What this guide shows
This poster separates node-level components from service and control-plane workloads for production AI platforms on Kubernetes.
It makes it easier to reason about:
- what must run on every node
- what should run as a Deployment or StatefulSet
- which operators manage GPU, networking, storage, and observability components
- the platform services commonly layered above the core GPU stack
Best for
- teams documenting a standard production AI platform bill of materials
- engineers explaining why some components must live on nodes while others stay in the control plane
- reviewers validating cluster architecture before workload onboarding
Operational value
This guide reduces design ambiguity. When teams confuse node agents, platform services, and application workloads, troubleshooting becomes slower and ownership gets blurred.