Source: Microsoft Foundry — Provisioned throughput billing and cost management
How Foundry bills PTUs (hourly on deployed capacity, prorated) versus 1-month/1-year Azure Reservations, why not to scale PTUs with traffic, and how reservation matching and overage work.
Contents
- What: vendor documentation
- Author: Microsoft
- Date: ms.date 2026-05-22, read 2026-11-14
- Original: Provisioned throughput billing and cost management
Summary
Covers the economics of PTU deployments and reservations, and the capacity risks of resizing.
Key claims
- PTUs are billed hourly on deployed count, whether used or not, prorated by the minute; deployments cannot be paused, only deleted.
- Hourly billing suits benchmarking and short events, not production; reservations are cheaper for sustained load.
- Don’t scale PTUs up and down with traffic: scale-up needs available regional capacity and can fail, and scale-down releases capacity permanently.
- Reservations (1 month or 1 year) are a billing discount, not capacity. They match deployments by deployment type, region (except Global) and scope, not by model. Excess deployed PTUs bill hourly. Deploy first, then reserve.
- One Global reservation can cover Global deployments across regions (e.g. 50 + 100 + 200 PTUs in three regions → one 350-PTU reservation).
- Reservations can be exchanged (resets term) or cancelled (possible fee); exchanges across deployment types are not supported.