Oracle offers Google TPUs in its cloud for the first time

Oracle is about to stock a chip that, until now, you could only rent from Google. The two companies said Google’s Tensor Processing Units, or TPUs, will be offered on Oracle Cloud Infrastructure (OCI) starting in 2027.

The deal makes OCI the first major cloud provider outside Google to offer the chips.

What Oracle and Google announced

The partnership was unveiled on October 27, 2026, at Oracle AI World in Las Vegas.

TPUs are set to debut on OCI in 2027. A phased rollout is then expected to scale up over several years.

The pitch to customers is choice. Companies building AI systems could run different chip architectures inside a single cloud platform, rather than stitching together multiple providers.

The hardware in question is Google’s latest generation. That includes the TPU 8t, designed for training AI models, and the TPU 8i, built for inference.

Training is the expensive phase where a model learns from data. Inference is the part where the finished model actually answers your questions, over and over, at scale.

The TPU 8t is billed as delivering nearly three times the compute performance per pod compared with the previous generation. A pod is a large cluster of chips linked together to work as one system.

Oracle hedges its chip bets

Oracle’s AI chip lineup has leaned heavily on Nvidia GPUs. Adding TPUs gives Oracle a second major option to put in front of customers.

Nvidia is not being shown the door, though. Oracle is also extending its collaboration with the chipmaker, working toward multi-gigawatt scale with Nvidia’s next-generation Vera Rubin systems.

Oracle is also bringing in hardware from Cerebras Systems and SambaNova Systems.

How markets reacted

Investors liked what they heard. Oracle shares rose nearly 5% following the announcement.

Alphabet, Google’s parent company, saw a modest uptick in its stock as well.

The move also builds on an existing relationship. Oracle and Google already have multicloud ties, including Oracle Database@Google Cloud, which brings Oracle’s database services into Google’s cloud.

What this means for the cloud AI race

For Oracle, the clearest benefit is differentiation. Being the first major cloud outside Google to offer TPUs gives OCI something competitors cannot currently match.

Customers who want TPUs but prefer not to build everything inside Google Cloud now have an alternative route, at least once the 2027 rollout begins.

The timeline deserves attention. TPUs arrive in 2027, and the scaling is expected to stretch over multiple years, so meaningful capacity will not appear overnight.

For Google, the arrangement opens a new distribution channel for its chips. TPUs have historically been a Google Cloud exclusive, so placing them inside a rival’s infrastructure widens their potential reach.

For enterprise buyers, the practical upside is optionality. Teams could match workloads to hardware, using training-focused chips like the TPU 8t for model development and inference-focused chips like the TPU 8i for serving users.

The early market response, with Oracle up nearly 5%, shows investors are willing to give the strategy credit before the chips even arrive. The harder test comes in 2027, when the TPUs are scheduled to start showing up on OCI.

Disclosure: This article was edited by Diego Almada Lopez. For more information on how we create and review content, see our Editorial Policy.