Oracle brings Google’s TPUs to its cloud, adds ChatGPT integrations

Oracle has announced that it will make Google's TPU artificial intelligence accelerators available through its public cloud. This marks the first time that Google is supplying chips to another hyperscaler. In addition, Oracle will also add Cerebras Systems' wafer-scale accelerators and OpenAI's Jalapeño inference processor to its cloud offerings.

The TPU chips utilize two data transfer technologies, ICI and OCS. ICI transmits data using copper wires while OCS uses optical links to skip the conversion of light signals to electricity and back. This optical link technology reduces power usage by employing tiny mechanical mirrors to steer light beams to the recipient servers.

The chips in Oracle's data centers will use both OCS and ICI technology to exchange data with one another. They will connect to the rest of the data center using Oracle's Acceleron networking technology, which is powered by a custom SmartNIC optimized for network management and cybersecurity tasks. Oracle plans to gradually add more TPU capacity to its cloud in the coming year.

The company will also deploy Cerebras' AI chips and OpenAI's Jalapeño inference accelerator. Oracle's partnership with OpenAI extends to integrating its applications with ChatGPT and powering new cloud services using OpenAI models. This integration allows customers to access Oracle applications through ChatGPT and use Oracle's applications to perform tasks on workers' behalf, such as managing customer records.

Oracle has also launched a new "free-to-start" product suite called the Oracle AI SMB Bundle, which is geared towards small businesses. The bundle includes AI Workplace, App Composer, the NetApp QuickStart accounting platform, ChatGPT Business, and Codex. Customers will receive $12,000 in cloud credits per year for using the suite.