Nvidia and Oracle said on October 27, 2026 that Nvidia Vera Rubin NVL72 rack systems and Nvidia Vera CPUs are now generally available on Oracle Cloud Infrastructure (OCI), and that the two companies plan to grow OCI Superclusters built on Vera Rubin to multi-gigawatt scale. Oracle also plans to add HGX Rubin NVL8 servers, a smaller eight-GPU option, for customers that want Rubin GPUs in an x86 configuration.
What was announced
The announcement has three parts. First, general availability: Vera Rubin NVL72 and the Vera CPU are open to OCI customers now. Second, a planned expansion of OCI Superclusters with Vera Rubin NVL72 to multi-gigawatt scale. Third, a planned HGX Rubin NVL8 platform that connects eight Rubin GPUs through NVLink.
Nvidia says each NVL72 rack links 72 GPUs with NVLink scale-up networking, while Oracle’s multiplanar network scales across hundreds of thousands of GPUs. A supercluster combines those racks with Oracle Acceleron networking, Nvidia Spectrum-X Ethernet with Spectrum-6 switches, BlueField-4 DPUs, and what the release calls gigawatt-class power and cooling. The expansion builds on OCI’s existing GB200 and GB300 NVL72 systems, and Nvidia stresses that its shared CUDA software foundation lets customers carry workloads across hardware generations.
Why CPUs are part of the story
Much of the release is about the processor work that surrounds model inference. AI agents call tools, execute code and retrieve data between model calls, and Nvidia argues that slow CPU steps leave GPUs idle. Vera CPUs on OCI are pitched as combining high single-thread performance with memory bandwidth to shorten those steps. OCI Kubernetes Engine lets customers scale CPU and GPU capacity independently, and Oracle Acceleron provides hardware-based isolation for workloads that run untrusted, AI-generated code. OpenAI is among customers with early access to Vera CPUs on OCI.
This echoes the pitch Nvidia made at GTC Berlin last week, which we covered in our report on its agent toolkit and confidential computing. The commercial point is that as agentic workloads grow, a share of spending shifts from GPUs alone to GPU-plus-CPU systems that one vendor can supply.
The HGX Rubin NVL8 option
The planned HGX Rubin NVL8 system uses two 64-core Intel Xeon 6 CPUs, 4TB of DDR5 memory and 6.4 Tbps of RDMA networking, according to the release. Nvidia positions it for training, fine-tuning and inference where an x86 software stack matters. The release gives no delivery date or pricing for NVL8, and it does not state how many gigawatts Oracle intends to build or by when.
OpenAI as the reference customer
Nvidia points to OpenAI’s work in Abilene, Texas, where it says OpenAI trained and served GPT-6 Astra on an OCI Supercluster of more than 100,000 Nvidia Blackwell GPUs. An OpenAI compute infrastructure leader is quoted saying Vera Rubin on OCI gives the company the performance and scale to speed up training and serve more demanding workloads. As with any vendor release, these are statements of intent from the companies involved rather than independently verified results.
Why it matters
The deal shows how tightly frontier model development is tied to a few infrastructure providers. Power is the constraint to watch: a separate Nvidia post this month says more than $300 billion of power equipment is backlogged and over 200 gigawatts of projects sit in U.S. interconnection queues, which is why “multi-gigawatt” plans are as much about electricity as about chips. For background on why customers stay on Nvidia’s platform, see our explainer on the CUDA moat.
What to watch next
- Which regions get Vera Rubin capacity first and how much of it is available to customers beyond OpenAI.
- A timetable and pricing for HGX Rubin NVL8 on OCI.
- How Oracle sources power and financing for gigawatt-class sites.
- Whether other clouds announce comparable Vera CPU offerings.
- Independent benchmarks for agentic workloads on Vera CPUs.
Key takeaways
- Vera Rubin NVL72 and Vera CPUs are generally available on Oracle Cloud as of October 27, 2026.
- Nvidia and Oracle plan multi-gigawatt OCI Superclusters, but gave no schedule or total.
- Vera CPUs target the tool-calling and code-execution work agents do between model calls.
- HGX Rubin NVL8 is a planned eight-GPU x86 option with two 64-core Xeon 6 CPUs.
- The performance claims come from the companies and have not been independently tested.
Frequently asked questions
What is Vera Rubin?
Vera Rubin is Nvidia’s next-generation platform pairing Rubin GPUs with Vera CPUs. The NVL72 version is a rack system that links 72 GPUs through NVLink.
Is Vera Rubin available on Oracle Cloud now?
Yes. Nvidia and Oracle say Vera Rubin NVL72 and Vera CPUs are generally available on OCI, while the HGX Rubin NVL8 systems and the multi-gigawatt expansion are planned.
Why does Nvidia sell CPUs for AI agents?
Agents spend time running tools, code and data retrieval on CPUs between model calls. Nvidia argues that faster CPUs keep GPUs busy and reduce idle capacity.
