During the GTC Paris conference, held alongside VivaTech, Jensen Huang, NVIDIA’s founder and CEO, presented a vision of Europe transitioning from adopting artificial intelligence to building its industry and infrastructure. Huang said that AI has become a new industry and part of an “intelligence infrastructure” that every country and community will use.
The presentation covered the growth of inference workloads, quantum computing, infrastructure, agentic AI, and robotics, with a focus on NVIDIA’s partnerships with European governments, telecommunications companies, and cloud service providers.
Blackwell Systems and AI Factories
Huang said that the GB200 NVL72 system, which he described as “one giant GPU,” had entered full production and powered applications including sovereign models and quantum computing. According to his presentation, the system was designed to be a “thinking machine” capable of inference and planning and of spending more time interacting with itself.
NVIDIA’s partners are currently producing 1,000 GB200 systems per week, Huang said, while the available systems include NVIDIA DGX Spark and rack-mounted RTX PRO Servers. The company is working with countries to build two types of infrastructure: AI infrastructure that provides services to third parties, and “AI factories” that companies build for their own use and to generate revenue.
NVIDIA also announced the expansion of its network of technology centers in Europe, including new centers in Finland, Germany, Spain, Italy, and the United Kingdom, with the goal of accelerating skills development and growth in quantum computing.
Quantum Computing and Sovereign Models
NVIDIA’s CUDA-Q platform is now available on Denmark’s Gefion supercomputer, opening the way for applications that combine AI and quantum engineering. The company also announced the availability of CUDA-Q on NVIDIA Grace Blackwell systems, alongside its collaboration with supercomputing centers and quantum hardware developers to support hybrid-computing research and quantum error correction.
Huang showcased NVIDIA Nemotron models designed to build large language models suited to local needs. These models will be made available to the Perplexity inference search engine, enabling the deployment of multilingual AI while taking into account each country’s language, culture, and sensitivities, according to the presentation.
Agents and the Industrial Cloud
NVIDIA presented a set of blueprints for building agentic AI, including an agent safety blueprint aimed at businesses and governments, as well as the NVIDIA NeMo Agent toolkit and a blueprint for building data loops that feed agent development. The company will work with governments, telecommunications companies, and cloud providers to deploy the DGX Cloud Lepton platform in the region, integrating it with Hugging Face to provide direct access to global computing capabilities.
In the industrial sector, NVIDIA announced the launch of the world’s first industrial AI cloud, to be built in Germany to help European manufacturers perform simulation, automation, and optimization at scale, using digital twins and the Omniverse platform.
Huang said that NVIDIA DRIVE, its end-to-end autonomous-driving platform, had entered production to accelerate the large-scale deployment of safe and intelligent transportation. He also discussed the Newton engine for training robots in physics, developed by NVIDIA in partnership with DeepMind and Disney.
Inference Growth and the Next Phase
Huang estimated that the number of inference users had risen from 8 million to 800 million, a hundredfold increase over a few years. He believes this growth requires computers designed for thinking and inference, with Blackwell systems adopted inside a new class of data centers he called “AI factories,” which generate the tokens that form the raw material of modern AI.