프랑스, NVIDIA 기술로 유럽 AI 미래 앞당김

Source: NVIDIA AI Blog

A year ago at NVIDIA GTC Paris at VivaTech, France laid out plans to advance local AI — from new AI factories and national compute capacity to open frontier models and industrial platforms.

Now, that AI infrastructure is coming online. AI agents are running in production, startups are deploying applications and the French AI ecosystem is developing models, datasets and platforms designed around local languages, cultural context and European requirements.

프랑스 AI 인프라가 형태를 갖추다

France’s AI ambitions are gaining momentum. Billions in investment commitments through France 2030, the 2025 AI Action Summit and this year’s Choose France Summit are reinforcing the country’s position as one of Europe’s leading destinations for AI infrastructure.

As part of these efforts, Mistral is building a new 44-megawatt data center in Bruyères-le-Châtel, a commune in northern France. Announced at GTC Paris last year, Mistral’s first deployment is already operational with 18,000 [NVIDIA GB200](https://www.nvidia.com/ en-us/data-center/gb200-nvl72/) systems — laying the foundation for the company’s roadmap of 200 megawatts of compute capacity across Europe by 2027.

The [NVIDIA Blackwell](https://www.nvidia.com/ en-gb/data-center/technologies/blackwell-architecture/) platform is designed to help AI factories maximize throughput within fixed power budgets, combining higher performance‑per‑watt silicon with software features that boost data center throughput in power‑constrained environments.

Mistral is also working with French public investment bank Bpifrance, AI and advanced tech investment company MGX and NVIDIA to expand [Campus AI](https://presse.bpifrance. fr/ bpifrance- mistral- etendent- campus- ai- a- lechelle- nationale- pour- batir- un- reseau- de- 3- gw- d’usines- dia/?lang=fra), a network of AI factories anchored by a planned 1.4-gigawatt facility, making it one of Europe’s largest AI campuses.

This momentum reflects a broader wave of AI infrastructure investment in France.

Scaleway, a European public cloud provider, now offers NVIDIA Blackwell B300‑SXM instances, giving developers and enterprises access to accelerated computing on demand.

Bull and Foxconn have announced the production of [NVIDIA Vera Rubin NVL72](https://www.nvidia.com/ en-gb/data-center/vera-rubin-nvl72/) in Europe. Systems will be manufactured and initially tested at Foxconn’s facilities in the Czech Republic before being assembled, integrated and fully validated at Bull’s factory in Angers, France.

And a consortium of eight leading French companies has submitted a bid to host a European AI gigafactory in France to strengthen European AI infrastructure and accelerate AI adoption.

Meanwhile, [Schneider Electric](https://www.se.com/ ww/en/about-us/newsroom/news/press-releases/ Schneider-Electric-teams-with-NVIDIA-to-develop-validated-blueprints-to-design-simulate-build-operate-and-maintain-gigawattscale-AI-Factories-69b82397e7fa28870e0cd5a3/) has teamed with NVIDIA to develop blueprints for gigawatt-scale AI factories, helping organizations accelerate AI infrastructure deployment.

오픈 모델이 AI 개발을 뒷받침한다

France’s AI ecosystem is producing models, datasets and platforms tailored to local languages, cultural context, and European business and regulatory requirements. As AI agents become more capable, organizations are increasingly adopting systems of models, using the right model for the right task to improve accuracy, reduce costs and accelerate outcomes.

On stage at this year’s VivaTech event, leaders from Gradium, H Company, LINAGORA, Pleias and NVIDIA explored the role of open models in enabling more transparent, customizable and locally relevant AI for governments, enterprises and developers.

“What we see now is a shift from building one isolated model to running continuous model infrastructure, where models train the next models, curate data, generate synthetic environments and verify reinforcement learning,” said Pierre- Carl Langlais, chief technology officer of Pleias. “Open model infrastructure is simply the way to ensure that many people can build AI and frontier-level practice can disseminate throughout the entire economy.”

The [discussion](https://vivatech.com/ sessions/ session/ f9e40c7f- 955a- f111- 8fcb- 6045bd954326) underscored a key theme: combining open models with energy‑efficient infrastructure gives organizations the control they need to inspect, adapt, deploy and audit AI that meets Europe’s compliance and trust requirements.

NVIDIA Nemotron is advancing this with open models, datasets and playbooks that help model builders accelerate workflows from training to deployment.

• Mistral, a founding member of the NVIDIA Nemotron Coalition — a network of AI builders collaborating on open frontier models — is contributing model-development expertise and multimodal capabilities to help advance open frontier models through open collaboration.

• LINAGORA is building multilingual large language models with strong focus on the French language with its Luciole model family, developed using [NVIDIA Nemotron](https://www.nvidia.com/ en-gb/ai-data-science/foundation-models/nemotron/) and [NeMo](https://www.nvidia.com/ en-gb/ai-data-science/products/nemo/) libraries and designed for local language and cultural context.

• Luciole 1B, 8B and 23B were pretrained on Jean- Zay — one of Europe’s most powerful and eco-efficient AI supercomputers — in collaboration with CNRS/IDRIS in the frame of the OpenLLM-France project. These open source models are distributed on Hugging Face (OpenLLM-France) along with their pretraining datasets.

• H Company, also part of the Nemotron Coalition, is developing Holotron, a family of AI agents built on open NVIDIA Nemotron models. These computer-use agents can interact with any software interface in the same way a human would, without needing application programming interfaces or custom integrations, and automate complex enterprise workflows from end to end.

• Pleias, in collaboration with NVIDIA, developed Nemotron-Personas-France and Nemotron-Personas-Belgium, privacy-preserving synthetic persona datasets grounded in French and Belgian demographics and cultural context. The startup is also using Jean Zay train compact language models entirely on open, well-documented datasets, making it easier for customers to address EU AI Act requirements around data provenance and transparency. The team is now building specialized versions for search, retrieval-augmented generation and public sector document workflows.

AI 생산이 성과를 내다

The shift from pilot to production is the defining story of the past year, as organizations across every major industry in France use AI to boost efficiency, quality and speed.

Initiatives like the collaboration — announced at Adopt AI — between AI Factory France (AI2F), led by GENCI, NVIDIA Inception and NVIDIA Connect programs are helping startups gain access to national supercomputing resources, including Jean Zay. Early participants, including Pleias, Nebula and Ryax Technologies, are already turning that access into deployable applications.

In healthcare, Sanofi is deploying AI agents across the value chain, from research, manufacturing and commercial to daily operations like procurement and IT, helping teams automate complex workflows at global scale. The company is also working with startups Owkin and Biolevate to develop autonomous agents for drug discovery and development.

Orange Business, the B2B subsidiary of telecom company Oran —