Austin, 27 October. NVIDIA and Oracle have agreed to expand Oracle Cloud Infrastructure (OCI) Superclusters built on NVIDIA Vera Rubin NVL72 systems to multi-gigawatt scale, and Vera Rubin NVL72 systems and NVIDIA Vera central processors have become generally available to the cloud’s customers, the companies said.
Each NVL72 rack links 72 graphics processors through NVLink, and Oracle’s network, according to the companies, connects hundreds of thousands of such processors. The new capacity will add to the NVIDIA GB200 and GB300 NVL72 systems already running on OCI. Oracle also plans HGX Rubin NVL8 servers — eight Rubin graphics processors with two 64-core Intel Xeon 6 processors, 4 TB of DDR5 memory and 6.4 Tbps networking — for customers who need a smaller configuration compatible with the familiar x86 architecture. The companies named neither timelines nor the value of the deal.
Particular emphasis is placed on central processors. Between calls to a model, AI agents call tools, run code and retrieve data; this is work for central rather than graphics processors, and if it is slow, expensive graphics processors sit idle. “We are bringing Vera into production on OCI to support these workloads at scale,” said Uday Ruddarraju, OpenAI’s chief technology officer of compute. According to NVIDIA, OpenAI trained and serves its GPT-6 Astra model in Abilene, Texas, on an OCI Supercluster with more than 100,000 NVIDIA Blackwell graphics processors. Customers can scale central and graphics processing capacity independently, so that less equipment sits idle.
What it means: The race for AI computing power is now measured in gigawatts of electricity rather than numbers of servers; Oracle and NVIDIA are betting that AI agents will need not only graphics but also fast central processors.
Who benefits: NVIDIA gains sales of a new chip generation. Oracle gains a place among the largest cloud providers for AI. OpenAI and other model developers gain capacity for training and running agents. Power companies gain new demand: multi-gigawatt clusters consume electricity on the scale of large power stations.
Background: Vera Rubin is the generation of NVIDIA’s computing platform that follows Blackwell; it is named after the American astronomer Vera Rubin. NVIDIA chief executive Jensen Huang visited Astana for the first time on 23 October and proposed making Kazakhstan a physical AI hub. NVIDIA, together with Firebird AI, is a technology partner of Kazakhstan’s Data Center Valley project in Ekibastuz: the 215 MW of its first phase are already fully reserved, the first 50 MW data centre is due to open in the second quarter of 2027, and the cluster is eventually to grow to 1 GW.
We have also reported that NVIDIA was proposed to open a regional physical AI centre in Kazakhstan.
