NVIDIA's Vera Rubin platform enters full production as Jensen Huang highlights agentic AI as the next computing paradigm

NVIDIA announced that Vera Rubin—a multi-rack, pod-scale system built for agentic AI workloads—is now in full production. The platform integrates Rubin GPU systems, Vera CPU racks, BlueField-based storage and security systems, and NVIDIA networking, with components from TSMC, Micron, SK hynix and Samsung. Microsoft, Dell and CoreWeave have operational Vera Rubin NVL72 engineering racks.

The Vera CPU, designed for agentic workloads (AI systems that reason, plan, use tools and complete tasks), delivers 1.8 times the agentic sandbox performance of x86 CPUs, according to NVIDIA. Huang said the company is building "millions and millions of Veras" with Taiwan's ODMs and computer makers.

NVIDIA introduced Nemotron 3 Ultra, an open model combining state-space models with Mixture of Experts and described as faster and cheaper to run than other open models. The company also announced Cosmos 3, a foundation model for physical AI that can function as a vision-language model, world model, simulator and foundation for action models.