On-demand billing in Microsoft Fabric

Important

On-demand billing is currently in preview. This information relates to a prerelease product that might be substantially modified before it's released. Microsoft makes no warranties, expressed or implied, with respect to the information provided here.

On-demand billing is a consumption-based billing model for Microsoft Fabric. Operations in an enabled on-demand billing category run outside the capacity's shared compute limit and are billed separately through Azure on-demand billing meters.

You can use on-demand billing in two ways:

  • On a Fabric capacity (F2 or higher) - Enable one or more on-demand billing categories on an existing capacity. The capacity then operates as a hybrid capacity. Operations in enabled categories no longer draw capacity units (CUs) from the shared capacity and are billed separately. Other operations continue to use the shared capacity.
  • On an F0 capacity - Create a capacity with zero provisioned CUs and no upfront capacity cost. All supported on-demand billing categories use on-demand billing.

Benefits of on-demand billing

On-demand billing offers the following benefits:

  • Pay for what you use. Operations are billed through their mapped on-demand billing meters.
  • Set spending controls. Each on-demand billing category has its own spend limit, independent of other categories and the capacity's CU limit.
  • Separate variable demand from provisioned capacity. Operations in an enabled category don't consume the capacity's provisioned CUs.
  • Burst beyond capacity limits. Operations in an enabled category can use compute beyond the provisioned capacity's CU limit without triggering capacity-wide throttling or requiring a capacity upgrade.
  • Start without upfront commitment. The F0 capacity lets you begin building on Fabric with no provisioned capacity cost.

Billing categories

A billing category groups related Fabric operations for on-demand billing. When you enable a category, Fabric bills consumption from that category separately from the capacity price.

The following billing categories are available:

On-demand billing category Included operations
AI data transform AI Functions and AI Services, Dataflow Gen2, notebook and Spark job runs, shortcut transformations, Project Osmos, SQL analytics endpoint and Warehouse queries, and AI Function Evaluation operations.
Application Development User Data Functions and API for GraphQL operations.
Copilot Copilot operations in Fabric.
Cosmos DB database Cosmos DB capacity usage operations.
Data agents AI query operations.
Data Factory Dataflow, data movement, pipeline activity, and Apache Airflow job operations.
VNet data gateway VNet data gateway uptime operations.
Mirroring Extended capabilities for mirroring billing operations.
Data Science Machine learning model endpoint operations.
Data Warehouse Warehouse query, SQL analytics endpoint query, snapshot query, and notebooks running T-SQL operations.
Fabric IQ Ontology modeling, discovery, AI and reasoning, Graph general, ingestion, and serving, and Fabric Planning background and session operations.
OneLake OneLake read, write, security, and diagnostic operations.
Power BI Core Services Full report email subscription, interactive and public API exports, subscription export, parameter processing, and report rendering operations.
Power BI Semantic Models Semantic model refresh, queries, AI Function Evaluation, VNet data gateway, web modeling, and XMLA read and write operations.
Real-Time Intelligence - Eventhouse Eventhouse uptime, KQL Database, and KQL Queryset operations.
Real-Time Intelligence Eventstream, event ingestion and operations, listeners and alerts, anomaly detection, Activator event computations, Investigation agent reasoning, Operations Agent reasoning and compute, and Maps operations.
SQL Database SQL database in Fabric usage operations.

Note

On-demand Billing for Apache Spark is a separate billing scenario with its own configuration and meter. To learn more, see On-demand Billing for Spark.

How on-demand billing works

To manage on-demand billing costs, understand the relationship between CUs and CU-hours.

CU vs. CU-hours

Concept Description Analogy
CU (capacity units) A unit of compute power. Some experiences use CUs to define the maximum concurrent compute available at a given time. Speed
CU-hours The amount of compute consumed over time, calculated as CUs multiplied by hours. Billing uses CU-hours, and spend limits evaluate the CU-hours consumed during the preceding rolling 24-hour window. Fuel

A billing category that uses 512 CUs continuously for 24 hours consumes 12,288 CU-hours (512 × 24). A category that uses 12,288 CUs continuously for one hour consumes the same 12,288 CU-hours.

Billing meters and pricing

Operations in an on-demand billing category map to separate Azure on-demand billing meters. A category can use multiple meters, and each operation-to-meter mapping has a multiplier against the regional pay-as-you-go price per CU-hour.

