A step-by-step guide to running isolated Kubernetes clusters for multiple teams on shared GPU hardware using two open source tools: KAI Scheduler and vCluster. KAI Scheduler handles GPU resource allocation with hierarchical queues and per-team quotas, while vCluster gives each team its own isolated Kubernetes control plane (API server, CRDs, RBAC) without splitting the physical infrastructure. The tutorial walks through setting up MicroK8s with the NVIDIA GPU Operator, installing KAI Scheduler with GPU sharing enabled, defining team queues with guaranteed GPU fractions, spinning up one vCluster per team, and deploying GPU workloads that each team can only see within their own cluster. The result is three teams sharing a single NVIDIA L40S GPU with full workload isolation and no wasted hardware.
