Germany's AI Dependency: A Closer Look at European Models

In an era where Europe aims to gain more technological independence from U.S. solutions, a recent analysis reveals that many German companies still heavily rely on American AI models. Despite the push for sovereignty, the data indicates that only a small fraction of businesses are utilizing European alternatives.

According to the European AI platform GreenPT, approximately 53% of German AI companies that disclose their providers are dependent on U.S. firms like OpenAI, Anthropic, Google, Microsoft, or AWS. The analysis, which examined 514 identifiable German AI companies, found that 221 of these companies openly stated which AI models they were using. Furthermore, about 43% of these companies are reportedly developing their own models.

The reliance on U.S. models is particularly pronounced among the 185 companies that specify their providers on their websites, where the dependency rises to around 60%. Mistral, the only frontier lab based in the EU, is mentioned by only about 10% of companies that disclose their AI models. When compared to other European nations, Germany's level of reliance on U.S. models and the development of proprietary solutions is average.

The AI models from OpenAI are the most frequently utilized among German firms, either directly or via Azure, with a usage rate of about 41% among those that disclose their usage. This figure is slightly below the European average. Following OpenAI, Google (Vertex) is used by approximately 20% of companies, while Anthropic accounts for 18%. Mistral's usage stands at around 10%, which is slightly above the European average of about 8%. Other models like Microsoft Copilot and Meta Llama are even less common.

The analysis also uncovers significant variations across different sectors. Sensitive industries tend to favor their own models, with approximately 85% of companies in healthcare and biotech developing proprietary solutions. In manufacturing, this figure is around 71%. Conversely, sectors such as cybersecurity, data, and infrastructure prefer either their own or open-source solutions, resulting in a lower U.S. dependency compared to the overall average. However, software-intensive areas like law, HR, finance, and EdTech show a stronger inclination towards U.S. models, with retail, media, and marketing exhibiting high direct usage of OpenAI.

A critical takeaway from the analysis is the notable sovereignty gap. Among the 514 companies studied, 173 actively market an EU or Germany-based data residency or on-premise deployment. Yet, 72 of these companies still utilize proprietary models from the U.S., signaling a disconnect between their claims of sovereignty and their actual technical dependencies. This trend is particularly pronounced in sectors such as law, HR, administration, customer service, and finance, where compliance pressures are highest. The findings reiterate the extent to which German companies remain reliant on American tech giants in the AI landscape.