The Fallacy of Sovereign Choice

European AI policy often frames the adoption of US cloud infrastructure as a moral or strategic failure by individual companies. However, the authors argue this is a structural necessity. Companies like DeepL, which built its reputation on privacy and sovereignty, pivoted to AWS because European providers lack the global distribution channels, low-latency performance, and integrated procurement marketplaces required to compete with Google Translate. For European firms, the choice is not between 'sovereign' and 'US' cloud, but between global reach and market irrelevance.

The 'Handover Blindness' in Industrial Strategy

Europe’s AI strategy is currently fixated on building 'AI Factories' and gigafactories for frontier-model training. The authors identify this as 'handover blindness': a failure to distinguish between the one-off, centralized nature of training and the distributed, elastic, and high-availability requirements of inference.

While policymakers subsidize training capacity, the real value—and the real lock-in—occurs at the inference layer. Because the most capable frontier models are proprietary and bundled with US hyperscalers (AWS, Azure, Google), any increase in AI adoption in Europe effectively acts as a subsidy for US infrastructure. European cloud providers are trapped in a 'chicken-and-egg' cycle: they cannot offer the latest proprietary models, and they lack the massive, captured customer base required to justify the financial risk of building out specialized GPU fleets for inference.

The Limits of Current Policy

Legislative efforts like the proposed Cloud and AI Development Act (CADA) fail because they treat cloud capacity as a fungible commodity. In reality, demand is driven by managed services, APIs, and proprietary model access—not just raw compute. Even when European providers attempt to compete, they face:

  • Contractual Lock-in: Hyperscalers use circular spending partnerships and discounted compute to ensure AI labs remain within their ecosystems.
  • Capital Constraints: European providers must build fleets from scratch, whereas hyperscalers can repurpose existing GPU capacity, lowering their risk.
  • Nvidia's Strategic Pivot: Even 'neoclouds' backed by Nvidia to diversify compute are increasingly becoming suppliers for US tech giants, as those are the only customers with sufficient demand to fill their capacity.