NVIDIA CEO Jensen Huang announced at GTC Berlin 2026 that the Vera Rubin platform has entered full production, marking a strategic shift from selling GPUs to delivering complete "AI factories." The new architecture features 100% liquid cooling, connects 72 GPUs per rack into supercomputers, and is projected to deliver 30x performance gains within two years. NVIDIA also unveiled the DSX POD modular AI factory reference design, which reduces on-site deployment time from six months to just 14 days while saving nearly 100% of water and up to 30% of electricity for equivalent output. Huang introduced the OpenShell security reference architecture for AI agents, emphasizing kernel-level sandboxing and least privilege as critical infrastructure components. The keynote showcased six open models—including Nemotron, Cosmos 3.5, and GR00T—and revealed physical AI products such as the Agile ONE SN humanoid robot and Hyperion autonomous driving platform, set for adoption by Volvo, Volkswagen, JLR, and Mercedes-Benz. Google Cloud confirmed that A5X instances powered by Vera Rubin NVL72 will launch next month with plans to deploy up to 960,000 Rubin GPUs. The event underscored NVIDIA's push to position Europe as an AI producer rather than consumer, following Huang's week-long tour across Paris, London, Stockholm, and Dublin where he pledged nearly $10 billion to the European technology ecosystem. Key partnerships include AstraZeneca, Ericsson, Saab, and SEB building the Sferical One supercomputer in Sweden, Siemens investing €500 million in digital twin capabilities, and T-Mobile and Nokia completing the first AI-driven commercial call using CUDA infrastructure.