Google TPUs to be offered on Oracle Cloud in 2027

Google and Oracle plan to make Google’s Tensor Processing Units available through Oracle Cloud Infrastructure in calendar 2027, giving customers access to the chips outside Google’s own cloud. Oracle said OCI will be the first hyperscaler beyond Google to offer the accelerators. The companies announced the plan on October 27. Oracle will provide the power, cooling and networking needed for TPU deployments in its data centers, according to the companies. They have not disclosed the commercial terms or the initial capacity, chip configurations, prices or locations. Oracle plans to introduce the service during 2027 and expand capacity over several years, but has not announced a specific start date. The offering is intended to let customers run workloads on Google’s custom AI chips through OCI rather than using Google Cloud, the companies said. The addition gives Oracle customers another accelerator option for training and running AI models. TPUs are designed for machine-learning workloads, while GPUs are used across a broader range of computing tasks. How either type performs, and the costs of using it, can depend on the model, workload and software involved. The announcement...