At its AI World conference, Oracle announced it is integrating Google TPUs, Cerebras Systems wafer-scale hardware, and Intel-backed SambaNova chips into Oracle Cloud Infrastructure (OCI). This major hardware diversification strategy marks the first time Google is selling its custom Tensor Processing Units (TPUs) to a third-party cloud data center rather than renting out capacity exclusively on Google Cloud.
Key Details of the Partnership
- Google TPUs — OCI will host Google’s custom TPUs to give customers wider structural flexibility over the throughput and economics of varied AI training and processing workloads.
- Cerebras Systems — Oracle will deploy Cerebras’s massive Wafer-Scale Engine computers inside its data centers starting in the first half of 2027. This setup integrates compute and memory on a single giant processor to run high-speed AI inference and AI agent processing at extremely low latency.
- SambaNova Systems — Oracle is deploying SambaRack systems equipped with the SN50 Reconfigurable Dataflow Unit (RDU) processor. These chips are specialized to power low-latency agentic inference at scale.
Strategic Objectives
- Reducing Nvidia Reliance — While Nvidia hardware still dominates AI model training, the market for “inference” (when an AI system processes a live user request) is highly fragmented. Oracle is building a multi-architecture cloud to prevent vendor lock-in and optimize costs for customers.
- Tailored Infrastructure Economics — By providing multiple chip variations, OCI allows corporate clients to pick hardware precisely tailored to whether they are building massive frontier models or running cheap, rapid-fire AI agents.
- Nvidia Stays in the Mix — Oracle is not abandoning Nvidia. Alongside these additions, Oracle expanded its partnership with Nvidia to install its latest Vera Rubin superclusters and Vera CPUs at a multi-gigawatt scale.
The Big Picture
Oracle maintained its massive fiscal 2027 capital-spending target of $90 billion to $95 billion to fund this comprehensive infrastructure expansion.
