Nvidia's Vera Rubin systems are now generally available on Oracle Cloud. The bigger story is what comes next: multi-gigawatt Superclusters built for AI agents.
October 28, 2026 — Nvidia and Oracle announced on October 27, 2026, at Oracle AI World in Las Vegas, that Nvidia Vera Rubin NVL72 systems and standalone Nvidia Vera CPUs are generally available on Oracle Cloud Infrastructure (OCI). The companies also plan to expand OCI Superclusters built on Vera Rubin NVL72 to multi-gigawatt scale.
Image: A large AI server rack with glowing cooling lines in a dark data centre hall. Credit: AI-generated illustration / Red Eye Eagle News.
Editor's Highlight: The Vera CPU is the part of this announcement most readers will overlook, and the part that matters most for the next wave of AI. Agents spend much of their time doing work between model calls — calling tools, running code, fetching information. If the CPU cannot keep up, the expensive GPU waits idle. Oracle and Nvidia are selling the fix for that idle time.
Table of Contents
- What Was Announced
- Inside the Vera Rubin NVL72 Rack
- The Vera CPU and the Agent Problem
- OpenAI, the Named Early Customer
- Oracle's Wider Chip Diversification
- What Is Planned, Not Yet Built
- Bottom Line
- FAQ
- Sources
Key Facts
| Detail | Information |
|---|---|
| Announced | October 27, 2026, at Oracle AI World, Las Vegas |
| Generally available | Vera Rubin NVL72 systems; standalone Vera CPUs on OCI |
| Expansion plan | OCI Superclusters to multi-gigawatt scale |
| Rack | 72 GPUs connected via NVLink |
| Deal value | Not disclosed |
| Pricing | Not published in sources read |
| Planned only | HGX Rubin NVL8; no date given |
| Named customer | OpenAI, for agentic sandbox workloads |
What Was Announced
Nvidia Vera Rubin NVL72 systems and Nvidia Vera CPUs are generally available on OCI, according to the joint announcement carried by GlobeNewswire and republished by Morningstar. The announcement was made on October 27, 2026, at Oracle AI World in Las Vegas.
The companies plan to expand OCI Superclusters built on Vera Rubin NVL72 to multi-gigawatt scale, for AI training, inference and agentic applications. That expansion is a plan, not completed capacity. No deal value was disclosed, according to Reuters reporting.
The expansion builds on OCI's existing Nvidia GB200 and GB300 NVL72 systems, on a common CUDA software foundation, according to the joint announcement. In other words, customers already running on Oracle's earlier Nvidia systems are being offered a newer generation on the same software base.
Tech Analysis: "Generally available" and "multi-gigawatt" describe two different things. The first means customers can use the systems now, in the regions where they are offered. The second is a buildout plan. Conflating them would overstate what exists today.
Inside the Vera Rubin NVL72 Rack
Each NVL72 rack connects 72 GPUs through Nvidia NVLink scale-up networking, according to the joint announcement. OCI's multiplanar network scales across hundreds of thousands of GPUs.
OCI Superclusters combine the NVL72 racks with Oracle Acceleron networking, Nvidia Spectrum-X Ethernet with Spectrum-6 switches, BlueField-4 DPUs, and gigawatt-class power and cooling, according to the announcement.
The video embedded above is a Supermicro product spotlight on the Vera Rubin NVL72 rack, filmed at Nvidia GTC Berlin — a look at the rack hardware itself, not footage of the Oracle announcement.
One further system is planned only. HGX Rubin NVL8, with eight Rubin GPUs in an x86 server with two 64-core Intel Xeon 6 CPUs, 4 TB of DDR5 memory and 6.4 Tbps RDMA networking, is planned for training, fine-tuning and inference. No availability date was given, according to the joint announcement and The Register.
The Vera CPU and the Agent Problem
Vera CPUs target the work AI agents do between model calls: calling tools, executing code and retrieving information. Slow CPU execution can leave GPUs waiting, according to the announcement. Customers can scale CPU and GPU resources independently through OCI Kubernetes Engine, and Acceleron provides hardware-based isolation for workloads running untrusted, AI-generated code.
Oracle's dual-processor Vera configuration, as reported from Oracle's Vera compute announcement, uses two 88-core Arm-compatible Nvidia Olympus processors — 176 cores and 352 threads — with 1 TB of LPDDR5X memory, 7.68 TB of local NVMe storage and 200 Gbps Acceleron networking. The configuration is offered as Bare Metal and Flex VM instances. A per-processor peak memory bandwidth figure of up to 1.2 TB/s is Nvidia's processor specification, not a measured OCI instance result.
Oracle plans an initial deployment of 150,000 Vera CPUs across its regions, according to the same reporting, with instances available in selected regions in the fourth quarter of 2026. Availability is therefore not in every region at once, and no pricing was published in the sources read for this article.
The performance claims need the same care. VentureBeat reported that Oracle's "up to 80% faster" Vera figure for agentic CPU performance, against leading x86 CPUs, comes from Nvidia platform benchmarks, not OCI instance tests. It is a company benchmark claim, not an independent measurement. VentureBeat also cited its own surveys, in which 79% of 218 respondents reported GPU utilisation at 50% or below — context for the efficiency pitch, not proof that Vera systems will save any particular customer money.
