Europe's AI Strategy Ignores the Structural Lock-In

Europe is pouring billions into AI infrastructure, hoping to replicate the success cycles seen in Silicon Valley. The strategy relies on co-financing large data centers, channeling institutional savings into venture capital, and loosening labor regulations. The underlying assumption is that by removing regulatory friction and concentrating capital, European firms will naturally rise to compete with dominant American players.

However, a recent analysis published in partnership with the AI Now Institute argues this approach is flawed. It suggests that the current ecosystem is structurally captured. Even when European AI companies succeed, the economic value often flows upstream to US-based labs and hyperscalers. These firms act simultaneously as indispensable infrastructure providers and direct competitors, creating a dependency that market incentives alone cannot break.

Market Tools Fail to Shift Power

The core issue is that current policy treats the trajectory of AI as a natural phenomenon, like a tsunami to be braced for rather than a tide that can be turned. Interventions focus on boosting supply and deregulation, but they do not address how to change the incentives in a market where choices are limited. For example, when European startups like DeepL partner with AWS or when platforms like Hugging Face are acquired by Nvidia, the structural dependence deepens rather than loosens.

This creates a paradox where Europe is trying to build sovereignty using the very tools that maintain the status quo. The analysis points out that neither consumers, businesses, nor the public sector currently have meaningful choices at any layer of the technology stack. The lack of alternative options means that value accumulation remains concentrated in the hands of a few dominant providers, regardless of local investment efforts.

Uncertainty Offers a Strategic Opening

While the market is heavily concentrated, it is also in flux. The dominant approach to building AI, which relies on ever-larger models and cloud-based training, is not yet fixed. The analysis suggests that policymakers should take this uncertainty seriously rather than assuming a single future. Different scenarios, such as the commoditization of models or a shift in where value accumulates, require distinct policy responses.

One promising signal is the growing demand for open-weight models. Customers are increasingly turning to these models because they are often sufficient for their needs and significantly cheaper. This trend suggests a potential shift toward a market where several providers offer interchangeable models, making switching costs low. If this trajectory holds, the model itself stops being a scarce asset, and the cost of compute becomes the primary differentiator.

Policy Must Adapt to Shifts

The report advocates for a policy framework that encourages different ways of thinking about sovereignty. Instead of just removing friction to let the market work, Europe needs to prepare for multiple possible futures. This includes scenarios where inference costs determine value or where leading labs move up the stack into enterprise products. By acknowledging that the market has not yet answered the question of where value will accumulate, policymakers can create more resilient strategies.

There is no silver bullet that solves these structural issues without causing some pain in the tech environment. However, the current approach of deepening dependence on US infrastructure is not the only option. By focusing on the signals of commodification and preparing for a more fragmented market, Europe can potentially escape the gravitational pull of dominant players. This requires moving beyond the idea that AI is an unstoppable force and recognizing that policy can shape the structure of the ecosystem itself.