DevOps Paradox DOP 344: KubeCon EU 2026 Review
Apr 1, 2026
A KubeCon EU 2026 review that spotlights Kubernetes shifting into an AI and inference platform. They cover vendor contributions like NVIDIA and Google, new CNCF sandboxes, and model routing becoming a networking primitive. Expect debates on micro VMs for secure inference, agents as first-class platform users, platform engineering culture bottlenecks, and whether CNCF is absorbing projects companies no longer maintain.
AI Snips
Chapters
Transcript
Episode notes
Kubernetes Became An AI Platform
- KubeCon EU 2026 showed Kubernetes shifting from a container orchestrator to an AI-first platform focused on inference and orchestration of non-container workloads.
- NVIDIA, Google, and Red Hat contributions (DRA drivers, KAI Scheduler, LLM-D) highlighted infrastructure-level AI integration across GPUs and TPUs.
Inference Pushes Operators Toward Micro VMs
- Inference workloads change the game: containers may be insufficient because models can 'escape', pushing operators toward micro VMs and different runtime choices.
- Victor and Whitney noted startup time and image size matter far less for terabyte-scale model weights than safe isolation and hardware support.
Model Routing Moves Into Kubernetes Networking
- Gateway API gained an inference extension, making model routing a first-class Kubernetes networking primitive tied to user sessions and multi-cluster inference.
- This elevates model routing needs into core networking concerns for large-scale, multi-node inference deployments.