Oracle to offer Google TPUs on OCI, a first outside Google

Oracle and Google said on October 27 that Google TPUs, the Tensor Processing Units Google designed for AI, will be offered through Oracle Cloud Infrastructure (OCI). Oracle says it is the first hyperscale cloud outside Google to sell them, according to the Oracle press release.

For anyone who builds on rented AI compute, this is a change in who sells what. Until now, TPUs ran in Google's own data centers. Oracle already sells NVIDIA GPU clusters, so OCI customers will be able to pick between the two kinds of chip in one cloud.

How Google TPUs will work on OCI

Oracle will supply the power, cooling and data center networking, and Google's TPUs will run inside it. Oracle's Mahesh Thiagarajan, president of OCI, said the point is choice: "different AI workloads have different compute needs."

The release names Google's eighth-generation TPUs. TPU 8t is built for training and TPU 8i for inference. Google says TPU 8t delivers nearly 3x the compute per pod of the previous generation. That is Google's claim, and the release gives no independent test or price.

When it arrives

Not soon. Oracle says Google TPUs on OCI will be available in calendar year 2027, ramping over several years. No pricing, regions or capacity figures were published.

Part of a bigger Oracle week

The same day, at its AI World event, Oracle announced a collaboration with OpenAI that puts ChatGPT and Codex into its business apps. NVIDIA also said Vera Rubin NVL72 systems and Vera CPUs are now generally available on OCI, with plans to expand to multi-gigawatt scale, in its own release. Taken together, Oracle is positioning OCI as a cloud that sells every major AI chip, not one vendor's.

What we don't know yet

• The price of a TPU hour on OCI, and how it compares with GPUs.
• Which regions get the chips first.
• Whether customers can run existing TPU code unchanged.

Watch for Oracle's pricing and region details as the 2027 launch gets closer.