Oracle Cloud to offer Google TPUs, the first hyperscaler outside Google to do so

Oracle and Google announced on October 27 at Oracle AI World in Las Vegas that Google's Tensor Processing Units (TPUs) will be available through Oracle Cloud Infrastructure (OCI), the first time Google TPUs will be offered on a hyperscale cloud platform outside Google. OCI will provide the power, cooling and data-center networking to deploy the TPUs at scale, with availability starting in calendar year 2027 and ramping over a multi-year period.

Eighth-generation TPUs split training and inference

Google's eighth-generation TPU comes in two purpose-built architectures: TPU 8t for training, which the companies say delivers nearly 3x the compute performance per pod over the previous generation and cuts frontier-model development cycles from months to weeks, and TPU 8i for inference, designed for the iterative work of many specialized agents. The companies say TPUs can support large-model training and post-training, including dense and mixture-of-experts models, as well as production inference for reasoning, agentic and long-context applications.

Oracle's bet: several architectures in one cloud

Mahesh Thiagarajan, president of Oracle Cloud Infrastructure, said different AI workloads have different compute needs, and bringing Google TPUs to OCI gives customers access to multiple leading AI compute architectures in one cloud. Amin Vahdat, Google's senior vice president and chief technologist for AI and infrastructure, said the next generation of AI will require a diverse range of compute and that TPUs give more organizations access to the technology.

Separately on the same day, Reuters reported Oracle is also expanding its deal with Nvidia to multi-gigawatt scale with the newest Vera Rubin AI servers, while skipping some Nvidia networking chips in favor of its in-house Acceleron networking chips, and planning to offer inference chips from Cerebras and SambaNova. Oracle did not disclose the value of those deals, and OCI pricing for TPUs was not announced; real-world benefits will depend on deployment details and workload benchmarks.