Nvidia, Oracle say Vera Rubin racks are now on Oracle Cloud

Nvidia’s Vera Rubin NVL72 rack-scale systems are now generally available on Oracle Cloud Infrastructure, and the two companies plan to expand Oracle’s AI Superclusters with the systems to multi-gigawatt scale, Nvidia and Oracle said in a joint release Tuesday. Nvidia’s Vera CPUs are also generally available on OCI, the companies said.

What’s driving it

The companies said the expansion supports growing customer demand for AI training, inference and agentic applications. According to the release, OCI Superclusters combine Nvidia NVL72 rack-scale systems with Oracle Acceleron networking, Nvidia Spectrum-X Ethernet networking featuring Spectrum-6 switches and BlueField-4 DPUs, and gigawatt-class power and cooling. Each rack connects 72 GPUs through Nvidia’s NVLink, the release said, and the planned Vera Rubin build-out extends OCI’s existing GB200 and GB300 NVL72 systems.

The release pointed to OpenAI as an example: In Abilene, OpenAI has trained and served GPT-6 Astra on an OCI Supercluster with more than 100,000 Nvidia Blackwell GPUs, the companies said.

“As our frontier models advance, we need infrastructure that can keep pace,” Sreekanth Pothanis, engineering leader for compute infrastructure at OpenAI, said in the release.

The companies also pitched Vera CPUs for the work AI agents do between model calls, such as running tools, executing code and retrieving data. “We are bringing Vera into production on OCI to support these workloads at scale,” said Uday Ruddarraju, OpenAI’s chief technology officer of compute.

Why it matters

The announcement shows how Nvidia and Oracle are framing the next phase of AI computing. By the companies’ account, demand is no longer only about the GPUs that train and serve large models but also about the general-purpose processors that handle agents’ tool calls and code execution. The release said slow CPU work can “leave GPUs waiting between model calls,” and it presented Vera CPUs, along with the ability to scale CPU and GPU resources independently, as a way to reduce idle capacity.

The pitch also stresses reuse. Ian Buck, Nvidia’s vice president of hyperscale and high-performance computing, said in the release that AI factories must be “productive, durable and fungible,” remaining useful as software advances. That framing addresses a question buyers face with each new chip generation: how long expensive hardware stays valuable.

The scale claims carry caveats. The multi-gigawatt plan is an expansion the companies describe as planned, with no cost or schedule disclosed, and Oracle’s product disclaimer in the release says development, timing and pricing may change. The OpenAI figures describe Blackwell-based capacity already in use, not Vera Rubin.

Who has a stake

Nvidia supplies the GPUs, CPUs and networking. Oracle sells the capacity through OCI; Mahesh Thiagarajan, president of Oracle Cloud Infrastructure, said in the release the collaboration extends OCI’s capabilities “across training, inference and agentic AI.” OpenAI is named as an early-access Vera CPU customer and the user of the Abilene Supercluster. Intel’s Xeon 6 processors are part of a planned server configuration.

What’s next

OCI plans to introduce Nvidia HGX Rubin NVL8, which the release said pairs eight Rubin GPUs with two 64-core Intel Xeon 6 CPUs, 4TB of DDR5 memory and 6.4 Tbps of RDMA networking in an x86-based server.