) and Alice Recoque (Jules Verne Consortium) provide European researchers and startups with access to frontier compute clusters.

Direct integration with renewable energy sources and industrial district heating sets new standards for energy-efficient large-scale inference.

1. Efficiency-Driven Architectures: Systematic adoption of sparse MoE and compressed open-weights models to circumvent hardware bottlenecks via software design.

2. Sovereign Regulatory Framework (CADA): Establishing clear assurance tiers (Level 3) that give public institutions and commercial enterprises verifiable, independent cloud options.

3. Domestic Compute Capacity (EuroHPC): Physical construction and dependable operation of modern high-performance infrastructure on European soil.

From JUPITER in Jülich to AI centers in Paris, routing hubs in Frankfurt, and green Nordic data centers in Finland, an interconnected European high-performance backbone is emerging.

The strategic initiatives and foundational investments are in motion – industry execution, research adoption, and public procurement will now determine European self-determination in the age of intelligence.