Skip to main content

Menu secondaire

  • News
  • Events
  • Newsletter

Réseaux sociaux

  • LinkedIn
  • Twitter
  • YouTube
  • Canal-U
  • Flux RSS
  • Bluesky
  • Instagram
  • Contact
    • FR
Logo GENCI
Logo GENCI

Menu principal

  • Learn about GENCI
    • About us
    • Our ecosystem
    • HPC.AI.Quantum
    • Towards exascale
    • Our reports and publications
  • Services
    • Computing resources
    • For academic researchers
    • At the service of companies
    • Training
    • Contact us
  • Results and projects
    • Scientific Focus

Je souhaite...

  • Submitting or renewing a request for resources
  • Learn more about high-performance computing, quantum computing and AI
  • Contact GENCI

Breadcrumb

  1. Home
  2. node
  3. DALIA’s NVIDIA GB200 NVL72 Fuels New AI Services in France in Its First Year

DALIA’s NVIDIA GB200 NVL72 Fuels New AI Services in France in Its First Year

Since entering in production in November 2025 as the first NVIDIA GB200 NVL72 system in Europe, DALIA’s supercomputer has driven a wave of new AI services for GENCI and AI Factory France including training AI For Science open models for academia and industry, processing data from large scale scientific instruments, running hackathons and for French Universities and engineering schools‘ trainings, experimenting federated learning across French supercomputers, building a first public/private continuum with Scaleway hyperscaler and will soon anchor a sovereign agentic AI offer addressing AI Co-Scientists.

16 November 2026

    Co-acquired by GENCI and AMIAD, the French Ministry of Defense’s AI Agency, DALIA supports dual-use AI, one of the 12 verticals of AI Factory France. Hosted and operated by IDRIS-CNRS, the GB200 NVL72 system has already supported nearly 40 open research projects from academics and startups since November 2025, leveraging NVIDIA Blackwell GPUs with NVFP4 and FP8 precisions, a 72-GPU NVIDIA NVLink domain and GB200 Superchips pairing two B200 GPUs and one NVIDIA Grace CPU via NVIDIA NVLink-C2C.Dalia is also built with NVIDIA AI Enterprise software and NVIDIA Mission Control, offering a turnkey operational environment that simplifies deployment and management for developers, researchers, and system administrators.

    Among the projects supported is Raidium’s FoundRa, a multimodal foundation model for precision radiology that combines radiological images and reports. Developed by Raidium, a French health startup member of NVIDIA Inception, it aims to improve diagnostic accuracy, workflow efficiency, and personalized patient care with selected model weights to be released to the community democratizing access to advanced AI. 

    In the science of humanities, Sciences Po researchers are evaluating how well open-source LLMs can simulate individual political opinions to predict French public opinion, using post-election ELIPSS survey data.

    By comparing four levels of input, from no personal data to detailed demographic and political profiles, and measuring individual accuracy, sample-level accuracy, and the ability to reproduce human response variability, the project will clarify the potential and limits of generative agents for political science and policy evaluation

    Finally in sovereign LLMs, LINAGORA and the OpenLLM consortium trained on Jean Zay LUCIOLE a set of multilingual SLMs of 1B to 23B parameters including Hybrid Mamba-Transformer architectures adapted to edge devices to larger reasoning and synthesis tasks and they will benefit from the support of NVFP4 on DALIA.

    An enabler of new AI services in France and in Europe 

    Beyond supporting scientific and industrial projects, DALIA has shaped many new AI services now offered by GENCI and AI Factory France including: 

