Oracle to bring Google TPUs to its cloud and adds ChatGPT integrations

Oracle plans to make Google's TPU AI accelerators available through its public cloud, the first time Google supplies the chips to another hyperscaler. According to CNBC, Oracle will also add Cerebras wafer-scale accelerators and OpenAI's Jalapeño inference processor. Oracle will start deploying Google's chips next year and plans to add more TPU capacity gradually after that.

The TPUs in Oracle's data centers will use Google's OCS and ICI technologies to exchange data, and will connect to the rest of the data center through Oracle's Acceleron networking, which runs on a custom SmartNIC. On the application side, Oracle is rolling out AI Workspace, which lets ChatGPT perform tasks in Oracle applications for workers and bring in external applications when needed.

Oracle is also launching App Composer, which lets non-technical users write custom software with AI assistants, scans it for vulnerabilities and deploys it on managed cloud infrastructure. Its AI Security Mesh tool uses OpenAI cybersecurity-optimized models to find weak points in customer applications and rank them by severity. A small-business bundle combines these services with ChatGPT Business and Codex, and Oracle says customers will receive $12,000 in cloud credits per year.

Teams building agents on Oracle Cloud may get a new accelerator option for running models, although deployment of the TPUs is only planned for next year. Agents connected through the ChatGPT integrations can act inside Oracle applications, so permissions and data access for those connections should be reviewed before rollout.