# Azure AI Foundry & Azure OpenAI — PTU commitment brief

> **Data sourced**: March 2026. Verify current figures at the [Azure Pricing Calculator](https://azure.microsoft.com/en-us/pricing/calculator/).
>
> GitBook Assistant

## Scope note

Azure AI Foundry and Azure OpenAI PTU commitments are unified under a single reservation product: **Microsoft Foundry Provisioned Throughput Reservations**. There is no separate Azure OpenAI reservation — _**a single PTU reservation covers any mix of Foundry Models sold directly by Microsoft, including GPT-4o, o1, DeepSeek, Llama, and others.**_ Reservations are model-independent: you commit to PTU count and deployment type, not to a specific model.

GitBook Assistant

## Coverage summary

PTU reservations apply to provisioned throughput deployments only — Global Provisioned, Data Zone Provisioned, and Regional Provisioned. The hourly billing model (no commitment) is available for short-term or testing scenarios. Term discounts of up to ~64% (1-month) and ~70% (1-year) vs. the hourly rate are available via Azure Reservations. Reservations and deployments are decoupled — purchasing a reservation does not create a deployment and does not guarantee capacity.

GitBook Assistant

**What is covered**: PTU-billed deployments across Global, Data Zone, and Regional Provisioned deployment types. All Foundry Models sold directly by Microsoft. Partial-hour deployments are pro-rated by the minute.

GitBook Assistant

**What is not covered**: Standard token-based pay-as-you-go deployments, fine-tuning hosting, Batch API usage, storage, networking. The legacy Commitment model (PTU-C / Provisioned Classic) is a separate mechanism and is not covered by the new Reservation model. Capacity availability is not guaranteed by purchasing a reservation.

GitBook Assistant

## Azure Hybrid Benefit

> Not applicable. PTUs are a model-independent compute unit with no SQL Server or Windows Server licensing component.
>
> GitBook Assistant

## Deployment type / SKU coverage

| Deployment type          | Reservation | Savings plan | 1-month discount | 1-year discount | Notes                                                                                         |
|--------------------------|-------------|--------------|------------------|------------------|-----------------------------------------------------------------------------------------------|
| Global Provisioned       | ✅ Yes      | ❌ No        | ~64%             | ~70%            | Routes globally. Not interchangeable with Data Zone or Regional reservations.                     |
| Data Zone Provisioned    | ✅ Yes      | ❌ No        | ~64%             | ~70%            | Data stays within a geographic zone (e.g. EU). Reservation must match Data Zone type.       |
| Regional Provisioned     | ✅ Yes      | ❌ No        | ~64%             | ~70%            | Strictest data residency. Reservation scoped to region + deployment type.                     |
| Standard (token PAYG)   | ❌ No       | ❌ No        | —                | —                | No commitment instrument. Batch API offers 50% discount for latency-tolerant workloads.    |
| Legacy Commitment (PTU-C)| ⚠️ Legacy   | ❌ No        | varies           | varies           | Not available to new customers or models post-Aug 2024. Migrate to Hourly/Reservation model. |

## 1-month vs. 1-year reservation

**1-month reservation**: Up to ~64% off the hourly rate. Lowest commitment risk — ideal for workloads that are ramping, models that may change, or where regional capacity availability is uncertain. Renews monthly.

GitBook Assistant

**1-year reservation** _(max savings)_: Up to ~70% off the hourly rate. Best for stable production workloads with a predictable PTU baseline. Because reservations and deployments are decoupled, you can change models freely during the term without re-purchasing. No Compute Savings Plan equivalent exists for PTUs — the Reservation is the only commitment instrument.

## Critical: deploy first, reserve second

`PTU capacity availability is dynamic and not guaranteed by quota.` `Always create your deployments before purchasing a reservation`. `If you buy a reservation for more PTUs than you can deploy due to regional capacity constraints, you are paying for committed units you cannot use. Reservations are not interchangeable across deployment types — Global, Data Zone, and Regional each require a separate reservation purchase.`

GitBook Assistant

## Legacy commitment migration

Customers on the old Commitment model (PTU-C / Provisioned Classic) can continue using it alongside the new Hourly/Reservation model, but it is not available for new customers and does not support models introduced after August 2024. Microsoft recommends migrating to the Reservation model for all new deployments.

* * *

**Sources**

- [Microsoft Learn — Save costs with Microsoft Foundry Provisioned Throughput Reservations](https://learn.microsoft.com/en-us/azure/cost-management-billing/reservations/microsoft-foundry)
- [Microsoft Learn — PTU costs and billing (Microsoft Foundry)](https://learn.microsoft.com/en-us/azure/foundry/openai/how-to/provisioned-throughput-onboarding)
- [Microsoft Learn — What is provisioned throughput?](https://learn.microsoft.com/en-us/azure/ai-services/openai/concepts/provisioned-throughput)
- [Microsoft Tech Community — Unveiling Azure OpenAI PTU reservations and hourly pricing](https://techcommunity.microsoft.com/blog/azure-ai-foundry-blog/unveiling-azure-openai-service-provisioned-reservations-and-hourly-pricing/4214560)

⚠️ Discount percentages (~64% monthly, ~70% annual) are based on GPT-4o Global Provisioned pricing as of January 2025. Actual savings vary by model, region, and deployment type. Always verify with the [Azure Pricing Calculator](https://azure.microsoft.com/en-us/pricing/calculator/).