Résumé IA
L'intelligence artificielle s'impose comme l'infrastructure technologique centrale de notre époque, portée par un écosystème diversifié de modèles — grands et petits, ouverts et propriétaires, généralistes et spécialisés. Lors d'une session spéciale consacrée aux modèles ouverts à la conférence NVIDIA GTC, Jensen Huang, fondateur et PDG de NVIDIA, a résumé la situation en une phrase : « Propriétaire versus open source n'est pas un débat. C'est propriétaire et open source. » Pour illustrer cet engagement, NVIDIA a annoncé la création de la Nemotron Coalition, une collaboration mondiale inédite regroupant des laboratoires d'IA et des développeurs de modèles pour faire avancer les modèles fondamentaux ouverts. Le premier projet issu de cette coalition sera un modèle de base codéveloppé par Mistral AI et NVIDIA, dont les membres apporteront données, évaluations et expertise sectorielle. Les modèles Nemotron ont déjà été téléchargés plus de 45 millions de fois sur Hugging Face, plateforme sur laquelle NVIDIA est désormais la plus grande organisation avec près de 4 000 membres d'équipe. Plusieurs panels réunissant des figures majeures du secteur — dont Mira Murati (Thinking Machines Lab), Aravind Srinivas (Perplexity), Michael Truell (Cursor) et Arthur Mensch (Mistral) — ont dégagé des tendances clés. Les agents IA s'apprêtent à devenir de véritables collègues capables de mener des tâches complexes sur plusieurs jours. L'IA n'est plus un modèle unique mais un système orchestré : « ce que vous voulez, c'est une orchestra multimodale, multi-modèles et multi-cloud », a déclaré Srinivas. L'ouverture des modèles est présentée comme un moteur d'innovation indispensable, aussi bien pour les grandes entreprises que pour la recherche académique. Murati a insisté sur ce point : « il y a beaucoup d'études à mener qui ne peuvent pas être réalisées uniquement dans les grands laboratoires — c'est là que l'ouverture est précieuse, elle fait avancer la science de l'intelligence. » Cette dynamique s'inscrit dans un tournant structurel où chaque secteur — santé, finance, industrie — a besoin d'une IA adaptée à ses données et workflows spécifiques, rendant la coexistence de modèles ouverts et propriétaires non seulement inévitable, mais souhaitable pour accélérer l'innovation à tous les niveaux.
Impact France/UEMistral AI, acteur français majeur, est cofondateur de la Nemotron Coalition aux côtés de NVIDIA, ce qui renforce son rôle stratégique dans l'écosystème mondial des modèles ouverts.
AI is the defining technology of our time, quickly becoming core business infrastructure. It’s fueled by a diverse ecosystem of models: large and small, open and proprietary, generalist and specialist. This variety is essential for a future where every application will be powered by AI, every country will build it and every company will use it. And it’s not a debate between open versus closed innovation. As NVIDIA founder and CEO Jensen Huang told attendees at a special session on open frontier models at NVIDIA GTC , “Proprietary versus open is not a thing. It’s proprietary and open.” That’s why the future of AI innovation isn’t about a single massive model. Every industry — healthcare, finance, manufacturing — tackles its own unique challenges. They all need AI that can reason about their data and workflows in various ways. And that requires systems of models, tuned and specialized for different modalities, domains and organizations, working together to solve a specific business problem. NVIDIA is a major contributor to open source AI: it’s now the largest organization on Hugging Face , with nearly 4,000 team members . And at GTC, the company announced the NVIDIA Nemotron Coalition , a first-of-its-kind global collaboration of model builders and AI labs working to advance open, frontier-level foundation models through shared expertise, data and compute. The first project stemming from the coalition will be a base model codeveloped by Mistral AI and NVIDIA, with coalition members contributing data, evaluations and domain expertise to support the model’s post-training and continued development. It’ll be shared with the open ecosystem and underpin the next generation of NVIDIA Nemotron models, which have been downloaded more than 45 million times from Hugging Face. Several Nemotron Coalition members joined other leaders building and consuming open models for a back-to-back panel session at GTC. The first panel featured LangChain cofounder and CEO Harrison Chase, Thinking Machines Lab founder and CEO Mira Murati, Perplexity CEO and cofounder Aravind Srinivas, Cursor CEO and cofounder Michael Truell, and Reflection AI cofounder and CEO Misha Laskin. The second included Mistral cofounder and CEO Arthur Mensch, OpenEvidence CEO Daniel Nadler, and Black Forest Labs cofounder and CEO Robin Rombach, alongside Hanna Hajishirzi, senior director of natural language processing at Ai2, and Anjney Midha, founder of AMP PBC. Five key points stood out from the conversation: 1. AI agents are becoming highly capable coworkers. “We’re soon going to see agents really be coworkers that can take on tasks that take many hours or many days, and do incredibly complex workloads,” said Cursor’s Truell. 2. AI is not a single model — it’s an orchestrated system. “What you want is a multimodal, multi-model and multi-cloud orchestra,” said Perplexity’s Srinivas. “All you’ve got to do is delegate your task. You don’t have to worry about which model is good at what — it’s for the orchestration system to figure it out.” 3. Openness fuels innovation across the model ecosystem. “Models are fundamental knowledge infrastructure, and fundamental knowledge infrastructure yearns for openness,” said Reflection AI’s Laskin. “There’s a flourishing ecosystem of powerful, closed models but equally capable open models that are going to be coming over the next couple years.” This combination of open and proprietary models drives advancements at frontier AI companies as well as in academia. “There’s a lot of study to be done, and it cannot be done completely in the large labs,” said Thinking Machines Lab’s Murati. “This is where openness can be very helpful…it advances the science of AI, the science of intelligence.” From left to right: NVIDIA founder and CEO Jensen Huang, LangChain cofounder and CEO Harrison Chase, Thinking Machines Lab founder and CEO Mira Murati, Perplexity CEO and cofounder Aravind Srinivas, Cursor CEO and cofounder Michael Truell, and Reflection AI cofounder and CEO Misha Laskin. 4. Open systems are trustworthy and accessible. “At the end of the day, you’re delegating trust…and it’s much easier to trust an open system,” said AMP PBC’s Midha. With a trusted system, developers can deploy long-running AI agents that can tackle virtually any task. “The models and the systems orchestrating the models are going to get much more capable,” said LangChain’s Chase. “And so you’ll be able to have personal productivity agents that can take on more complex tasks that run for longer.” Open ecosystems also foster collaboration, helping democratize access to AI. “We believe that open-wide models should be the basis for building all the AI software in the world,” said Mistral’s Mensch. “By having an open ecosystem of people that have aligned incentives to create assets that are going to be great for humanity, we can accelerate progress and make sure that everybody gets access in a fair way across the world to artificial intelligence.” From left to
