Oracle to Offer Google's TPU AI Chips on Its Cloud From 2027

Oracle and Google Cloud will begin deploying Google's artificial intelligence chips on Oracle Cloud Infrastructure in 2027, the companies announced on 27 October at Oracle AI World in Las Vegas. Under the deal, Oracle will become the first large-scale cloud provider outside Google to offer Google's Tensor Processing Units, or TPUs, specialised processors built to speed up the maths at the heart of machine learning. Prices and initial service regions have not yet been announced, and the rollout will scale up over several years.
The idea is to let businesses run AI work close to their Oracle databases, rather than moving data between different clouds. Google's TPU 8t chip is designed for training large models, while the TPU 8i is tuned for inference, the day-to-day job of running a trained model for users. The deal is part of a wider Oracle push to offer customers a choice of AI hardware: alongside Google's TPUs, Oracle is adding Cerebras inference systems and scaling up Nvidia's latest Vera Rubin systems, positioning its cloud as neutral ground rather than a single-chip shop.
For Oracle, the timing matters. Companies are spending heavily on AI infrastructure but increasingly want proof that each kind of chip earns its place, and chip choice has often forced a cloud choice by default. Oracle says its customers will now be able to pick the architecture that fits each job without leaving its platform. Investors reacted positively on Tuesday, with Oracle shares rising 3.5 percent and Cerebras shares jumping 10 percent after its own Oracle partnership was confirmed.
Source: read more
Comments
Post a Comment