Nvidia and Oracle bring Vera Rubin and Vera CPUs to OCI

Nvidia and Oracle bring Vera Rubin and Vera CPUs to OCI

27 October 2026: Nvidia and Oracle say Nvidia Vera Rubin NVL72 systems and Nvidia Vera CPUs are now generally available on Oracle Cloud Infrastructure, so customers can rent them today. The companies also announced plans to expand OCI Superclusters built on Vera Rubin NVL72 to multi-gigawatt scale. The release gives no figure for that expansion and no timetable.

Not yet independently verified. This rests on Nvidia's release; we have not read Oracle's own announcement, and the statement about the model trained on OCI is Nvidia's, not confirmed by its developer. We will update this when it can be confirmed, and remove this note.

What is available now

Each NVL72 rack connects 72 GPUs through Nvidia's NVLink scale-up networking. OCI Superclusters combine these racks with Oracle's Acceleron networking, Nvidia Spectrum-X Ethernet with Spectrum-6 switches and BlueField-4 data processing units, and gigawatt-class power and cooling. The release says OCI's multiplanar network scales across hundreds of thousands of GPUs, and that the Vera Rubin systems build on the GB200 and GB300 NVL72 systems already in service.

A planned eight-GPU server

A second system is planned but not yet available. The HGX Rubin NVL8 connects eight Rubin GPUs through NVLink in an x86 server with two 64-core Intel Xeon 6 processors, 4 TB of DDR5 memory and 6.4 Tbps of RDMA networking. It is aimed at training, fine-tuning and inference. Oracle's product plans are described in the release as forward-looking and subject to change.

Vera CPUs for agent work

Vera CPUs are pitched at the CPU-heavy parts of agent workloads: tool calls, code execution and data retrieval between model calls. Nvidia says they combine high single-thread performance with high memory bandwidth. OCI Kubernetes Engine lets customers scale CPUs and GPUs independently, and Acceleron provides hardware isolation for workloads that run untrusted, AI-generated code.

Uday Ruddarraju, OpenAI's chief technology officer of compute, says Vera CPUs on OCI perform well under load for agentic sandbox workloads, and that OpenAI is bringing Vera into production on OCI.

The OpenAI claim

The release says that in Abilene, OpenAI trained and served GPT-6 Astra on an OCI Supercluster with more than 100,000 Nvidia Blackwell GPUs. The release does not describe the model.

Sreekanth Pothanis of OpenAI says frontier models need infrastructure that keeps pace, and that Vera Rubin on OCI gives it performance and scale. Mahesh Thiagarajan, president of OCI, and Ian Buck of Nvidia also give statements.

What buyers should check

Availability is not the same as capacity. The release gives no price, no regions and no count of racks in service. Customers planning training runs should ask Oracle which regions hold Vera Rubin capacity, what the contract commitment is, and when the NVL8 option will arrive. Our AI model board lists 136 models.

Sources

• Nvidia: Nvidia and Oracle expand Nvidia Vera Rubin to multi-gigawatt scale and bring Nvidia Vera CPUs to Oracle Cloud Infrastructure, 27 October 2026