Scott Hanselman talks with Omar Shorbaji from the Anyscale engineering team about how Anyscale on Azure scales Python AI workloads from a single notebook to thousands of CPUs and GPUs. Built on Ray, the most widely adopted AI compute engine, Anyscale gives you a unified runtime to build, train, and serve, running directly on Azure Kubernetes Service without the complexity of managing Kubernetes. See a live demo that fine-tunes a vision-language-action robotics policy, with the metrics you need to push GPU utilization higher. Chapters 00:00 - Introduction 00:52 - Ray and the Anyscale platform 03:11 - Start of demo: Workspaces 04:38 - Running a job and viewing utilization metrics 05:24 - Choosing the right scale 06:53 - Abstracting Kubernetes on AKS 08:53 - Wrap up and where to learn more Recommended resources Learn Docs Anyscale on Azure Connect Scott Hanselman | Twitter/X: @SHanselman Anyscale | Twitter/X: @anyscalecompute Azure Friday | Twitter/X: @AzureFriday Azure | Twitter/X: @Azure