Summary
On October 27, 2026, Nvidia and Oracle announced that OCI Superclusters will run Vera Rubin NVL72 and expand to multi-gigawatt scale; Vera Rubin NVL72 and Vera CPU went generally available (GA) on OCI the same day, and OpenAI already trains GPT-6 Astra on over 100,000 Blackwell GPUs in the Abilene OCI Supercluster.
Scale and Availability
What does "multi-gigawatt" deployment mean in concrete terms?
Per Nvidia's official press release, OCI Superclusters will run Vera Rubin NVL72 and expand to multi-gigawatt scale. Per the release, OCI's multi-plane networking scales to hundreds of thousands of GPUs, backed by gigawatt-class power and cooling, Oracle Acceleron networking and Spectrum-X Ethernet (Spectrum-6 switches, BlueField-4 DPUs).
Is Vera Rubin NVL72 already available on OCI?
Yes. Per Nvidia's press release, Vera Rubin NVL72 and Vera CPU are generally available (GA) on OCI starting October 27, 2026.
What makes up an OCI Supercluster?
Per Nvidia's press release, a single rack interconnects 72 GPUs via NVLink, and OCI's multi-plane networking scales to hundreds of thousands of GPUs. Networking is paired with Oracle Acceleron and Spectrum-X Ethernet (Spectrum-6 switches, BlueField-4 DPUs), with gigawatt-class power and cooling design.
New Configuration and CPU
What is the planned HGX Rubin NVL8 configuration, and for whom?
Per Nvidia's press release, a planned HGX Rubin NVL8 pairs 8 Rubin GPUs interconnected via NVLink inside an x86 server, with two 64-core Intel Xeon 6 CPUs, 4TB of DDR5 memory and 6.4 Tbps of RDMA networking. It targets training, fine-tuning and inference workloads.
What problem does Vera CPU solve for agent workflows?
Per Nvidia's press release, Vera CPU targets CPU-intensive tasks between agent model calls (tool use, code execution, data retrieval). It supports independent CPU and GPU scaling to reduce GPU idle time.
Customers and Generations
What is OpenAI already running on Oracle Cloud, and at what scale?
Per Nvidia's press release, OpenAI trains and serves GPT-6 Astra on over 100,000 Blackwell GPUs in the Abilene OCI Supercluster. OpenAI engineering leadership has stated it will put Vera into production.
How does Vera Rubin relate to the earlier GB200/GB300?
Per Nvidia's press release, Vera Rubin NVL72 is the next rack-scale platform after Blackwell (the GB200/GB300 families). OpenAI has deployed over 100,000 Blackwell GPUs in Abilene and stated it will move the new Vera platform into production; the two generations extend sequentially on OCI. Specific cross-generation performance comparisons are not given in the release.