Oracle and Nvidia are widening their partnership on AI computing as customer demand for capacity grows. Nvidia's Vera Rubin NVL72 racks and its Vera processors are now generally available on Oracle Cloud Infrastructure, and the two firms plan to grow OCI's Superclusters to multi-gigawatt scale, the companies said in a release. The pairing divides the work: graphics processors train and serve large models, while the CPUs handle the chores agents run between model calls, such as executing code and fetching data. Customers can add either kind of capacity independently, which Oracle presents as a way to keep costly GPUs from sitting idle.
Vera marks a shift for Nvidia, which built its data center business on accelerators rather than general-purpose chips. Oracle bills itself as the first large cloud provider to sell Vera CPUs commercially, its engineering blog said, while Nvidia on its second-quarter call said that it continues to see demand for about $20 billion in total server CPUs. Shipments of Vera to lead partners, OCI among them, were already under way by then.
Rack-scale systems are only part of the plan. Oracle also intends to offer an eight-GPU HGX Rubin NVL8 server built on Intel Xeon 6 processors for customers who want to stay on x86 software, the release said. The smaller configuration gives buyers a route onto the new architecture without committing to a full rack-scale system. As a gauge of the scale OCI already runs, the companies pointed to OpenAI's cluster of more than 100,000 Blackwell GPUs in Abilene, Texas, where GPT-6 Astra was trained.
The two companies have worked together across several chip generations, and the Rubin cycle is now Nvidia's main growth engine. Data center revenue reached $89.0 billion in the quarter ended July 26, 2026, Nvidia reported in August, and executives on the earnings call said they expect Vera Rubin to mark the fastest product ramp in the company's history. Racks were already running at OCI, CoreWeave, Google Cloud, Microsoft Azure and Nebius, as per quarterly highlights.