Skip to main content
PRIVATE BETAThis feature is currently in Private Beta, contact sales@metronome.com for early access.Don’t have a Metronome account? Sign up to get started.
Provisioned throughput pricing is a pattern where a customer pays for a reserved baseline capacity (for example, “100 H100 GPUs”) and then pays an additional on-demand rate for usage above that reserved baseline. Metronome supports modeling these reserved instance commitments as contract-level pricing inputs. You store a customer’s reserved capacity (which may change over time) on the contract, and Metronome uses it when rating usage to split covered vs. overage consumption.

​
When to use this model

Use provisioned throughput pricing when:
  • Customers reserve capacity in advance (often with a recurring fee).
  • Actual usage is measured continuously (heartbeats or high-frequency usage events).
  • Billing needs to allocate usage into reserved (covered) usage up to the baseline and on-demand (overage) usage above the baseline.
  • The reserved baseline may change over time (for example, a customer increases reserved GPUs mid-contract).

​
Configuration overview

A typical Metronome configuration has four pieces:
  1. Ingest usage (for example, 30-second GPU heartbeats).
  2. Store reserved capacity on the contract as a time-varying value using Named Schedules (for example, “reserved A100 GPUs = 120 from Jan 1; 200 from Jun 15”).
  3. *Compute on-demand quantities using SQL billable metrics that reference the Named Schedules.
  4. Charge by pricing rates on the rate card

​
Products and rates

A common product setup for provisioned throughput billing:
  • Reservation fee (subscription): The recurring price the customer pays for the reserved capacity.
  • On-demand overage (usage): Metered overage above the reserved baseline.
  • Reserved-covered usage (optional): A $0 tracking product to display how much usage was covered by the reservation on invoices.

​
How Metronome allocates usage

At invoice time, usage is evaluated in consistent time windows (for example, 30-second slices), then split using the reserved capacity effective for that window:
  • covered = min(total_usage_in_window, reserved_capacity_in_window)
  • overage = max(total_usage_in_window - reserved_capacity_in_window, 0)
Metronome converts those quantities to your billing unit (for example, GPU-seconds to GPU-hours) and applies rates.