Summary
NVIDIA and Oracle say Vera Rubin NVL72 systems and NVIDIA Vera CPUs are generally available on Oracle Cloud Infrastructure. The companies also plan a multi-gigawatt expansion and an eight-GPU Rubin system.
NVIDIA and Oracle said on October 27, 2026, that NVIDIA Vera Rubin NVL72 systems and NVIDIA Vera CPUs are generally available on Oracle Cloud Infrastructure (OCI). They also plan to expand OCI Superclusters using the systems to multi-gigawatt scale, aiming to support AI training, inference and agent workloads.
The announcement combines two kinds of computing capacity: GPUs for intensive model training and serving, and CPUs for tasks agents perform around model calls. NVIDIA and Oracle say customers can scale those resources independently to better match different workloads.
GPU scale and a planned eight-GPU system
OCI Superclusters combine NVIDIA NVL72 rack-scale systems with Oracle Acceleron networking, NVIDIA Spectrum-X Ethernet networking, Spectrum-6 switches and BlueField-4 data processing units. NVIDIA says each NVL72 rack connects 72 GPUs through NVLink, its scale-up interconnect. OCI’s network is designed to scale across hundreds of thousands of GPUs, while the systems are paired with gigawatt-class power and cooling infrastructure.
The planned Vera Rubin expansion builds on OCI’s existing NVIDIA GB200 and GB300 NVL72 systems. NVIDIA says its shared CUDA software foundation and ongoing software optimisation are intended to help customers carry workloads across hardware generations.
Oracle also plans to offer NVIDIA HGX Rubin NVL8, a smaller configuration with eight Rubin GPUs connected through NVLink in an x86-based server. The planned system includes two 64-core Intel Xeon 6 CPUs, 4TB of DDR5 memory and 6.4 Tbps of high-speed RDMA networking. Oracle describes it as an option for AI training, fine-tuning and inference that can run x86-based applications.
CPUs handle work between model calls
AI agents do more than generate text or other model outputs. They may call tools, execute code and retrieve data; these operations can require substantial CPU work. If those steps take longer, GPUs can wait before they are needed for the next model call.
NVIDIA and Oracle say Vera CPUs on OCI are intended to accelerate such tasks through high single-thread performance and memory bandwidth. OCI Kubernetes Engine lets customers scale CPU and GPU resources separately, while Oracle Acceleron provides hardware-based isolation for workloads running untrusted, AI-generated code.
OpenAI, which had early access, said it is bringing Vera CPUs into production on OCI for agent sandbox workloads. The announcement also cited OpenAI’s use of an OCI Supercluster with more than 100,000 NVIDIA Blackwell GPUs to train and serve GPT-6 Astra in Abilene. These examples illustrate the distinct roles of the infrastructure: GPUs provide large-scale model compute, while CPUs can run the surrounding code and tools.