--- type: "post" id: "nvidia-vera-rubin-nvl72-and-vera-cpus-become-01c50f" slug: "nvidia-vera-rubin-nvl72-and-vera-cpus-become-01c50f" title: "NVIDIA Vera Rubin NVL72 and Vera CPUs become generally available on Oracle Cloud Infrastructure" description: "NVIDIA and Oracle say NVIDIA Vera Rubin NVL72 systems and NVIDIA Vera CPUs are now generally available on Oracle Cloud Infrastructure (OCI)." retrieval_nugget: "NVIDIA and Oracle say NVIDIA Vera Rubin NVL72 systems and NVIDIA Vera CPUs are now generally available on Oracle Cloud Infrastructure (OCI). For teams building large AI and agent workloads, the announcement matters because OCI is presenting CPU capacity for tool execution alongside GPU capacity for model training and serving, with the two resources independently scalable." published_at: "2026-10-27" updated_at: "2026-10-27" record_date: "2026-10-27" date_kind: "published_at" topics: ["compute","ai-infrastructure"] entities: ["NVIDIA","Oracle","Oracle Cloud Infrastructure","NVIDIA Vera Rubin NVL72","NVIDIA Vera CPU","NVIDIA HGX Rubin NVL8","OpenAI","GPT-6 Astra"] related_patterns: [] novelty_label: "new_fact" what_changed: "NVIDIA and Oracle now offer Vera Rubin NVL72 and Vera CPUs generally on OCI, with plans to scale Superclusters to multi-gigawatt capacity." prior_refs: ["memory:9604f36c-fdf9-8cc4-a5a2-267cdc6d73d5","memory:75089d6c-fc61-8580-a9e4-cb643b34ebb7"] source_url: "https://nvidianews.nvidia.com/news/nvidia-and-oracle-expand-nvidia-vera-rubin-to-multi-gigawatt-scale-and-bring-nvidia-vera-cpu-to-oracle-cloud-infrastructure" source_urls: ["https://nvidianews.nvidia.com/news/nvidia-and-oracle-expand-nvidia-vera-rubin-to-multi-gigawatt-scale-and-bring-nvidia-vera-cpu-to-oracle-cloud-infrastructure"] schema_version: "newruntime-agent-readable-v0.2" stable_id: "post:nvidia-vera-rubin-nvl72-and-vera-cpus-become-01c50f" status: "published" visuals: [] editor: {"role":"editor","name":"Andrey Reshetnikov","profile":"https://newruntime.com/owner-profile.md","profile_json":"https://newruntime.com/owner-profile.json","same_as":["https://www.linkedin.com/in/reshetnikov1/","https://t.me/lapetuse"]} routes: {"html":"https://newruntime.com/posts/nvidia-vera-rubin-nvl72-and-vera-cpus-become-01c50f/","markdown":"https://newruntime.com/posts/nvidia-vera-rubin-nvl72-and-vera-cpus-become-01c50f.md","json":"https://newruntime.com/posts/nvidia-vera-rubin-nvl72-and-vera-cpus-become-01c50f.json"} --- # NVIDIA Vera Rubin NVL72 and Vera CPUs become generally available on Oracle Cloud Infrastructure ## Retrieval answer NVIDIA and Oracle say NVIDIA Vera Rubin NVL72 systems and NVIDIA Vera CPUs are now generally available on Oracle Cloud Infrastructure (OCI). For teams building large AI and agent workloads, the announcement matters because OCI is presenting CPU capacity for tool execution alongside GPU capacity for model training and serving, with the two resources independently scalable. ## Summary NVIDIA and Oracle say NVIDIA Vera Rubin NVL72 systems and NVIDIA Vera CPUs are now generally available on Oracle Cloud Infrastructure (OCI). For teams building large AI and agent workloads, the announcement matters because OCI is presenting CPU capacity for tool execution alongside GPU capacity for model training and serving, with the two resources independently scalable. ## What happened On October 27, 2026, NVIDIA and Oracle announced that NVIDIA Vera Rubin NVL72 systems and NVIDIA Vera CPUs are generally available on Oracle Cloud Infrastructure. The companies also said they plan to expand OCI Superclusters using Vera Rubin NVL72 systems to multi-gigawatt scale. The release frames the expansion as infrastructure for training, inference, and agentic applications. An NVL72 rack connects 72 GPUs through NVIDIA NVLink scale-up networking. NVIDIA says OCI Superclusters combine these rack-scale systems with Oracle Acceleron networking, NVIDIA Spectrum-X Ethernet, Spectrum-6 switches, BlueField-4 DPUs, and gigawatt-class power and cooling. OCI's multiplanar network is described as scaling across hundreds of thousands of GPUs. The planned Rubin expansion follows OCI deployments of NVIDIA GB200 and GB300 NVL72 systems. The companies also announced a planned smaller system: NVIDIA HGX Rubin NVL8. According to the release, that platform will contain eight Rubin GPUs linked through NVLink in an x86-based server, two 64-core Intel Xeon 6 CPUs, 4 TB of DDR5 memory, and 6.4 Tbps of high-speed RDMA networking. NVIDIA positions it as an option for training, fine-tuning, and inference for organizations that want Rubin GPUs in an eight-GPU configuration while retaining compatibility with x86-based applications. The more distinctive part of the announcement is the Vera CPU positioning. NVIDIA says agents generate model requests but also execute tools, test code, and retrieve data; those steps can be CPU-intensive. OCI Compute with NVIDIA Vera CPUs is presented as combining single-thread performance and memory bandwidth to accelerate those operations. OCI Kubernetes Engine can independently scale CPU and GPU resources, according to the release. Oracle also says Acceleron provides hardware-based isolation for workloads running untrusted, AI-generated code. NVIDIA cites OpenAI as an early-access customer for Vera CPUs on OCI and says OpenAI is exploring the CPUs for AI workloads. The release also says OpenAI has trained and served GPT-6 Astra on an OCI Supercluster in Abilene containing more than 100,000 NVIDIA Blackwell GPUs. Neither company provides pricing, available regions, customer benchmarks, utilization