Europe Can Compete in AI Without Matching US Model Scale

Europe produced just two notable AI models in 2025 compared to 59 from the United States and 35 from China, according to analysis published by Reuters Breakingviews. Yet this gap doesn't condemn the continent to an "AI dark age" or complete dependence on American hyperscalers.

The region has viable alternatives that balance capability with sovereignty concerns—particularly for enterprises wary of sending sensitive data to centralized US cloud infrastructure.

The local deployment advantage

French AI lab Mistral, valued at $24 billion with backing from Samsung and ASML, has built its business model around customer control. The company's models can run on a client's own infrastructure rather than requiring queries to travel to distant data centers. Mistral also offers open-weight models that enterprises can download and customize for specific use cases.

Canadian-German firm Cohere follows a similar approach, selling private versions of AI systems that companies can operate internally.

While these models lack the advanced reasoning capabilities of frontier systems from OpenAI or Anthropic, they prove sufficient for most business tasks. Stanford researchers found that nearly 90% of over one million chatbot queries could be handled by small models running on local infrastructure, according to a study published earlier this year.

Europe's data advantage

The continent holds valuable datasets that could power AI applications without requiring the most sophisticated models. Large public health systems represent one significant data repository, while manufacturing and defense industries contain another.

Recent controversy surrounding Palantir's contract with the UK's National Health Service signals growing public resistance to foreign companies accessing sensitive information. European AI vendors can extract value from these datasets while keeping data within regional borders.

Former ECB president Mario Draghi has proposed that Europe's largest companies pool commitments to purchase compute from local providers, potentially accelerating development of the continent's underdeveloped data center financing market. The OECD estimates that widespread AI adoption could increase annual aggregate labor productivity growth by up to 1.3 percentage points over the next decade in countries with high AI exposure.

Why it matters

European governments and companies face a strategic choice: wait for homegrown frontier models that may never materialize, or deploy available technology that meets current needs while maintaining data sovereignty. The productivity gains from AI adoption are substantial enough that delaying implementation to achieve model parity with the US could prove more costly than accepting a capability gap for non-critical applications. Certain challenges like pharmaceutical innovation will still require frontier LLMs, but most enterprise use cases don't demand cutting-edge reasoning abilities.

European AI companies including Mistral have pushed back against calls from Anthropic to slow AI development, suggesting some US incumbents are using safety concerns to consolidate market position through favorable regulation. Germany's digital affairs ministry has indicated that halting AI progress isn't viable for Europe.

These details were first reported by Reuters Breakingviews columnist Jennifer Johnson.