On October 27, 2026, at Oracle AI World in Las Vegas, NVIDIA and Oracle announced the expansion of OCI Superclusters with NVIDIA Vera Rubin NVL72 systems to multi-gigawatt scale. The companies also made the Vera Rubin NVL72 racks and standalone Vera CPUs generally available on Oracle Cloud Infrastructure.
Oracle NVIDIA Vera Rubin Systems Now Available on OCI
Customers can immediately deploy NVIDIA Vera Rubin NVL72 rack-scale systems that combine 72 GPUs with NVLink scale-up networking. Standalone NVIDIA Vera CPUs are also live on OCI, giving users separate scaling of CPU and GPU resources through the OCI Kubernetes Engine.
The rollout builds on existing GB200 and GB300 NVL72 deployments. Oracle plans to introduce HGX Rubin NVL8 systems later, each pairing eight Rubin GPUs with dual 64-core Intel Xeon 6 processors, 4 TB of DDR5 memory, and 6.4 Tbps RDMA connectivity.
Vera Rubin delivers up to 4 times the training performance and up to 30 times higher inference throughput per watt compared with the prior Blackwell generation, according to vendor specifications. The systems use HBM4 memory and NVLink 6 interconnects. Vera CPUs target agentic workloads with 88 Olympus cores and up to 1.2 TB/s memory bandwidth.
Supercluster Architecture Supports Multi-Gigawatt Expansion
OCI Superclusters combine NVL72 racks with Oracle Acceleron networking based on NVIDIA Spectrum-X Ethernet, Spectrum-6 switches, and BlueField-4 DPUs. The multiplanar fabric design supports hundreds of thousands of GPUs within dense campuses that operate inside a roughly 2 km radius with liquid cooling.
Power and cooling infrastructure is scaled for gigawatt-class loads. Customers can adjust CPU and GPU capacity independently to limit idle resources during mixed training, inference, and agentic tasks such as tool execution and data retrieval.
OpenAI previously used an Abilene, Texas OCI Supercluster with more than 100,000 Blackwell GPUs to train and serve its GPT-6 Astra model. Oracle becomes the first hyperscaler to offer Vera CPUs, with an initial deployment of approximately 150,000 units planned.
Context and Next Steps for AI Infrastructure
The announcement extends a long-standing NVIDIA-Oracle collaboration on cloud AI infrastructure. Oracle is also adding offerings from other chip vendors, including Google and Cerebras, to broaden customer choice.
Further Supercluster scaling to multi-gigawatt levels will proceed on a demand-driven basis. Additional software features, including continued CUDA compatibility, Dynamo inference services, and cuVS vector search, are scheduled for integration. No specific contract values or exact rollout timelines were disclosed.
Enterprises evaluating large-scale AI deployments can use the new OCI capacity to test separate CPU and GPU scaling and reduce underutilized hardware during agentic workflows.
Frequently asked questions
When did NVIDIA Vera Rubin become available on OCI?
General availability of Vera Rubin NVL72 systems and Vera CPUs on Oracle Cloud Infrastructure began on October 27, 2026.
What performance gains does Vera Rubin offer over Blackwell?
Vendor data shows up to 4 times training performance and up to 30 times higher inference throughput per watt, supported by HBM4 memory and NVLink 6.
How large can OCI Superclusters grow with Vera Rubin?
The architecture supports multi-gigawatt campuses that can scale to hundreds of thousands of GPUs within a compact physical footprint.
Which workloads benefit most from separate CPU and GPU scaling?
Agentic tasks that combine model inference with tool execution, code testing, and data retrieval gain from independent resource allocation.
By YasTech Editorial
Comments 1