Back to home
Use Case
AI / GPU Infrastructure
AI agents are evolving from running on VMs to managing entire clouds. They need real infrastructure primitives — not sandboxed environments or limited APIs.
kplane gives each agent its own Kubernetes control plane — full API access over shared GPU infrastructure, provisioned in seconds.
Why agents need Kubernetes
- Full Kubernetes API — not a sandboxed subset
- Native resource management (pods, services, jobs, CRDs)
- Standard tooling works: kubectl, Helm, operators
- Orchestrate multi-step pipelines with real primitives
- Scale infrastructure programmatically via the API
Why kplane
- ~3 MB overhead per control plane — run thousands concurrently
- Provision in ~2 seconds — fast enough for ephemeral sessions
- Share GPU node pools across all control planes
- Full isolation at the API level — agents can't see each other
- No per-cluster infrastructure to manage or pay for
The infrastructure shift
Early AI agents ran simple scripts on VMs. The next generation manages fleets of containers, schedules GPU workloads, and orchestrates multi-step pipelines. Giving every agent its own Kubernetes cluster is expensive and slow. With kplane, you provision a control plane in seconds — agents get full Kubernetes semantics while you keep GPU pools shared.