Oracle adds Google, Cerebras and SambaNova chips alongside Nvidia

Oracle has signed agreements involving Google, Cerebras and SambaNova to expand its cloud hardware lineup while retaining Nvidia systems. Google is selling its Tensor Processing Units, or TPUs, for use in a third-party data center for the first time, rather than offering access only through rented computing capacity. Oracle also plans to offer Nvidia’s Vera Rubin NVL72 superclusters and Vera CPUs.

The additions focus largely on inference, the processing needed to run trained models and generate responses. Cerebras chips will support inference and AI-agent processing, while SambaNova’s SN50 processor will handle agentic inference. Agents can make repeated model calls while reasoning, checking results and choosing actions, adding delays throughout a workflow. Cerebras says its wafer-scale design combines computing and memory to reduce data movement between chips and accelerate token generation.

The broader strategy responds to demand for alternatives to dependence on a single chip supplier and greater control over costs, but no measured savings have been established for Oracle’s new services. Cerebras Inference on Oracle Cloud Infrastructure is planned for calendar year 2027. The companies caution that timing and pricing may change and that the announcement is not a delivery commitment.