In this episode of the Kubernetes Podcast from Google, recorded live at Google Cloud Next 2025, hosts Abdel and guests Nick Eberts and Jon Li discuss Multi Cluster Orchestrator (MCO), a new open-source tool announced at KubeCon EU 2025. Jon explains the problem MCO aims to solve, addressing the limitations of the "infinite and uniform" cloud assumption with the rise of accelerators like GPUs, which face stock outs and varying generations. Nick elaborates on the need for a middle ground between one massive cluster and numerous small ones, highlighting the importance of bin packing applications and ensuring high availability across regions. He details how MCO helps manage workloads across multiple Kubernetes clusters, especially for expensive accelerated hardware, by recommending optimal cluster placement based on priorities and capacity, and scaling workloads from zero to one based on demand. The discussion also covers the Cluster Inventory API and Cluster Profile API, components MCO builds upon, and how these profiles are automatically generated and updated to reflect cluster capabilities. Jon discusses how MCO will integrate with the inference extension, creating a two-layer approach to routing, first by picking a region and then balancing the load among accelerators within that region.
