Oracle will begin running Google's tensor processing units (TPUs), the chips Google designs for artificial intelligence, on its cloud platform, according to The Information.
Oracle announced the move on Tuesday. It is the first time Google's AI chips will be available on a cloud platform outside Google's own.
Availability is set for 2027, with capacity expected to expand over several years.
Google designed TPUs to handle the full AI workload cycle, including large-model training, post-training, inference, reasoning and agentic applications.
The deal splits chip design from cloud infrastructure.
Google supplies the TPU technology and software ecosystem. Oracle provides the data centre power, cooling and networking to run it at scale.
The move fits Oracle's wider push to diversify the accelerator options on Oracle Cloud Infrastructure.
Alongside the Google deal, Oracle has announced expanded access to Nvidia Vera Rubin systems and CPUs, plus offerings from Cerebras and SambaNova.
The aim is to let customers match chips to workloads rather than depend solely on Nvidia GPUs.
Read more about Cloud Infrastructure
