Summary
At Oracle AI World on October 27, 2026, Oracle and NVIDIA announced OCI superclusters will scale Vera Rubin NVL72 to multi-gigawatt deployment, with Vera Rubin NVL72 and Vera CPU reaching GA on OCI the same day; the planned HGX Rubin NVL8 pairs dual 64-core Xeon 6 with 4TB DDR5, and the Abilene, Texas supercluster already hosts over 100,000 Blackwell GPUs.
Background and Data Scope
- Event: At Oracle AI World in Las Vegas on October 27, 2026, Oracle and NVIDIA announced that OCI superclusters will scale Vera Rubin NVL72 to multi-gigawatt scale; Vera Rubin NVL72 and Vera CPU reached general availability on OCI the same day, per Oracle's official press release.
- Sources: Deployment form and node configuration data come from Oracle's official press release (2026-10-27).
- Scope: Each NVL72 rack interconnects 72 GPUs via NVLink scale-up; Oracle's multi-plane network scales across hundreds of thousands of GPUs; the transaction value was not disclosed.
Core Data Table
| model | deployment form | GPUs per node | CPU/memory | network | current status |
|---|---|---|---|---|---|
| Vera Rubin NVL72 | rack-scale OCI supercluster | 72 GPUs per rack (NVLink scale-up) | — | OCI multi-plane network, scales across hundreds of thousands of GPUs | GA on 2026-10-27 |
| HGX Rubin NVL8 | x86 server node (planned) | 8 Rubin GPUs | dual 64-core Intel Xeon 6, 4TB DDR5 memory | 6.4 Tbps RDMA network | planned |
| GB200 / GB300 NVL72 | existing rack-scale deployment | 72 GPUs per rack | — | Oracle Acceleron + NVIDIA Spectrum-6 switches + BlueField-4 DPU | deployed |
| Abilene OCI supercluster (Texas) | hyperscale supercluster | over 100,000 NVIDIA Blackwell GPUs | — | OCI supercluster network | OpenAI trains and serves GPT-6 Astra here |
Supplementary Data
- Jensen Huang previously said each gigawatt of Vera Rubin deployment can bring NVIDIA up to about $40 billion in revenue; Oracle expects customer-related chip spending of about $90–95 billion in fiscal 2027 (per IT Home).
- The value of this transaction was not disclosed.