Oracle will offer Google TPUs through its public cloud starting next year, the first time Google has supplied the chips to another hyperscaler. The plan also covers Cerebras wafer-scale accelerators and OpenAI Jalapeño inference processors. In parallel, Oracle will connect its business applications to ChatGPT and use OpenAI models for new cloud services. For companies, access to scarce AI capacity and ready chatbot workflows now comes from one vendor.
TPU deployment and new OpenAI services
Oracle plans to gradually expand TPU capacity after the initial deployment next year. The TPUs in its data centers will exchange data with each other using Google ICI interconnects over copper and OCS optical switching. Connection to the rest of the data center will run over Oracle Acceleron networking, built around a custom SmartNIC for network management and cybersecurity tasks. CNBC reported that Oracle also intends to deploy Cerebras chips, each the size of a silicon wafer with trillions of transistors, alongside OpenAI Jalapeño.
Google OCS avoids the usual conversion of light signals into electricity and back by steering light beams with tiny mechanical mirrors, which lowers power usage. Oracle describes Acceleron as combining a SmartNIC with hardware-level tenant isolation, High Performance Fabric and Zero Trust Packet Routing. The stated result is high throughput and low latency movement of data between Oracle databases, storage and TPU clusters. Senior executives Karan Batta of Oracle and Gurmeet Goindi of Google Cloud detailed the design in a joint blog post. The architecture keeps Google interconnects inside the cluster and Oracle networking outside it.
OpenAI said in August that Jalapeño can run some open-source models 1.5 times faster than competing silicon. The broader Oracle-OpenAI partnership adds application access through ChatGPT and several model-powered services. AI Workspace will let the chatbot perform tasks inside Oracle applications for managing customer records and other daily operations, while calling external applications when needed. App Composer will let non-technical users build software with AI assistants, scan it for vulnerabilities and deploy it on managed infrastructure. AI Security Mesh will use cybersecurity-optimized OpenAI models to find, rank and suggest fixes for weak points.
What this means for AI buyers
For midsize and large companies running Oracle databases and business applications, the combination reduces friction between compute and data. AI Workspace fits multi-step work such as resolving a customer order threatened by a supplier delay, from finding an alternative product and checking inventory to validating transport, assessing financial impact and updating the order. IT keeps control over access, approved data, policies and costs while Oracle manages deployment and operations. Small Oracle shops gain less from TPUs but can test automation without rebuilding their application landscape.
The limits deserve attention before any decision. Oracle has not disclosed TPU volumes, pricing, regions or service-level terms, and Cerebras and Jalapeño timelines remain unspecified beyond next-year start for TPUs. AI Security Mesh prioritizes vulnerabilities by severity, yet buyers should verify coverage, false-positive rates and remediation workflow with their own code. App Composer promises managed deployment, but governance over data sources and cost controls needs contractual clarity. The announcement alone does not guarantee availability or performance in a specific workload.
A practical marker is the start of TPU provisioning in Oracle Cloud next year and publication of capacity, pricing and supported configurations. A second marker is general availability of AI Workspace, App Composer and the Oracle AI SMB Bundle with $12,000 in annual cloud credits, NetApp QuickStart, ChatGPT Business and Codex. If both appear on schedule with reference customers, the bundle of chips plus chatbot workflows will have moved from partnership statements to procurement options.
