Google Cloud: Google TPUs on OCI for AI Workloads on Oracle TV at Oracle AI World 2026

Scaling AI training and inference workloads while managing latency, infrastructure costs, and power consumption is a growing challenge for enterprises, and Google TPUs on Oracle Cloud Infrastructure (OCI) offer another option for deploying specialized AI computing systems. Explore how Google and Oracle are bringing TPU technology to OCI to support large-scale AI training, low-latency inference, and agentic AI workloads. Register for Oracle AI World On Air so you don’t miss a thing from Oracle AI World 2026: https://social.ora.cl/6003BGN9vJ How can Google TPUs on Oracle Cloud Infrastructure (OCI) help organizations scale AI workloads while improving performance, managing costs, and addressing power requirements? In this interview on Oracle TV at Oracle AI World 2026, George Ellis, vice president of product management for Google Cloud and AI infrastructure, discusses the announcement bringing Google's Tensor Processing Units (TPUs) to OCI and what it means for enterprise AI customers. The conversation explores what differentiates Google's TPU architecture from other AI computing technologies. Rather than focusing exclusively on individual chips, Google designs TPUs as integrated systems optimized for AI workloads, combining specialized processors, optical networking, and memory architecture to support large-scale computing. Ellis explains how this approach supports both vertical and horizontal scaling for demanding AI applications. A central focus is the growing diversity of AI workloads, including model pretraining, post-training, batch inference, low-latency inference, and agentic AI systems. As AI agents increasingly communicate with other agents, the speed of model inference becomes especially important. Ellis discusses how TPU architecture and high-speed static random-access memory (SRAM) can help improve performance per cost and performance per watt. The interview also examines Google's eighth-generation TPU technology, including separate designs optimized for AI training and inference. Ellis describes how these specialized architectures address different workload requirements, from large-scale interconnected training systems to inference environments where low latency is essential. He also discusses Google's reported 80% reduction in inference costs with its newest TPU generation. Why bring Google TPUs to OCI? Ellis highlights Oracle's experience delivering high-density, large-scale, liquid-cooled data centers and the engineering requirements involved in deploying specialized AI infrastructure. He also discusses the existing collaboration between Google Cloud and Oracle, including the availability of Oracle databases in Google Cloud, and how the companies are extending that relationship to support shared customers. Part of Oracle TV's coverage of Oracle AI World 2026, this interview explores Google TPU architecture, AI infrastructure scalability, agentic AI performance, inference economics, and the customer benefits of bringing Google's specialized AI computing technology to Oracle Cloud Infrastructure. Like what you learned about Oracle AI World 2026? Subscribe now https://social.ora.cl/60025k87j Follow us here: LinkedIn: https://social.ora.cl/6000B6oR6p X: https://social.ora.cl/6002B6oRDG Instagram: https://social.ora.cl/60095eXDN Facebook: https://social.ora.cl/60065enQp