Explore the advantages and challenges of serverless computing for machine learning inference on Kubernetes in this insightful conference talk. Delve into the results of extensive benchmarking experiments comparing serverless and traditional computing for inference workloads running on Kubernetes, using KubeFlow and the ModelDB MLOps Toolkit. Gain valuable insights into various model types, data modalities, hardware configurations, and workloads. Learn how to architect your own Kubernetes-based ML inference system and understand the trade-offs between flexibility, operating costs, and performance. Discover whether serverless computing is truly a panacea for elastic compute in ML inference or if its limitations outweigh its benefits.