NVIDIA unveils modular AI factories and agent tools at GTC Berlin

NVIDIA introduced infrastructure and software for Europe’s AI economy at GTC Berlin, with CEO Jensen Huang emphasizing systems that support autonomous agents and physical AI. Its new DSX POD reference design packages complete Vera Rubin AI factories into modular containers, integrated and validated before delivery. NVIDIA says onsite deployment can shrink from six months to just 14 days once site utilities are ready. The company also says its DSX platform can reduce cooling water consumption by up to 100% and deliver the same AI output with up to 30% less power.

Agent security was another focus. The NVIDIA OpenShell Reference Architecture for Safe AI Agents uses an open source, policy-controlled runtime with kernel-level sandboxing to restrict and monitor access to systems. The new Agent Knowledge Foundation connects agents to enterprise documents and databases, while Kumo models make predictions from structured data.

Huang highlighted applications in drug research, factory planning and engineering, alongside work with Siemens Healthineers on adaptive radiation therapy. Volvo Cars is adopting Hyperion and Alpamayo Stack software for automated driving, while other mobility companies demonstrated potential vehicle and robotaxi applications.

For robotics researchers, NVIDIA announced Agile ONE SN, an Isaac GR00T-based reference humanoid expected next year. Huang also showcased RTX Spark laptops, launched last week, and demonstrated a workflow combining local simulation, cloud training and deployment to Hugging Face’s Microduck robots.