    • A new orchestration layer based on SchedMD Slinky unifying Slurm and Kubernetes, enabling an elastic use of DALIA by allocating compute partitions on the fly to HPC, AI, training, hackathons and instrument data ingestion workloads. 
    • Training sessions and hackathons, including the first world models hackathon at VivaTech 2026, where more than 200 students worked with models from AMI Labs, the startup founded by Yann LeCun early 2026. DALIA has also joined MESONET, the network of computing resources from the French Universities for AI training.
    • Support to European ODISSEE, enabling real-time ingestion data from large-scale scientific instruments such as SKA radiotelescope. OSKAR, a simulator of the SKA Interferometer, developed by the French ECLAT consortium to simulate real SKA data workflows is now fully operational on DALIA and its Blackwell GPU.
    • Through the ConcordIA collaboration with BriCS – Bristol Supercomputing Center, DALIA validated the deployment of NVIDIA FLARE for federated learning across distributed sites, in France, soon extending to UK and also German HPC centers, enabling the training of larger models or models using sensitive data that cannot be shared.
    • At CNRS AISSAI Hackathon on “AI for Particle Accelerators” - October 2026- held with the ARTIFACT (ARTificial Intelligence For Accelerators, user Communities and associated Technologies) network, 40 participants  - from 9 French research organizations (CEA, CNRS/IJCLab, GANIL, GIP ARRONAX, IDRIS, Synchrotron SOLEIL) and 14 international institutions (CERN, ESS, KEK, STFC, BNL, Elettra, HZB, KIT, etc.) plus 3 startups (Cosylab, Hopsworks, Disobedient Geometry Labs) - used DALIA for challenges including anomaly detection, beam power and size optimization, electron beam charge prediction from laser parameters and AI surrogates for PIC numerical simulations.

    "Slinky represents the future of supercomputer orchestration, enabling the rigor required for high-performance computing or the training of large AI models to coexist with the flexibility needed to accommodate new communities or AI services. Slinky optimizes the utilization of computing resources by allowing these two worlds to operate together on the same silicon" said Antoine Regnier, AI infrastructure engineer at GENCI.

    Paving the path to sovereign AI for Science tools offering

    DALIA is only the beginning. Its capacity will soon double thanks to a new collaboration with French universities, laying the first foundations of a sovereign agentic AI infrastructure for AI co-scientists. In February 2027, it will power the first ‘1000 Scientists AI Jam’, connecting scientists across at least four European (France, Italy, Spain and Germany) and four Latin American and Caribbean (Chile, Brazil, Mexico and Costa Rica) countries with collaborators like NVIDIA. 

    Together with Jean Zay and Adastra, DALIA helps GENCI and AI Factory France support nearly 2,000 AI projects in 2026, an unprecedented number in Europe. 

    Figure 1 - Picture of DALIA, a NVL72 supercomputer at  IDRIS-CNRS

    About GENCI

    Created by the public authorities in 2007, GENCI (Grand Équipement National de Calcul Intensif) is a major research infrastructure, a public operator aiming to democratize the use of digital simulation through high-performance computing associated with the use of artificial intelligence and quantum computing to support French scientific and industrial competitiveness.

    GENCI pursues three missions :

    • Implement the national strategy of equipping France with high-performance computing, storage and massive data processing resources associated with AI (Artificial Intelligence) technologies and future quantum computing technologies for the benefit of French and European open scientific research, drawing on the three national computing centers;
    • Support the development of an integrated ecosystem on a national and European scale 
    • Promote the use of numerical simulation, artificial intelligence and quantum computing among academic and industrial teams through open research.

    GENCI is a civil company owned 49% by the French State represented by the Ministry in charge of Higher Education, Research and Space, 20% by CEA, 20% by CNRS, 10% by the Universities represented by France Universités and 1% by Inria.

    About AI Factory France

    AI Factory France is a European-scale platform co-funded by the French government and the EuroHPC Joint Undertaking under grant agreement No 10126039. It sets a one stop shop providing public and private actors with the computing resources and related services (training, mentoring, technical support) they need to accelerate their research on AI development and deployment. Led by GENCI, it brings together a consortium of 16 prestigious partners including large research organizations (Inria, CNRS, CEA, AMIAD), French Universities and innovation partners, fostering AI adoption among 12 verticals including Health, Materials Science, Robotics, Defense & Cybersecurity, Finance, Earth Science or Aerospace.

    More info here: https://aifactory-france.eu

    Media Contacts

    GENCI: nicolas.belot@genci.fr

    Share

    Opens in a new window Opens in a new window

    6 bis rue Auguste Vitu

    75015 PARIS

    +33 1 42 50 04 15

    Menu footer

    • Join us
    • Public procurement
    • Newsletters
    • Terms of use
    • Site map
    • Cookies

    Follow us

    Réseaux sociaux

    • LinkedIn
    • Twitter
    • YouTube
    • Canal-U
    • Flux RSS
    • Bluesky
    • Instagram