European AI Sovereignty Finds a Poster Child in Mistral

Notes from the Mistral AI Now Summit are circulating on HN, and the concrete use cases are interesting. BNP Paribas is running Mistral models on-prem for KYC in Belgium, keeping sensitive data inside the bank's walls. Abanca is using agent orchestration for 2 million customers. These are not experiments. These are production deployments from regulated financial institutions that would not touch OpenAI or Anthropic models for this work.

The pattern is European data sovereignty driving real enterprise AI adoption. The comment thread shows genuine enthusiasm from European builders who see Mistral as the rare case of a credible alternative to US hyperscaler AI, with the transparency and on-prem deployment story that compliance teams can actually accept. Liquid AI also dropped an 8B MoE model trained on 38 trillion tokens today, adding another datapoint to the 'efficient smaller models are getting competitive' story.

The 'homeopathic AI' joke in the Liquid AI thread signals some skepticism about marketing-driven benchmark claims, but the underlying architecture interest, particularly for vision-language-action models and sparse inference, is genuine among researchers.