European firms widen model mix as US AI curbs raise risks

Limits on access to some US AI services are accelerating efforts by major European companies to reduce reliance on any single provider. The US government ordered Anthropic to suspend access to its Fable 5 and Mythos 5 models for foreign nationals, a move executives said underscored the risks of proprietary services that can be cut off remotely.

Siemens, Renault Group, Orange and ChapsVision said they already use a mix of US, Chinese and European models. Siemens uses DeepSeek, Alibaba’s Qwen and Nvidia’s Nemotron alongside other models, while Renault Group works with Google, Microsoft, Mistral, DeepSeek and Dataiku. Executives framed sovereignty as flexibility and credible alternatives rather than isolation from foreign technology.

Europe still has few general-purpose AI providers, led by Mistral, while companies such as DeepL remain stronger in narrower fields. Open-source and open-weight models are gaining attention because they can run on company servers, limiting data exposure and provider control.

Rising token costs are adding pressure as companies deploy software agents that process more information automatically. Orange warned that cost per token is becoming a central concern, while Renault Group said sharply higher token costs are forcing it to adapt its AI strategy.