Oracle and NVIDIA Take Vera Rubin to Multi-Gigawatt Scale on OCI

Multi-gigawatt Superclusters for training and inference

The announcement was made at Oracle AI World in Las Vegas on October 27, 2026. According to the companies, the expansion responds to growing customer demand for AI training, inference and agentic applications.

OCI Superclusters combine NVIDIA NVL72 rack-scale systems, Oracle Acceleron networking with NVIDIA Spectrum-X Ethernet featuring NVIDIA Spectrum-6 switches and NVIDIA BlueField-4 DPUs, and gigawatt-class power and cooling. Each rack connects 72 GPUs through NVIDIA NVLink scale-up networking, while OCI's multiplanar network scales across hundreds of thousands of GPUs, the companies said.

The Vera Rubin expansion builds on existing NVIDIA-powered infrastructure on OCI, including NVIDIA GB200 and GB300 NVL72 systems. The companies point to NVIDIA's common CUDA software foundation as a way to support evolving models across hardware generations.

OCI also plans to introduce NVIDIA HGX Rubin NVL8, which delivers eight Rubin GPUs connected through NVLink in an x86-based server. The planned configuration includes two 64-core Intel Xeon 6 CPUs, 4 TB of DDR5 memory and 6.4 Tbps of RDMA networking. It is aimed at training, fine-tuning and inference, and lets customers adopt Rubin GPUs in a smaller eight-GPU setup while keeping compatibility with x86-based applications.

Vera CPUs target the work agents do between model calls

A central argument of the announcement is that AI agents rely on more than GPUs. According to the companies, agents call tools, execute code and retrieve information, and slow execution of those CPU-heavy steps can leave GPUs waiting between model calls.

OCI Compute with NVIDIA Vera CPUs combines high single-thread performance with memory bandwidth. The companies say single-thread speed accelerates individual tool calls and code execution, while memory bandwidth keeps cores supplied with data when agents run concurrent tasks. The stated result is less GPU idle time and more consistent performance as workloads grow.

Two supporting features were named. OCI Kubernetes Engine lets customers scale CPU and GPU resources independently, and Oracle Acceleron provides hardware-based isolation for workloads running untrusted, AI-generated code.

OpenAI as reference customer

The release cites OpenAI as an example of the infrastructure at work. In Abilene, OpenAI has trained and served GPT-6 Astra on an OCI Supercluster with more than 100,000 NVIDIA Blackwell GPUs, according to the announcement.

Sreekanth Pothanis, engineering leader, compute infra at OpenAI, said: "NVIDIA Vera Rubin on OCI gives us the computing performance and scale to accelerate model training and serve increasingly demanding AI workloads."

OpenAI is also among the customers with early access to Vera CPUs on OCI. Uday Ruddarraju, chief technology officer of compute at OpenAI, said: "We are bringing Vera into production on OCI to support these workloads at scale," referring to the company's growing agentic sandbox workloads.

What it means for customers

For OCI customers, the release positions the combination of Vera Rubin GPUs and Vera CPUs as a way to run both model training and the surrounding agent workloads in one cloud. Because CPU and GPU capacity can be scaled separately, the companies say organizations can reduce idle capacity and use more of what they pay for.

Mahesh Thiagarajan, president of Oracle Cloud Infrastructure, said the aim is to help customers "keep GPUs productive, secure, and scale efficiently as their needs grow." Ian Buck, vice president of hyperscale and high-performance computing at NVIDIA, said AI factories must be "productive, durable and fungible."

The multi-gigawatt expansion and the HGX Rubin NVL8 systems are described as plans, and the release carries a disclaimer that timing and features may change. Pricing and regional availability were not given in the announcement.