Oracle used its AI World conference on Oct. 27 to make two announcements about Oracle Cloud Infrastructure (OCI) that point in different directions. The first ties OCI more closely to Nvidia. Oracle and NVIDIA said Vera Rubin NVL72 rack systems and NVIDIA Vera CPUs are now generally available on OCI, and they described plans to grow OCI Superclusters built on Vera Rubin to what the companies call "multi-gigawatt scale." The second spreads OCI across more chipmakers. Oracle said OCI will host Google's Tensor Processing Units, Cerebras systems and other non-Nvidia hardware, with the TPUs and Cerebras expected in 2027.

Taken together, the announcements show a cloud provider adding a lot of Nvidia capacity while also making room for competing accelerators in its own data centers. Here is what was announced, what is shipping now and what is still only planned.

What is generally available today?

According to Oracle's press release, two NVIDIA products are now generally available on OCI: the Vera Rubin NVL72 rack-scale system and the Vera CPU. In each NVL72 rack, 72 GPUs are linked over NVIDIA's NVLink interconnect.

The release says OCI Superclusters built on these racks use Oracle's Acceleron networking together with NVIDIA Spectrum-X Ethernet, running on Spectrum-6 switches and BlueField-4 data processing units. Oracle describes the Superclusters as having "gigawatt-class power and cooling." The release gives no sites, megawatt totals for particular campuses or construction schedules for the multi-gigawatt expansion.

What is planned but not yet available?

Oracle also described a planned HGX Rubin NVL8 offering. This is a smaller eight-GPU configuration than the 72-GPU NVL72 rack. According to the press release, each system would have eight Rubin GPUs, two 64-core Intel Xeon 6 CPUs, 4 TB of DDR5 memory and 6.4 Tbps of RDMA networking. The release gives no availability date for it.

The multi-gigawatt Supercluster expansion and the HGX Rubin NVL8 option are both covered by the release's forward-looking-statement disclaimers. Neither should be treated as a commitment to a particular capacity or date. Tech Wire Asia reported that the companies did not disclose the financial value of the Nvidia expansion.

What does OpenAI have to do with it?

Oracle's release offers an existing customer deployment as evidence of what OCI Superclusters can do. According to the release, OpenAI trained and served its GPT-6 Astra model on an OCI Supercluster with more than 100,000 NVIDIA Blackwell GPUs in Abilene. The release includes statements from OpenAI's Sreekanth Pothanis and Uday Ruddarraju, along with Oracle's Mahesh Thiagarajan and NVIDIA's Ian Buck.

BigGo Finance reported that OCI will also offer OpenAI's in-house chip, which BigGo's report calls Jalapeno. Oracle's NVIDIA press release does not mention this chip, and the sources listed here give no technical details or availability date for it.

Why put Google TPUs inside OCI?

According to Tech Wire Asia, OCI will be the first hyperscale cloud other than Google Cloud to offer Google's TPUs. The arrangement covers two variants: TPU 8t for training and TPU 8i for inference. The outlet reported that the TPUs are expected to be available in 2027. Oracle will supply the data center infrastructure, power, cooling and networking for them.

That division of labor matters for facility planning. Under the arrangement Tech Wire Asia described, Oracle provides the data center infrastructure, power, cooling and networking for the TPUs. So the TPU deal could add to the physical load Oracle has to plan for, on top of its Nvidia buildout, although the sources give no capacity figures.

Where do Cerebras and SambaNova fit?

Tech Wire Asia reported that Cerebras systems will be available on OCI in 2027 and that SambaNova SN50 inference racks are also planned. BigGo Finance, citing a joint statement, put the Cerebras deployment more precisely in the first half of 2027. Neither outlet gave capacity figures for the Cerebras or SambaNova deployments.

Why is Oracle doing this?

BigGo Finance reported that Thiagarajan, president of OCI, pointed to supply. He said tight chip supply remains a challenge for the industry. He also said the multi-vendor approach does not mean Oracle is turning away from Nvidia, which he called the "best and most general-purpose" option, according to BigGo.

Thiagarajan's description fits the announcements themselves. The Nvidia news covers hardware that is generally available now plus plans to scale it to multiple gigawatts. The Google TPU and Cerebras additions are scheduled for 2027 and are described mainly in terms of training and inference workloads.

What were the financial and market figures?

Tech Wire Asia reported the following figures for Oracle's first quarter of fiscal 2027, which ended Aug. 31, 2026:

  • Cloud infrastructure revenue of $7.4 billion, up 121%.
  • Total revenue of $19.3 billion, up 30%.
  • Remaining performance obligations of $664 billion.

These are Tech Wire Asia's figures as published. BigGo Finance reported that Oracle shares rose nearly 5% on the day of the announcements and that Cerebras shares rose about 10% at one point. DataCenterFeed does not provide investment advice. These figures describe reported results and one day's trading, and they say nothing about future performance.

Why it matters beyond the cloud

For utilities and local officials, the most important phrase in the announcement is "gigawatt-class power and cooling," and it comes with no locations attached. Oracle's release commits only to expanding Vera Rubin Superclusters to multi-gigawatt scale. It does not name new campuses, interconnection requests or water arrangements. Separately, the TPU arrangement as Tech Wire Asia described it makes Oracle responsible for power and cooling for Google hardware too. That means the facility demand from these announcements is not limited to Nvidia systems. The only site Oracle's release names is the existing Abilene Supercluster that OpenAI used; BigGo Finance separately mentions Oracle's Project Jupiter campus in Sunland Park, New Mexico, in the context of local opposition, but ties it to none of these chip announcements.

What Happens Next

The sources give the following schedule:

  • Vera Rubin NVL72 and Vera CPUs are generally available on OCI now, according to Oracle's press release.
  • Google TPU 8t and TPU 8i are expected on OCI in 2027, according to Tech Wire Asia.
  • Cerebras is scheduled for the first half of 2027, according to the joint statement cited by BigGo Finance.
  • No availability dates were given for the HGX Rubin NVL8 configuration, the SambaNova SN50 inference racks, the OpenAI chip or the multi-gigawatt Supercluster expansion. Oracle describes the planned items as forward-looking.

Sources