Oracle will deploy Google’s Tensor Processing Units inside its own data centres and offer them on Oracle Cloud Infrastructure, making it the first hyperscaler outside Google to offer the chips since Google said in April 2026 it would sell TPUs to a select group of customers. Oracle cloud chief Karan Batta described the arrangement to Datacenter Dynamics as a multi-year, multi-gigawatt partnership, with new Google TPU generations available on OCI on day one. Google TPUs on OCI will be available in calendar year 2027, ramping over a multi-year period.
The current generation splits by workload. The TPU 8t is designed for training, while the TPU 8i is an inference chip scalable to 1,152 chips in a single pod, delivering 11.6 exaflops of FP8 compute with 331.8 terabytes of high bandwidth memory capacity per pod.
Alongside the Google deal, Oracle announced general availability of Nvidia Vera Rubin NVL72 systems and Vera CPUs on OCI. Each NVL72 rack connects 72 GPUs through NVLink, and Oracle said the systems will be added to its Superclusters offering at multi-gigawatt scale, building on existing Grace Blackwell capacity that OpenAI used with a Supercluster of more than 100,000 Blackwell GPUs.
The same event filled out the rest of the shelf. Oracle announced a multi-year Intel Xeon collaboration, said AMD Epyc based E7 instances are nearing launch, and said Cerebras inference is planned on OCI in calendar year 2027, with SambaNova and Positron systems also in the line up. The Register noted Oracle also unveiled OpenAI powered products, Oracle AI Workspace and Oracle AI Security Mesh, at the event. Taken together, the announcements make OCI the broadest multi-vendor AI compute menu among the large clouds.