Many German AI companies use language models and advertise with digital sovereignty. However, a new study shows that a large proportion still rely on US technology. We show where the sovereignty gap is greatest.
Whether in customer service, in the legal department or in product development: artificial intelligence is an integral part of everyday work in many companies, including in Germany. A key word here is “digital sovereignty”.
What this means is that when it comes to artificial intelligence, companies and authorities remain independent of foreign providers, store their data in Europe and are not dependent on US companies.
Some companies advertise that they are independent of US providers. A new study by the European AI platform GreenPT now shows which AI models German companies really use.
German companies use these AI models
For the report, GreenPT examined 514 publicly discoverable German AI companies. Important: There is no official register of German AI companies. So GreenPT could only count which companies were publicly discoverable.
Only 221 of these companies disclose which AI provider they use, for example in their data protection declaration or on their own website. The investigation therefore relates exclusively to these companies. The most important finding: Of the German AI companies that publicly name a provider, a good half rely on US technology.
Models from US providers are clearly ahead
According to the official announcement of the research report, around 53 percent of German AI companies that publicly state which AI model they use depend on a US frontier provider. Of the US models, 41 percent are from OpenAI, 18 percent from Anthropic, ten percent from Google, five percent from Microsoft, ten percent from AWS and six percent from Meta’s Llama.
A similar proportion (around 43 percent) develop their own models. Mistral, the only frontier laboratory based in the EU, achieves around ten percent. According to the report, a frontier model hosted in Germany does not currently exist.
In a European comparison, Germany is in the middle field. What is striking is that German companies use Google and Anthropic slightly more often than the European average, which relies more heavily on OpenAI.
How heavily an industry relies on US technology depends largely on the use case. According to the study, around 85 percent of disclosing companies in health and biotech develop their own models, often because strict approval rules require this.
In industry it is around 71 percent, for example for image recognition or maintenance forecasts. In areas such as law, human resources and educational technology, however, ready-made US solutions clearly dominate.
AI models in German companies reveal a large gap in sovereignty
Of the 514 companies examined, 173 actively advertise sovereignty, EU data residency or on-prem operation. Of these 173 companies, 72 still use a US proprietary model in the background, i.e. a closed, commercial AI system from a US provider.
This gap occurs particularly often in the areas of law, human resources, administration and finance, i.e. precisely where compliance and data protection should play a particularly important role.
GreenPT’s investigation is a snapshot from mid-2026. However, provider relationships could change quickly, according to the statement. In addition, the European benchmark combines country studies with current research from 2026. The numbers should therefore be read as a trend rather than the result of a controlled comparison.
GreenPT is a European AI provider that offers chat and API access to open source models and says it works exclusively with renewable energy. The company positions itself as an alternative to US services and touts EU hosting and data protection.
Also interesting:
