Ambient Global Compute - Orchestrating the Non-Elastic Cloud With Kubernetes

Explore how Kubernetes is evolving beyond single clusters to orchestrate workloads across non-elastic, global cloud environments in this 26-minute conference talk. Discover the reality that cloud compute is no longer the infinitely elastic utility it once appeared to be, as customers increasingly depend on reservations and sophisticated capacity management to handle scarce, specialized hardware like GPUs and TPUs. Learn about the concept of "Ambient Global Compute" and how Kubernetes is adapting to become the "operating system of the cloud for distributed systems." Gain insights from Google's engineering director leading open-source Kubernetes on key initiatives supporting this transformation, including Dynamic Resource Allocation (DRA), Multi-Kueue, Karpenter/ComputeClasses, Multi-Cluster Orchestrator (MCO), ArgoCD, and other tools that enable unified multi-cluster and often multi-cloud deployments. Understand how these components work together to support modern, demanding workloads while maintaining a simpler mental model despite the growing complexity in infrastructure, and discover practical strategies for leveraging Kubernetes' evolution in your own distributed systems.