Vera Rubin arrives on Oracle Cloud
On 27 October 2026 NVIDIA and Oracle announced that NVIDIA Vera Rubin NVL72 rack systems and NVIDIA Vera CPUs are generally available on Oracle Cloud Infrastructure (OCI), and that OCI plans to expand its Vera Rubin superclusters to multi-gigawatt scale and to add HGX Rubin NVL8 servers. The release says OpenAI has trained and served GPT-6 Astra on an OCI supercluster in Abilene with more than 100,000 Blackwell GPUs.
Why it matters
NVIDIA's newest generation, announced in January as in production, can now be rented from a hyperscaler, with a separate line of CPUs for the work agents do between model calls. There are no capacity figures: "multi-gigawatt" is a direction and the NVL8 servers are only a plan.
What the release says. Two statements in one headline: Vera Rubin NVL72 and Vera CPUs "are now generally available" on OCI; and "plans" to expand OCI superclusters with Vera Rubin to multi-gigawatt scale and to introduce HGX Rubin NVL8 (eight Rubin GPUs on NVLink, two 64-core Xeon 6 CPUs, 4 TB of DDR5, 6.4 Tbps RDMA networking). Customers with early access to Vera CPUs, "including OpenAI", are exploring them; OpenAI's chief technology officer of compute says Vera is being brought "into production" on OCI for agent sandboxes. On GPT-6 Astra: "In Abilene, OpenAI has trained and served GPT-6 Astra on an OCI Supercluster with more than 100,000 NVIDIA Blackwell GPUs": NVIDIA's sentence, with no date for the training. Oracle's notes say the description is "not a commitment" to deliver anything. What the record does not claim. The page gives no price, no number of racks, no regions and no customer other than OpenAI. The claim about GPT-6 Astra is NVIDIA's own wording in its release, unchecked. Oracle's own pages were not read. This is a release of two companies.