NVIDIA announced a modular data-center reference architecture called the DSX POD and committed nearly $10 billion to European technology partnerships during GTC Berlin, according to a report published by NVIDIA on Wednesday.
Chief executive Jensen Huang confirmed that Vera Rubin computing systems have entered full production. Partners will build the new DSX POD reference designs inside prefabricated containers that undergo validation before delivery. NVIDIA stated that this modular container system reduces onsite data-center deployment times from six months to 14 days once power and site utilities are connected. The platform enables operators to eliminate water consumption for cooling by up to 100% and run AI workloads using up to 30% less electricity.
European enterprises working with NVIDIA on AI include L’Oréal, Revolut, Spotify, Merck, Vodafone, Zara, Porsche, Nestlé, and HSBC. To run autonomous software inside corporate IT environments, NVIDIA introduced the OpenShell Reference Architecture for Safe AI Agents. The open-source runtime applies kernel-level sandboxing between automated software agents and external systems, resetting and enforcing access permissions dynamically on every launch.
Enterprise tools and agents
Enterprise data access will rely on the newly introduced NVIDIA Agent Knowledge Foundation, which links software agents to company databases. The toolkit pairs Nemotron language models for unstructured text processing with newly announced Kumo models designed to make predictions from structured tabular data. NVIDIA also confirmed that more than 11 million developers now build on its CUDA-X library ecosystem.
Industrial demonstrations highlighted autonomous agent workflows across European manufacturing and research facilities. Lila Sciences uses virtual experiments and software agents to plan B-cell lymphoma therapies, Siemens tests automated production changes in a digital twin at its Erlangen factory, and Dassault Systèmes runs simulation agents to alter vehicle body designs.
Telecom, automotive, and robotics
Network operators are also converting physical infrastructure into distributed compute nodes. Working with Nokia, NVIDIA carried out its first commercial voice call over T-Mobile’s telecommunications network using CUDA-accelerated infrastructure across an environment targeting 100,000 distributed sites.
Automakers expanded their adoption of the NVIDIA Hyperion computing platform for level 3, level 4, and robotaxi vehicle fleets. Bolt and Lucid, Mercedes-Benz, and Volkswagen Commercial Vehicles demonstrated autonomous mobility projects on Hyperion, while Volvo Cars committed to using both Hyperion hardware and NVIDIA Alpamayo Stack software for automated driving. Next-generation vehicles from Jaguar Land Rover are also set to support the Hyperion platform.
Robotics researchers received a reference hardware platform called the Agile ONE SN. Built on Isaac GR00T, the modular humanoid body includes 31 degrees of freedom and is scheduled to become available next year to cut initial development setup times. Closing the presentation, Huang demonstrated Hugging Face’s Microduck robots trained inside Isaac Sim using RTX Spark laptops, which feature 128 gigabytes of memory and one petaflop of processing capacity.
