NVIDIA ramps Vera Rubin CPU platform into full production
NVIDIA CEO Jensen Huang announced at GTC 2026 that the Vera Rubin CPU platform has entered full production, with shipments scheduled to begin this fall. Vera Rubin integrates Redpanda for real-time streaming and delivers 5.5x lower latencies for agent workloads while pairing with NVIDIA DOCA and BlueField-4 STX for line-speed policy enforcement and visibility. The platform uses HBM memory from Micron, SK Hynix, and Samsung, NVIDIA expanded Vera Rubin's supply chain to twice the size of Blackwell, and rack assembly time dropped from two hours to five minutes. Supermicro released DCBBS designs for NVL72 and HGX Rubin NVL8 supporting 5 MW to 1 GW deployments, and early customers including Anthropic, OpenAI, and SpaceX position Vera Rubin to power large-scale agentic AI factories worldwide.
