M365Con 2027 – free online conference · 11–22 Jan 2027 · 120 sessions m365con.net

Provisioned throughput units (PTU)

Concept

Provisioned throughput units are the normalized unit used to reserve dedicated model processing capacity for Azure OpenAI deployments, as an alternative to pay-as-you-go token billing.

✦ CONCEPTProvisioned throughputunits (PTU)M365.FM
Title graphic: Provisioned throughput units (PTU) (Concept)

Also discussed

The episode about Provisioned throughput units (PTU) also covers LLMLingua, Model router (Microsoft Foundry), Pay-as-you-go pricing, Semantic caching, Chargeback and Microsoft Cost Management.

Description

Provisioned throughput units (PTU) are the normalized unit used to reserve dedicated model processing capacity for Azure OpenAI deployments. Microsoft's documentation describes a provisioned throughput unit as the unit of measure for provisioned deployments in Microsoft Foundry, where each unit represents a fixed amount of model processing capacity. PTUs serve as an alternative to pay-as-you-go token billing, in which capacity is shared and cost follows the number of tokens consumed.

The central idea is reservation. The chosen number of units is reserved exclusively for the deployment whether or not requests arrive. Because the capacity is not shared the way it is in standard deployments, latency is more predictable. The documentation positions provisioned throughput for steady traffic, latency-sensitive applications, high production volumes and real-time scenarios such as chats, copilots or agents.

Several parameters shape how PTUs are obtained and used. Quota is model-independent but is granted per subscription, region and deployment type. Each model has a minimum unit count and its own tokens-per-minute yield per unit, so the same number of PTUs delivers different throughput depending on the model. Deployment types include global provisioned, data zone provisioned and regional provisioned. An optional spillover feature can send overflow requests to a standard deployment when the provisioned capacity is exceeded, although it is not available for every model family.

Billing for PTUs is hourly per unit, regardless of how many tokens are consumed. Azure Reservations are offered as a discounted commitment option. Sizing a deployment relies on request rates, prompt and response sizes and cache rates, and the Azure portal includes a capacity calculator to support that estimate. The documentation also cautions that quota does not guarantee capacity, which can fluctuate, and that reservations do not guarantee it either. For that reason, creating the deployment first is the advised order of steps before committing to a reservation.

A typical decision around PTUs is whether workload patterns justify reserved capacity or whether pay-as-you-go billing is sufficient. Steady, high-volume or latency-sensitive production traffic matches the scenarios the documentation names for provisioned throughput, while variable or low-volume usage is the situation standard token billing is built for. The hourly billing model means the cost is tied to the reserved units rather than to usage, which is why sizing matters.

The M365.FM episode "How to Build Low-Cost AI Agents in Microsoft Cloud" places PTUs within the wider cost structure of building agents. In its overview of billing layers, the episode lists Provisioned Throughput Units as the cost of throughput in Foundry and gives a figure of $1 per PTU. It sets this beside other cost items such as agent logging and observability costs, which vary with logging volume, and the Microsoft Agent Pre-Purchase Plan, described as a unified plan with discounts on AI services at $19,000 for 20,000 ACUs. The episode says the pre-purchase plan covers over 30 services and offers a single governance path for AI projects, and it frames an understanding of these billing layers as a way to make better decisions and keep AI costs under control.

Overview compiled with AI from M365.FM episodes and the facts on this page. You can improve it with the edit button below.

TypeConcept
DocumentationProvisioned throughput units (PTU) – Microsoft Learn learn.microsoft.com

Episodes about Provisioned throughput units (PTU)

Episodes in which Provisioned throughput units (PTU) is discussed:

Frequently asked questions

Questions marked with AI were written with AI from the facts on this page.

What is a provisioned throughput unit (PTU)? (AI)

A PTU is the unit of measure for provisioned deployments in Microsoft Foundry and Azure OpenAI, where each unit represents a fixed amount of model processing capacity reserved exclusively for the deployment.

How are PTUs billed compared with standard deployments? (AI)

PTUs are billed hourly per unit regardless of tokens consumed, with Azure Reservations available as a discounted commitment option. Standard deployments use pay-as-you-go token billing on shared capacity.

Does having PTU quota guarantee capacity? (AI)

No. The documentation states that quota does not guarantee capacity and that reservations do not guarantee it either, so creating the deployment first is advised.

Glossary

Community wiki

Short info above, community knowledge below: add practical tips, examples or corrections. New contributions are reviewed first; trusted members publish directly.

Experts and companies for this topic

Nobody has listed this topic yet. Add it to your profile.

Sources

Maintained by Mirko Peters (Microsoft MVP) · Last updated: