Nvidia and Oracle announced on October 27, 2026 that Oracle Cloud Infrastructure superclusters will reach multi-gigawatt capacity using Nvidia’s Vera Rubin platform. Vera Rubin NVL72 racks with Nvidia Vera CPUs are already available on OCI, while HGX Rubin NVL8 configurations with eight GPUs are aimed at frontier model training, inference and agentic AI workloads. Each Vera Rubin GPU features up to 288 GB of HBM4 memory and 22 TB/s of bandwidth, with NVLink 6 for chip-to-chip interconnection and BlueField-4 handling networking, storage and security. Nvidia says Vera Rubin can deliver up to 30 times higher inference throughput per watt than Blackwell and four times its training performance, and estimates a $40 billion revenue opportunity per gigawatt for the platform.
Source: Read the original article
