NVIDIA, through its CEO Jensen Huang, announced during a keynote speech at the GTC Berlin 2026 conference held in Germany, a package of technical innovations and infrastructure aimed at developing the AI economy in Europe. The announcements included the unveiling of new infrastructure platforms, advanced security architecture for intelligent agents, and expanded open models.
Reducing AI Factory Launch Time via "DSX POD"
The company unveiled the "NVIDIA DSX POD" reference design based on the next-generation "Vera Rubin" platform currently in mass production. The concept relies on assembling AI factories within standard containers by partners and conducting pre-integration and verification tests, which contributes to reducing the field deployment period from six months to just 14 days once power infrastructure is available. The "NVIDIA DSX AI Factory Platform" also contributes to reducing cooling water consumption by up to 100% and achieving 30% higher energy efficiency for the same output.
Security Standards for Intelligent Agents Using "OpenShell"
Regarding the security of intelligent systems, the company introduced the "NVIDIA OpenShell Reference Architecture for Safe AI Agents." This architecture relies on an open-source runtime environment called "OpenShell," which acts as a bridge between the intelligent agent and the systems it controls, to impose strict restrictions on agent behavior using kernel-level sandboxes and granular control policies.
Expanding the Portfolio of Open Models and Intelligent Physics Applications
The announcements included expanding the company's range of open models, featuring a new language family that includes "Nemotron 3.5 Super VL" and "Nemotron 3.5 Bolt," in addition to the "Nemotron Retriever" model dedicated to document search and analysis, the "Kumo" model for structured data prediction, the "Cosmos 3.5" world model, and the "Alpamayo 2 Edge" autonomous driving model.
In the robotics and automotive sectors, the company announced that next-generation vehicles from companies such as Volkswagen, JLR, Volvo, and Mercedes-Benz are joining the "NVIDIA Hyperion" computing platform. It also unveiled the "Agile ONE SN" reference humanoid robot based on the "Isaac GR00T" model, equipped with 31 degrees of freedom, which is scheduled to be available to researchers next year to reduce setup times.
What does this development change in practice?
These announcements set new standards for the speed of creating data centers and specialized infrastructure via standard containers, as well as providing tangible solutions for controlling the security of autonomous agents through isolated runtime environments, addressing the operational and security concerns of companies wishing to deploy AI at scale.