Nvidia and Oracle expand Vera Rubin to multi-gigawatt scale
Nvidia and Oracle have announced on October 27, 2026 that Oracle Cloud Infrastructure Superclusters will grow to multi-gigawatt capacity using Nvidia's Vera Rubin platform.
The announcement goes beyond a promise. Vera Rubin NVL72 racks and Nvidia's Vera CPUs are now available on OCI, so customers can rent the hardware today rather than read about it in a roadmap. This expansion builds on Oracle's earlier deployments of Nvidia Blackwell NVL72 systems, with production shipments of Vera Rubin beginning to ramp in fall 2026.
Technical specifications
Vera Rubin is the successor to Nvidia's Grace Blackwell generation. The flagship configuration is the NVL72 rack, which packs dozens of GPUs into a single cabinet connected by NVLink 6. Each GPU uses HBM4 memory supporting up to 288 GB and 22 TB/s of bandwidth.
Oracle will wrap these racks in advanced Ethernet networking and Nvidia's BlueField-4 DPUs, which handle networking, storage and security tasks. The deal also includes HGX Rubin NVL8 systems for smaller eight-GPU configurations aimed at frontier model training and inference.
Performance claims
Nvidia promises significant generational jumps with Vera Rubin designed to deliver up to 30 times higher inference throughput per watt compared to Blackwell. For training workloads, it's projected to deliver four times the performance of the previous generation.
The OCI setup allows CPU and GPU resources to scale independently, giving customers more flexibility than fixed processing bundles. Nvidia estimates a $40 billion revenue opportunity per gigawatt for the Vera Rubin platform.
Related
