France shifts from ambition to implementation in national-scale agentic computing

Announcements from France over recent months show that an entire country has decided to stop talking about sovereign AI and start building it. The AI Action Summit in February 2025 and the Choose France summits in 2025 and 2026 laid the financial and political groundwork, and the local ecosystem now demonstrates a complete supply chain, from servers manufactured in Europe to satellites running models in orbit.

Bull is expanding production of its European AI servers to NVIDIA Vera Rubin Ultra NVL72 systems, with delivery planned for 2028. The boards are made in Europe and rack integration takes place at the Angers factory, officially inaugurated on 2 October. In parallel, the Altair-1 satellite — the first of a planned 50-satellite constellation backed by a $1 billion investment — launched this month and reached orbit carrying NVIDIA Jetson AGX Orin modules at the edge and running Mistral models for Earth observation and autonomous space operations. The satellite was developed by Loft Orbital and built in Abu Dhabi by Orbitworks, a joint venture of Loft Orbital and Marlan Space.

On the software side, NVIDIA's open Nemotron models are already in production at French organisations that require sovereign, trusted data. LINAGORA and Docaposte won the CANUT framework agreement, making Twake.ai, a Nemotron-based digital workplace, the only officially approved solution for more than 2,500 public bodies in France. In the financial sector, SBS is launching a sovereign platform that lets banks run generative and agentic services in their own environment, in partnership with NVIDIA. Last week Mistral announced Mistral Large 4 in private preview: an open-weights model that delivers state-of-the-art performance in cybersecurity, coding, finance and multimodal tasks, and serves as a base for fine-tuning in key industries. The Inria Foundation, with NVIDIA support, is forming a working group on advanced modelling and open-source tool adoption, with findings expected next spring.

The pattern repeats across every vertical: open models on NVIDIA infrastructure yield measurable results. L'Oréal is integrating NVIDIA ALCHEMI into its research processes, letting scientists explore formulation possibilities in silico and better direct lab work, a combination that accelerates discovery up to 100 times versus traditional methods and maximises the potential of proprietary active ingredients. Mirakl has uploaded more than 70 million products to catalogues using fine-tuned open models running on NVIDIA GPUs and inference software, cutting seller catalogue onboarding to under two hours. Photoroom and Vestiaire Collective demonstrate measurable business impact from advanced visual modelling on marketplace performance. Omi enables brands to place accurate products in AI-generated marketing visuals at scale, combining approved 3D digital twins with open models optimised on NVIDIA for a 4x inference speed-up.

In the public sector, robotics and telecoms, digital twins and social robots are moving into production. In Côte-d'Or, ThinkDeep and Akila built a digital-twin agent based on NVIDIA's open Cosmos world models, Nemotron and Omniverse libraries. It manages nearly 5,800 kilometres of roads and buildings, has cut total operating costs by 10% and targets 30% as deployment progresses. Port Sud de France runs cargo processing and port security on ThinkDeep's DeepBrain platform powered by Nemotron and Cosmos. Enchanted Tools, a Paris-based robotics company, is deploying social robots for patient support in healthcare settings, including paediatric oncology. In telecoms, Bouygues Telecom and Prisme.ai have launched — the text breaks off here, but the direction is clear: agents are already operating in production.