Oracle signs deals for Google and other AI chips, keeping Nvidia in the mix

Oracle’s cloud customers are set to get more ways to run AI without relying solely on Nvidia hardware. Oracle has signed deals to bring Google, Cerebras and SambaNova chips into its cloud infrastructure, Yahoo Finance reported on October 27. Oracle also said it is bringing Nvidia’s Vera Rubin NVL72 superclusters and Vera CPUs to Oracle Cloud Infrastructure, known as OCI.

Google chips take a different route

The Google agreement changes how its Tensor Processing Units, or TPUs, reach customers. According to Yahoo Finance, this is the first time Google is selling those processors for use inside a third-party data center, rather than simply renting customers computing capacity.

Yahoo Finance places the agreements within a push by companies to avoid dependence on a single chip supplier and contain costs. That rationale is not a measured savings result for Oracle’s new offerings.

As agents take on more complex work, customers demand compute that matches each application’s performance needs.

Mahesh Thiagarajan, president, Oracle Cloud Infrastructure

The repeated work behind an AI agent

Cerebras and SambaNova are being positioned for inference—the work of running an AI model to generate a response. Oracle said Cerebras chips will supply inference capacity and AI-agent processing, while SambaNova’s SN50 processor will handle agentic inference, or model processing for agents carrying out tasks.

The Cerebras announcement explains the intended benefit through the delays that accumulate in multistep applications. Each additional model call adds waiting time. An agent may need inference while reasoning, checking its work and deciding on an action, so response speed affects more than the wait for a single answer.

Cerebras addresses that problem with its Wafer-Scale Engine, which combines computing and memory on one wafer-scale processor. The company says this design reduces data movement and communication between chips. Its stated benefit is fast generation of tokens—the pieces of an AI response—during lengthy reasoning and multistep agent workflows.

The partners are targeting AI labs, enterprises and developers building interactive applications. The partnership builds on Oracle’s existing inference capabilities, including Dedicated AI Clusters in OCI Enterprise AI.

In their October 27 announcement, the companies said Cerebras Inference on OCI is planned for calendar year 2027. The announcement cautions that this is not a delivery commitment and should not guide purchasing decisions; timing and pricing may change.