GPU fleet occupancy simulator

Price occupancy points, idle hours between contracts, and fractional sales — the numbers a leasing business actually manages.

A provider that leases GPU-hours does not optimise tokens; it optimises billed hours as a share of leasable hours. This simulator prices the three levers of that business: occupancy itself, the unbillable gap between one tenant leaving and the next arriving, and how divisible the inventory is.

01: The fleet and the rate card

02: Re-lease friction

Leasable hours, where they go

Each cell is 1% of leasable GPU-hours. Blue is billed. Amber is lost to re-lease gaps. Hollow is unsold.

  • Billed
  • Re-lease gaps
  • Unsold

More Info

Occupancy is a leasing provider's version of goodput: the share of paid-for capacity that is actually earning. The levers are the same ones that improve token throughput on an inference platform—isolation, policy, and provisioning that live at the network layer rather than in host configuration—expressed in a different currency. A tenant handover that needs a re-image is unbillable time; a handover that is an API call is not.

Frequently asked questions

Where this leads

From AI factory math to AI factory control

Tenant isolation enforced below the host is what makes fractional and multi-tenant sales safe, and API-driven policy is what turns a three-day re-lease into an afternoon. Both land directly in this arithmetic.