Nvidia and Oracle Plan Multi-Gigawatt Vera Rubin Cloud Expansion
On 27 October 2026, Nvidia and Oracle announced plans to expand Oracle Cloud Infrastructure (OCI) Superclusters with Nvidia Vera Rubin NVL72 systems to multi-gigawatt scale. Vera Rubin NVL72 systems and Nvidia Vera CPUs are already generally available on OCI, and Rubin HGX NVL8 systems are planned. The companies say CPU and GPU capacity can be scaled separately, which matters for AI agents that run tools between model calls. The expansion is a plan, not a schedule: no delivery dates were given for each site.
In short
- Nvidia and Oracle announced on 27 October 2026 plans to take OCI Superclusters with Vera Rubin NVL72 to multi-gigawatt scale.
- Vera Rubin NVL72 systems and standalone Vera CPUs are already generally available on OCI.
- Rubin HGX NVL8 systems are planned, and no fixed delivery dates were given for each site.
- CPU and GPU capacity can be scaled separately, which suits agent workloads that run tools between model calls.
- For most businesses the effect arrives through existing cloud and AI tools, so check your vendor's infrastructure and cost per finished task.
What happened
Nvidia and Oracle announced on 27 October 2026, at Oracle AI World in Las Vegas, plans to expand OCI Superclusters with Nvidia Vera Rubin NVL72 systems to multi-gigawatt scale. According to the joint press release, the same announcement made Vera Rubin NVL72 systems and Nvidia Vera CPUs generally available on OCI. Oracle's own cloud blog adds that Rubin NVL8 GPUs are being introduced on OCI, and Datacenter Dynamics reports that the eight-GPU Rubin HGX NVL8 system will be added at a later date.
The multi-gigawatt part is a plan. The companies did not publish fixed delivery dates for each site, so readers should treat the expansion as a direction rather than a delivery calendar.
Current status by component
| Component | Status in the October 2026 announcement |
|---|---|
| Vera Rubin NVL72 rack systems | Generally available on OCI |
| Standalone Nvidia Vera CPUs | Generally available on OCI |
| OCI Superclusters at multi-gigawatt scale | Planned |
| Rubin HGX NVL8 systems | Planned, with no date given in the release |
Why it matters for businesses that automate work
Automated business work rarely stops at one model call. A finance agent might read an invoice, check a ledger, call a payment tool, then write a note back to the system. The release says the CPU-heavy steps between model calls, such as executing tools, testing code and retrieving data, sit alongside the GPU work of training and serving models. Oracle and Nvidia say customers can scale the two independently, which they argue reduces idle capacity.
Those are the companies' own claims. Oracle's blog attributes to Nvidia's design figures that Vera Rubin delivers up to 30 times the inference throughput per watt of the prior Blackwell generation, and a 4x leap in training performance. These are vendor figures from 2026, not independent measurements. Tom's Hardware ran its own benchmarks on a Vera server, but it states that those tests were not run on OCI infrastructure.
What changes in practice
For most businesses, nothing changes on the day of the announcement. A company that runs reconciliations or payment follow-ups does not buy NVL72 racks. The effect reaches it indirectly, through the cloud platforms and AI tools it already uses, if those providers run their services on this hardware. The announcement does not describe customer pricing, so no cost change can be read from it.
For operations heads who plan to use AI agents, the announcement points to three checks:
- Ask your AI or cloud vendor which infrastructure its agent workflows run on, and whether CPU and GPU capacity are priced separately.
- Measure cost per finished task, such as one matched reconciliation, rather than cost per model call. Tool steps can dominate the time a task takes.
- Keep approval rules in place for payments, refunds and tax filings. Faster infrastructure does not change who is accountable for the output.
The limit is clear. More capacity does not repair a broken process. If invoices arrive in inconsistent formats, or the accounting system holds duplicate customer records, an agent will move that error faster.
What to watch next
- Opening dates for each multi-gigawatt Supercluster site, which the release does not give.
- The launch date of Rubin HGX NVL8 systems on OCI.
- Whether other clouds follow. Tom's Hardware reports that Google Cloud announced general availability for Vera Rubin the week before.
- Any independent utilisation data for OCI, since the idle-capacity benefit is currently the companies' claim.
Sources
How AiStaffo would automate this
AiStaffo would start with one routine workflow, such as matching payments to invoices, and map each tool call the agent makes to your accounting, billing and CRM systems. The AI worker would run those steps automatically, pull the records, draft the match or follow-up, and log each action. The owner or finance lead would still approve payments, refunds and any exception the agent flags. Before any change, the free automation audit measures how many tool calls each workflow uses, so cost per finished task is known rather than guessed. Book a free automation audit.
Questions people ask
What did Nvidia and Oracle announce?
Is Vera Rubin available on Oracle Cloud now?
Why does CPU scaling matter for AI agents?
Does this change the cost of AI automation for my business?
What is Rubin HGX NVL8?
Book a free automation audit
Thirty minutes. We look at one process you run every week and tell you exactly what an AI worker would take off your desk, and what it would not.
