What Is the Tenant Plane?
The tenant plane is the layer of a token factory that makes infrastructure safely multi-tenant. It covers metering, isolation, quotas, per-token billing, and governance.
In a three-plane model:
| Plane | Responsibility |
|---|---|
| Data plane | Where inference actually runs: GPUs, engines, model weights |
| Control plane | Scheduling, routing, autoscaling, placement |
| Tenant plane | Who is allowed to use what, how much they used, and what they owe |
Why It Is the Missing Layer
The data plane and the control plane are well served by open source. Engines, schedulers, and orchestration tools such as NVIDIA Dynamo have become a commodity layer that any operator can adopt.
No open source component ships the tenant plane. Metering that is accurate enough to bill from, isolation strong enough to put two customers on the same fleet, quotas that hold under load, and rate cards that map token counts to invoices are all things an operator otherwise has to build. This is the layer that turns a GPU cloud into a sellable, billable service rather than a cluster.
Saturn Cloud runs the tenant plane so operators do not have to construct metering, isolation, and billing from scratch. It deploys on infrastructure the operator already owns, above whatever data plane and control plane they have chosen.