Practical Takeaway: If you run AI agents at scale, the question to ask a cloud provider is no longer only "how many GPUs?" It is "how much of the time are those GPUs actually working?" The Vera CPU pitch is aimed squarely at that idle time.
OpenAI, the Named Early Customer
OpenAI is the named early customer in the joint announcement. Uday Ruddarraju, OpenAI's chief technology officer of compute, said: "We are bringing Vera into production on OCI to support these workloads at scale," referring to agentic sandbox workloads.
Customers with early access, including OpenAI, were already exploring Vera CPUs on OCI, according to the announcement. That is a statement made in the companies' own announcement, reported here as such.
Oracle's Wider Chip Diversification
The Vera Rubin expansion was not Oracle's only chip news on October 27. Oracle also said it will begin offering Google TPUs on OCI, with TPU availability from 2027, and plans inference chips from Cerebras and SambaNova, according to Reuters and The Register. Some Nvidia-based systems will use Oracle's in-house Acceleron networking chips instead of some Nvidia networking chips, Reuters reported.
That diversification is background to the Vera Rubin story, not its headline. It does, however, explain Oracle's strategy: offer customers more than one kind of AI hardware on the same cloud, rather than betting on a single supplier.
For context only, Reuters noted that Nvidia chief executive Jensen Huang has previously said Vera Rubin systems can generate up to $40 billion in revenue per gigawatt for Nvidia. That is a prior general statement, not a figure for this Oracle deal, whose value was not disclosed.
What Is Planned, Not Yet Built
Several elements of the announcement are plans or expectations, and should be read that way:
- Multi-gigawatt Supercluster expansion: planned, not completed.
- HGX Rubin NVL8: planned, with no availability date.
- Vera CPU instances in selected regions in Q4 2026: a rollout plan, not availability everywhere.
- Google TPUs on OCI: from 2027.
- No pricing was published for Vera CPU instances or NVL72 capacity.
Reuters also noted that Oracle did not disclose how it will finance its chip purchases, and that Oracle plans an investor day this week. Those are financing and calendar details, not part of the product announcement, and they are noted here only for completeness.
Bottom Line
Oracle and Nvidia have put Vera Rubin NVL72 systems and standalone Vera CPUs on general availability on OCI, with a multi-gigawatt Supercluster expansion planned. The hardware story is real and available now in selected regions; the scale story is a plan. What to watch next is the Q4 Vera CPU rollout, any pricing disclosure, and whether other cloud providers answer with their own CPU-plus-GPU agent infrastructure.
FAQ
What is the Vera Rubin NVL72?
It is Nvidia's rack-scale AI system, with 72 GPUs connected through NVLink networking. The systems are now generally available on Oracle Cloud Infrastructure, according to the October 27, 2026 announcement.
What is a Vera CPU?
A Vera CPU is Nvidia's Arm-compatible processor designed to handle the work AI agents do between model calls — calling tools, executing code and retrieving information — so GPUs spend less time waiting. Standalone Vera CPUs are generally available on OCI, in selected regions.
Is the multi-gigawatt expansion already built?
No. The companies plan to expand OCI Superclusters built on Vera Rubin NVL72 to multi-gigawatt scale. It is a plan, and no completion date or deal value was disclosed.
Who is using Vera CPUs on OCI?
OpenAI is the named early customer. Its compute chief technology officer, Uday Ruddarraju, said OpenAI is bringing Vera into production on OCI for agentic sandbox workloads.
What about the HGX Rubin NVL8?
It is planned only, with no availability date given. It pairs eight Rubin GPUs with two Intel Xeon 6 CPUs for training, fine-tuning and inference.
Sources
- Nvidia / Oracle joint announcement, via GlobeNewswire and Morningstar — "NVIDIA and Oracle Expand NVIDIA Vera Rubin to Multi-Gigawatt Scale and Bring NVIDIA Vera CPU to Oracle Cloud Infrastructure", October 27, 2026: https://www.morningstar.com/news/globe-newswire/9853667/nvidia-and-oracle-expand-nvidia-vera-rubin-to-multi-gigawatt-scale-and-bring-nvidia-vera-cpu-to-oracle-cloud-infrastructure
- Reuters, via SRN News — "Oracle expands chip supply deal with Nvidia, adds Google and Cerebras offerings", October 27, 2026: https://srnnews.com/oracle-expands-chip-supply-deal-with-nvidia-adds-google-and-cerebras-offerings/
- VentureBeat — "Nvidia and Oracle expand Vera Rubin cloud infrastructure, pitching better utilization", October 27, 2026: https://venturebeat.com/infrastructure/nvidia-and-oracle-expand-vera-rubin-cloud-infrastructure-pitching-better-utilization-as-enterprises-report-idle-gpus
- The Register — "Big Red gets cozy with Google to offer TPUs to AI customers", October 27, 2026: https://www.theregister.com/ai-and-ml/2026/10/27/big-red-gets-cozy-with-google-to-offer-tpus-to-ai-customers/5305846
- AiCybr — "Oracle / Nvidia Vera CPU / Rubin NVL72 OCI specs and agent workloads": https://aicybr.com/blog/oracle-nvidia-vera-cpu-rubin-nvl72-oci-specs-agent-workloads
- Supermicro video — "NVIDIA Vera Rubin NVL72 Spotlight (Now Shipping)": https://www.youtube.com/watch?v=qTZN74vyevU
- Hero image source: original AI-generated illustration created for this article.
Join the conversation
Comments
Post a Comment