Nvidia and Oracle plan multi-gigawatt Vera Rubin cloud capacity
On October 27, 2026, Nvidia and Oracle said they plan to expand Oracle Cloud Infrastructure (OCI) Superclusters built on Vera Rubin NVL72 systems to multi-gigawatt scale, citing demand for training, inference and agentic applications. The same announcement said Vera Rubin NVL72 racks and Nvidia Vera CPUs are already generally available on OCI. The multi-gigawatt build-out is the planned part. The expansion builds on Oracle's existing GB200 and GB300 NVL72 infrastructure rather than starting a new Nvidia relationship.
Each NVL72 rack links 72 GPUs through Nvidia NVLink scale-up networking. Superclusters add Oracle Acceleron networking, Nvidia Spectrum-X Ethernet with Spectrum-6 switches and BlueField-4 DPUs, and gigawatt-class power and cooling. Oracle says its multiplanar network can scale across hundreds of thousands of GPUs. Customers can scale CPU and GPU capacity independently, which the companies say reduces idle capacity. That split suits GPU-heavy training and inference alongside CPU-heavy agent work between model calls. Oracle also plans to introduce HGX Rubin NVL8 systems, an eight-GPU x86 configuration.
The RIO Rundown Oct. 28 roundup flagged the item in one line about NVIDIA and Oracle expanding multi-gigawatt capacity, and this brief draws the detail from the companies' announcement and trade coverage. Pricing, delivery dates and named customers were not disclosed in the material reviewed, so treat the gigawatt figure as a target.
Why it matters
Power and building space now limit how quickly AI capacity can come online, so a gigawatt-scale plan says as much about data center readiness as about GPUs. Nvidia has cited a rough benchmark of about $40 billion in revenue opportunity per gigawatt for the Vera Rubin generation, a figure that would be very large if it converts to contracts. For founders, the practical question is which clouds can actually rent this hardware in the near term. For investors, the gap between announced and delivered capacity is where risk sits.
Operators building agents can match CPU and GPU pools to their own workloads instead of paying for idle accelerators, which the companies say should improve utilization. Vendors selling model serving or infrastructure software should expect buyers to compare Oracle with Google Cloud, since both are moving Vera Rubin into their offerings. Investors should ask for lead times, power commitments and general availability dates before treating gigawatt figures as revenue.
Key points
- 1Nvidia and Oracle announced plans on October 27, 2026 to expand OCI Superclusters to multi-gigawatt scale.
- 2Vera Rubin NVL72 racks and Vera CPUs are already generally available on Oracle Cloud Infrastructure.
- 3Oracle plans HGX Rubin NVL8 systems, an eight-GPU x86 server configuration, but has not dated them.
- 4Nvidia cites about $40 billion in revenue opportunity per gigawatt for Vera Rubin, a vendor estimate.
- 5
