Details
- NVIDIA and Oracle announced the general availability of NVIDIA Vera Rubin NVL72 systems and Vera CPUs on Oracle Cloud Infrastructure (OCI), alongside plans to expand OCI Superclusters to multi-gigawatt scale.
- The collaboration targets frontier-model training, inference and agentic workloads, combining NVIDIA GPUs, Vera CPUs, Oracle Acceleron networking, Spectrum-X Ethernet and BlueField-4 DPUs.
- NVL72 racks connect 72 GPUs through NVLink, while OCI Kubernetes Engine allows customers to scale CPU and GPU resources independently for model calls, tool execution, code testing and data retrieval.
- Oracle plans to introduce HGX Rubin NVL8, an eight-GPU, x86-based system with two 64-core Intel Xeon 6 CPUs, 4TB of DDR5 memory and 6.4 Tbps of RDMA networking for training, fine-tuning and inference.
- OpenAI said it is deploying Vera CPUs for agentic sandbox workloads; the announcement also cites OpenAI training and serving GPT-6 Astra on an OCI Supercluster with more than 100,000 NVIDIA Blackwell GPUs.[1][3][8]
Impact
The announcement reflects a shift toward heterogeneous AI infrastructure, where CPU capacity is provisioned alongside GPUs to reduce idle accelerator time as agents execute tools and code. Over the next 12–24 months, independently scalable CPU-GPU architectures and smaller Rubin configurations could broaden cloud access to advanced workloads while encouraging operators to prioritize utilization, networking and workload portability alongside raw accelerator scale.