Nvidia's Vera Rubin chips officially arrived on Oracle Cloud on 27 October 2026. At Oracle AI World in Las Vegas, Nvidia and Oracle announced that Vera Rubin NVL72 systems and Vera CPUs are now generally available on Oracle Cloud Infrastructure (OCI), and that the two companies plan to expand Superclusters built on the platform to multi-gigawatt scale.
What is live, and what is still planned
Three layers of this announcement need to be kept apart: what can already be rented, what is being expanded, and what is still only planned.
| System | Status | Key point |
|---|---|---|
| Vera Rubin NVL72 | Generally available | 72 GPUs per rack, linked by NVLink |
| Vera CPU | Generally available | CPU and GPU resources scale independently |
| HGX Rubin NVL8 | Planned | Eight Rubin GPUs in an x86 server form factor |
The Superclusters combine NVL72 racks with Oracle Acceleron networking, Nvidia Spectrum-X Ethernet, Spectrum-6 switches and BlueField-4 data processing units, with networking that scales to hundreds of thousands of GPUs. The systems build on OCI's existing GB200 and GB300 NVL72 infrastructure, and Nvidia says a common CUDA software foundation lets customers carry existing work across hardware generations. No contract value or timetable was disclosed for the expansion, and no availability date was given for the HGX Rubin NVL8.
Agent workloads are the selling point
Oracle and Nvidia are pitching the platform on agents: between model calls, agents run tools, test code and retrieve data — work that consumes CPUs rather than GPUs. With CPU and GPU resources able to scale independently, GPUs need not sit idle while waiting. In its press release, Nvidia says OpenAI is bringing Vera CPUs into production on OCI for agent sandbox workloads, and has trained and served GPT-6 Astra on an OCI Supercluster containing more than 100,000 Blackwell GPUs. That claim comes from Nvidia's release and has not been separately confirmed by OpenAI.
What to watch next
For Oracle, this is part of the compute diversification announced at the same event: Google TPUs are also coming to OCI, widening the choice of inference chips. For Nvidia, general availability of Rubin at a major cloud provider is the key step from launch to delivery for the new platform. But the press release says nothing about where funding for the multi-gigawatt expansion will come from, or how long customer prepayments and external financing can sustain it. Whether Rubin delivers will be judged by the workloads actually running on OCI over the coming quarters, not by the scale figures quoted at the launch event.