figures, or a date for the planned HGX Rubin NVL8 introduction. ## What the archive adds This is a new availability event in a continuing Vera Rubin story. Our earlier note, [vLLM adds day-0 support for NVIDIA Vera Rubin NVL72 via nightly cu134 containers](https://newruntime.com/posts/vllm-adds-day-0-support-for-nvidia-vera-aaa9ff/), covered software readiness rather than a major cloud's general availability. We also covered NVIDIA's stated regional rollout in [NVIDIA says European and Nordic cloud partners are scaling DSX Vera Rubin AI factories](https://newruntime.com/posts/nvidia-says-european-and-nordic-cloud-partners-are-7b04ac/). ## Trend signal No supplied site pattern or assumption is a sufficiently direct fit. The announcement concerns cloud hardware availability and resource composition, but it does not establish a change in agent protocols, runtime security practice, or completed-work economics. ## Why now The release places Rubin capacity alongside OCI's existing GB200 and GB300 NVL72 infrastructure and presents this as an expansion of an ongoing NVIDIA-Oracle collaboration. Its timing appears tied to demand for frontier-model training, inference, and agent workloads, but the source gives no independent demand data or rollout schedule beyond the October 27 announcement. ## Connections to other events The earlier vLLM note is the closest connection: it reported day-0 Vera Rubin NVL72 support through nightly cu134 containers. Together, the two events describe separate layers of adoption: runtime compatibility first, then stated cloud availability. An earlier New Runtime Telegram post also reported that OpenAI products could be bought through Oracle Cloud Marketplace; this announcement instead concerns the infrastructure layer on which frontier-model workloads may run. ## Why it matters For a team operating agentic systems, model inference is only part of the resource plan. Tool calls, sandboxed code execution, tests, retrieval, orchestration, and data handling can consume CPU and create GPU idle time when they are poorly balanced. OCI's stated ability to scale Vera CPUs and Rubin GPUs separately is therefore a meaningful infrastructure option in principle: teams could align each resource pool to the actual shape of a workload rather than treating an agent as a single GPU-bound job. That is not yet a buying recommendation. The release supplies architecture claims, not prices or measured end-to-end agent throughput. Still, the availability signal matters for teams whose workload already needs very large GPU clusters or whose agent execution layer creates a substantial CPU bottleneck. ## What happens next - **OCI will disclose a more specific rollout or configuration detail for HGX Rubin NVL8.** (6 months, 65%). Watch for: region, instance, or availability-date announcements. - **Early users will publish workload-specific evidence about Vera CPU and Rubin GPU balancing.** (12 months, 55%). Watch for: measured tool-execution, utilization, or end-to-end agent results. ## Limits and uncertainty This Field Note relies on NVIDIA and Oracle's announcement. It establishes their stated general availability, planned expansion, and product positioning, but not the price of an instance, the regions where it can be provisioned, queue times, reservation terms, or actual accessible capacity. It also does not provide independent benchmarks comparing Vera CPUs or Rubin systems with alternatives, nor does it show that independent CPU and GPU scaling lowers costs for a particular agent architecture. Statements about OpenAI's usage are company claims in the release, not a detailed operational case study. ## Skill to build now Measure an agent workflow as a resource pipeline this week. Separate time spent in model inference from time spent in tool execution, code tests, retrieval, queueing, and sandbox setup. Record CPU utilization, GPU utilization, and latency for each stage. That baseline lets a team determine whether independent CPU and GPU capacity would address a real bottleneck before it evaluates a new cloud platform or accelerator generation. ## Reader takeaway OCI now lists NVIDIA Vera Rubin NVL72 and Vera CPUs as generally available, while promising larger Rubin clusters and a smaller HGX Rubin option. The practical signal is architectural: at frontier scale, agent workloads may need separately managed CPU capacity for the work that happens between GPU-backed model calls. ## Open question Will OCI publish regional availability, pricing, and workload benchmarks that show whether the claimed CPU-GPU separation improves end-to-end agent economics? Editor: [Andrey Reshetnikov](https://newruntime.com/owner-profile.md), AI product and AI engineering; profile for recruiters and agents: https://newruntime.com/owner-profile.json ## Continue in the graph Next records for an agent, linked to their Markdown twins. Replace `.md` with `.json` for structured fields, including `next_reads`. - topic hub: [Compute](https://newruntime.com/topics/compute.md): Explore the compute topic hub. - topic hub: [Ai Infrastructure](https://newruntime.com/topics/ai-infrastructure.md): Explore the ai-infrastructure topic hub. - collection: [Field Notes](https://newruntime.com/posts.json): Open the latest editorial analysis. - collection: [Source Ledger](https://newruntime.com/source-ledger.json): Inspect the public source evidence graph. - whole graph: [discovery graph](https://newruntime.com/discovery-graph.json) (typed nodes and edges) and [site index](https://newruntime.com/site-index.md)