How France Is Building a Sovereign AI Ecosystem for Agentic Computing

French companies are deploying agentic computing across space, financial services, the public sector, retail, telecommunications and scientific research. Together, these applications show how France’s artificial intelligence ecosystem is moving from ambition to real-world implementation.

France’s progress is built on a planned foundation. The AI Action Summit in February 2025 and the Choose France Summits in 2025 and 2026 strengthened the country’s investment in computing infrastructure and AI. Since then, French organizations have developed capabilities across the entire AI stack, from sovereign infrastructure and open models to industry-specific applications.

Sovereign AI infrastructure in France

France-based companies are expanding their agentic AI infrastructure using NVIDIA technology.

Bull is expanding its production of AI servers in Europe to include NVIDIA Vera Rubin Ultra NVL72 systems, planned for 2028. The systems will use trays produced in Europe, with racks integrated at Bull’s Angers factory starting Friday, October 2.

Altair-1, the first satellite in a planned $1 billion constellation of 50 autonomous satellites, launched this month and successfully reached orbit. The spacecraft was integrated by Loft Orbital and built in Abu Dhabi by Orbitworks, a joint venture with Loft Orbital and Marlan Space.

The satellite uses an NVIDIA Jetson AGX Orin edge module and runs a Mistral model for real-time Earth observation and autonomous operation in orbit.

Open models and sovereign AI

NVIDIA Nemotron, an open model, is being used across French organizations that require sovereign and trusted data management.

Linagora and Dokaposte have secured the CANUT Framework Agreement. Twake.ai, built on NVIDIA Nemotron, is the only referenced sovereign digital workplace for more than 2,500 French public institutions.

In financial services, SBS is collaborating with NVIDIA to launch a sovereign AI platform. The platform enables banks to deploy managed generative AI and agent services within their own environments.

Last week, Mistral introduced the Mistral Large 4 open model in private preview. The model features open-weight performance for cyber defense, manufacturing, finance and multimodal tasks, providing a foundation for customizing AI models across key industries.

With support from NVIDIA, the Inria Foundation is launching a working group focused on adopting open-source advanced models and tools. Its findings are expected to be published next spring.

Agentic AI applications across French industries

The same pattern is emerging across retail, public services, telecommunications and scientific research. Open models combined with NVIDIA infrastructure are helping French organizations develop and deploy AI applications with measurable results.

L’Oréal: AI-accelerated cosmetic research

L’Oréal is moving cosmetic formulation discovery from laboratory chemistry toward open-source, in-silico screening. The company’s GPU cluster runs the NVIDIA Alchemi toolkit, which aims to make discovery up to 100 times faster.

Mirakl: faster product catalog onboarding

Mirakl has onboarded more than 70 million products using fine-tuned open models running on NVIDIA GPUs and inference software. The technology has reduced seller catalog onboarding to less than two hours.

AI-generated visual marketing

PhotoRoom and Vestiaire Collective demonstrate the potential business impact of advanced visual models on market performance.

Omi combines approved 3D digital twins with NVIDIA-optimized open models to deliver three-times-faster inference. This helps brands bring accurate products to AI-generated marketing visuals at scale.

Digital twins for infrastructure and public services

ThinkDeep and Akila from Côte d’Or built a digital twin agent using NVIDIA Cosmos, the Nemotron open model and Omniverse libraries.

The system manages approximately 5,800 kilometers of roads and buildings, reducing total operating costs by 10%. The project aims to reach a 30% reduction as deployment expands.

Paul Sud de France uses ThinkDeep’s DeepBrain platform, powered by Nemotron and Cosmos, to support goods processing and port security.

Paris-based robotics company Enchanted Tools is deploying social robots to support patients in medical settings, including pediatric cancer treatment.

Telecommunications and employee-built AI agents

Bouygues Telecom and Prisme.ai announced AI Studio V2, powered by NVIDIA’s full-stack infrastructure. The platform allows employees to build their own agents in a secure and controlled environment.

Quantum computing and GPU-accelerated error correction

In quantum computing, Alice & Bob has integrated its superconducting Cat qubit platform with NVIDIA NVQLink. The integration demonstrates how GPUs can improve quantum error correction and accelerate work across different qubit modalities.

Quandela is also using NVQLink with DGX Spark systems and photonic qubits. The work advances QPU-GPU supercomputing and supports research into hybrid quantum-classical applications.

France’s next phase of AI development

From sovereign cloud and open models to autonomous satellites, digital twins, robotics and quantum computing, France’s AI ecosystem is increasingly focused on deployment. The combination of domestic companies, open AI models and NVIDIA infrastructure is creating a foundation for agentic computing across critical industries.

Hear from Jensen Huang, founder and CEO of NVIDIA, in the GTC Berlin keynote speech on Wednesday, October 21, at 11 a.m. CET.