NVIDIA Vera Rubin to Multi-Gigawatt Scale on Oracle OCI

Oracle and NVIDIA announced at Oracle AI World that Oracle Cloud Infrastructure will expand its superclusters to multi-gigawatt scale built on NVIDIA Vera Rubin NVL72 rack systems, with Vera Rubin NVL72 and the new Vera CPU now generally available on OCI. The plan targets surging demand for AI training, inference and agentic workloads, and gives buyers a fresh read on where next-generation datacenter hardware spend is heading.

Each NVL72 rack connects 72 Rubin GPUs over NVLink, while OCI's multiplanar fabric scales across hundreds of thousands of GPUs. Oracle is pairing the racks with its in-house Acceleron networking alongside NVIDIA Spectrum-X Ethernet, Spectrum-6 switches and BlueField-4 DPUs, a hybrid signal that networking silicon remains one of the most contested lines in the component bill of materials.

A notable shift for hardware planners: Oracle plans to introduce HGX Rubin NVL8, an eight-GPU x86 server with two 64-core Intel Xeon 6 CPUs, 4TB of DDR5 and 6.4 Tbps of RDMA networking. Smaller configurations lower the entry point for enterprises that cannot absorb full NVL72 racks, broadening demand beyond hyperscalers.

OpenAI said it has trained and served GPT-6 Astra on an OCI supercluster with more than 100,000 Blackwell GPUs in Abilene, and is bringing Vera CPUs into production for agentic sandbox workloads. NVIDIA CEO Jensen Huang has previously estimated Vera Rubin systems can generate up to $40 billion in revenue per gigawatt, a figure that underscores how much board, power and cooling hardware rides on each deployment.

For component buyers, the multi-gigawatt commitment translates into sustained demand for high-current power stages, busbars, liquid-cooling fittings and 800G-class optics through 2027. sourcing teams watching AI capex announcements should track the OCI buildout as a leading indicator for PCB and module orders upstream.