NVIDIA and Oracle plan multi-gigawatt Vera Rubin expansion on OCI

The companies plan to expand OCI Superclusters with NVIDIA Vera Rubin NVL72 systems to multi-gigawatt scale, citing demand for AI training, inference and agentic applications.[1] Oracle says the expanded infrastructure is intended for GPU-intensive frontier-model training and serving, as well as CPU-intensive agent work such as tool use, code testing and data retrieval.[1] The announcement also says Vera Rubin NVL72 systems and Vera CPUs became generally available on OCI as part of the collaboration.[1]

Oracle says OCI customers can scale CPU and GPU resources independently to fit their workloads and reduce idle capacity.[1] OCI Supercluster racks connect 72 GPUs using NVIDIA NVLink scale-up networking, while OCI’s multiplanar network is designed to scale across hundreds of thousands of GPUs.[1] Those are infrastructure design details, not a stated guarantee of performance for a particular customer workload.[1]

Oracle says Vera Rubin offers up to 30 times higher inference throughput per watt and four times the training performance of the previous Blackwell generation.[2] NVIDIA reports up to 80% faster agentic CPU performance for Vera compared with leading x86 CPUs.[4] Oracle cautions that the 80% figure comes from NVIDIA platform benchmarks, not OCI instance-level testing, and says actual performance varies by workload and configuration.[4] Teams evaluating the figures should treat them as vendor-reported comparisons, not as results established for their own OCI deployments.[4]

Oracle plans to introduce HGX Rubin NVL8, an x86-based server configuration with eight Rubin GPUs connected through NVLink.[1] The planned configuration includes two 64-core Intel Xeon 6 CPUs, 4 TB of DDR5 memory and 6.4 Tbps of RDMA networking.[1] Oracle describes this smaller deployment option as maintaining compatibility with x86-based applications.[1]