Nvidia reframes AI as factory, agents as workload

At GTC Berlin, Nvidia CEO Jensen Huang argued AI constraints have shifted from chip supply to deployment speed, power, and security. Nvidia introduced DSX POD, a pre-cabled reference design for Vera Rubin AI factories that cuts site deployment from six months to 14 days. The company also launched the OpenShell Reference Architecture for agent security, expanded its Agent Toolkit with an Agent Knowledge Foundation and Kumo structured-data models, and deepened telecom partnerships. Huang framed a delayed 1-gigawatt AI factory as costing roughly $10 billion in lost annual revenue, underscoring infrastructure economics as the new competitive battleground.