Oracle to Offer Google TPUs in Public Cloud, Adds ChatGPT Integrations

Oracle will begin deploying Google’s chips in its data centers next year and gradually add more TPU capacity, the company said. CNBC reported that Oracle also plans to add Cerebras Systems Inc.’s wafer-scale accelerators and OpenAI Group PBC’s Jalapeño inference processor; the latter chip is part of a broader partnership that includes several newly announced cloud services. In August, the ChatGPT developer said the Jalapeño chip can run some open-source models 1.5 times faster than competing silicon.

Google’s TPUs exchange data using two technologies called ICI and OCS. ICI transmits data via copper wires, while OCS uses optical links. Sending data over an optical link usually requires servers to turn light signals into electricity and then back into light; Google’s OCS technology skips that step, lowering power use, and uses tiny mechanical mirrors to steer light beams toward recipient servers.

In Oracle’s data centers, the TPUs will use OCS and ICI to exchange data with one another and connect to the rest of the host data center through Oracle’s Acceleron networking technology. The acceleration is powered by a custom SmartNIC, a chip optimized for network management and cybersecurity tasks.

“Acceleron’s SmartNIC with hardware-level tenant isolation, High Performance Fabric, and Zero Trust Packet Routing moves data between Oracle databases, storage, and TPU clusters with high throughput and low latency,” Oracle senior vice president Karan Batta and Google Cloud senior director Gurmeet Goindi wrote in a blog post.

Oracle’s partnership with OpenAI also extends to applications and cloud services. The companies announced integrations that will let customers access Oracle applications through ChatGPT, and Oracle will use OpenAI models to power a series of new cloud services.

Oracle is rolling out a service called AI Workspace that will integrate its cloud applications with ChatGPT. The integrations will let the chatbot perform tasks in Oracle applications on workers’ behalf and bring in external applications when necessary.

“Consider a customer order put at risk by a supplier delay,” Oracle product marketing director Michelle Meng wrote in a blog post. “Workspace can maintain the goal throughout the process while coordinating the applications, agents, data, and tools needed to address it.”

Oracle also introduced App Composer, a tool that lets non-technical users write custom software using AI assistants. App Composer scans that software for vulnerabilities and deploys it on managed cloud infrastructure.

“Oracle manages deployment and operations, while IT controls access, approved data, policies, and costs,” said Blair Bozada, a senior product marketing manager at Oracle.

App Composer is rolling out alongside AI Security Mesh, a vulnerability remediation tool that uses OpenAI’s cybersecurity-optimized AI models to find weak points in customers’ applications. It prioritizes vulnerabilities by severity and suggests a fix for each one.

A new “free-to-start” Oracle AI SMB Bundle for small businesses includes AI Workplace, App Composer, the NetApp QuickStart accounting platform, ChatGPT Business and Codex. Oracle said customers will receive $12,000 in cloud credits per year.