NVIDIA and Oracle Plan Multi-Gigawatt Vera Rubin Expansion on OCI

NVIDIA and Oracle intend to grow OCI Superclusters by deploying additional NVIDIA Vera Rubin NVL72 systems at multi-gigawatt capacity to address rising demand for AI training, inference and agentic applications.

The companies have made NVIDIA Vera Rubin NVL72 and NVIDIA Vera CPUs generally available on OCI. This broadens their joint effort to supply infrastructure that handles both GPU-heavy model operations and CPU-heavy agent tasks such as tool execution, code testing and data retrieval. Customers may adjust CPU and GPU allocations separately to limit idle capacity.

OCI Superclusters target frontier AI workloads at scale. The architecture pairs NVL72 rack-scale systems with Oracle Acceleron networking, NVIDIA Spectrum-X Ethernet using Spectrum-6 switches and BlueField-4 DPUs. NVLink connects 72 GPUs inside each rack while the network extends across hundreds of thousands of GPUs. The Vera Rubin rollout extends prior GB200 and GB300 NVL72 installations. OCI also plans to add HGX Rubin NVL8 configurations that link eight Rubin GPUs via NVLink inside an x86 server containing two 64-core Intel Xeon 6 CPUs, 4TB of DDR5 memory and 6.4 Tbps RDMA networking.

OpenAI has already used comparable OCI infrastructure. The company trained and served GPT-6 Astra on an OCI Supercluster containing more than 100,000 NVIDIA Blackwell GPUs. Sreekanth Pothanis, engineering leader for compute infra at OpenAI, stated: “As our frontier models advance, we need infrastructure that can keep pace.” He noted that NVIDIA Vera Rubin on OCI supplies the performance and scale required for advancing frontier models.

NVIDIA Vera CPUs address agent workload demands. The processors provide high single-thread performance and memory bandwidth to accelerate tool calls and code execution between model inferences. OCI Kubernetes Engine permits independent scaling of CPU and GPU resources while Oracle Acceleron supplies hardware isolation for untrusted AI-generated code.