To calculate the pay-as-you-go price per CU-hour, find the hourly price for a Fabric capacity SKU in your region and currency on the Microsoft Fabric pricing page, and then divide that price by the number of CUs in the SKU. For example, divide the F2 hourly price by 2.

Note

The Azure Retail Prices REST API returns the regional retail rate for an on-demand billing meter. The rate already includes the on-demand billing multiplier. Don't apply the multiplier again.

For operation-to-meter mappings and multipliers, see Understand on-demand billing meters and multipliers.

On-demand consumption isn't smoothed. The service sends reported usage to the billing system as it is reported.

On-demand billing spend limits

Each billing category has a spend limit, which is the CU-hour threshold used to reject new operations based on consumption during the preceding rolling 24-hour window. A spend limit isn't a hard consumption or billing cap. Because the service evaluates usage every five minutes and in-progress operations continue to completion, actual consumption can exceed the configured limit.

Key characteristics of spend limits:

  • 24-hour rolling window: the service evaluates usage every five minutes against the total consumption in the past 24 hours.
  • When the limit is reached: the service rejects new jobs for that category. In-progress jobs continue to completion and are charged.
  • Setting a limit to 0 CU-hours: rejects new operations for that category. In-progress operations continue to completion, and operations that don't currently enforce spend limits can continue to consume CUs.
  • Default value: if you don't configure a spend limit, it defaults to the full quota for that category.
  • Maximum value: the spend limit can't exceed the quota assigned to that category.

On-demand billing quotas

Microsoft manages quotas to protect against fraud and maintain service health. On-demand billing has two quota types:

  • On-demand billing quota - Configured for a billing category, subscription, and region. The configured category limit applies independently to each capacity where the category is enabled and determines the upper bound for the spend limit.
  • On-demand billing-enabled capacity quota - Limits the number of F0 and hybrid capacities in a subscription and region. The default limit is 20.

The default on-demand quotas depend on the region. Due to regional availability, some quotas can default to 0 CU-hours. You can request an increase from Quotas or by creating an Azure support request.

For more information, see View and increase on-demand billing quotas.

F0 capacity

The F0 capacity SKU has no provisioned CUs and no upfront capacity cost. All supported billing categories use on-demand billing.

F0 characteristics include:

  • No upfront capacity cost. You pay for the CUs consumed through on-demand billing.
  • Category-level controls. Each enabled billing category has its own spend limit and quota.
  • Same quotas and limits. The same category quotas and spend limits that apply on paid capacities apply to F0.
  • Pause and resume. You can pause an F0 capacity like any other capacity. When paused, billing categories stop accepting requests.
  • Upgrade anytime. You can upgrade from F0 to a paid SKU (or downgrade from a paid SKU to F0) at any time.

F0 vs. paid capacity with on-demand billing

F0 capacity Paid capacity with on-demand billing categories
Provisioned CUs None Based on SKU (F2 and higher)
On-demand billing categories All supported categories use on-demand billing You enable available categories
Upfront cost None Capacity price
When to use You don't know your capacity requirements yet, or you want pure pay-per-use You have predictable baseline usage but need burst capacity for specific billing categories

Transition behavior

When you change on-demand billing settings or switch SKUs, billing transitions apply to new operations as follows.

Scenario New operations Unreported or smoothed consumption Spend limit window
Enable on-demand billing for a category Billed through on-demand billing Consumption already in smoothing continues to bill to the capacity. Starts fresh
Disable on-demand billing for a category Billed to the capacity Smoothed consumption is billed to the capacity. Not applicable
Switch from a paid SKU to F0 Categories use on-demand billing. Smoothed consumption is reported and charged to the capacity at the time of scaling. Unreported consumption is billed to the capacity. Starts fresh for enabled billing categories
Switch from F0 to a paid SKU On-demand billing is disabled for all categories. New operations use the shared capacity. Unreported consumption is billed to the capacity. Not applicable

Choose the right capacity model

The right model depends on where you are in your Fabric journey and how predictable your Fabric workloads are.

Stage Recommended model Rationale
Evaluating Fabric Free trial, then F0 Start with a trial to learn the platform. When the trial ends, create an F0 capacity.
Early production F0 You don't yet know your capacity requirements. Pay only for what you use while you monitor consumption trends.
Stable, predictable usage Paid capacity (F2+) When you understand your steady-state consumption, upgrade to a paid SKU to benefit from discounted pricing (reserved instances).
Predictable baseline with variable demand Paid capacity with on-demand billing Move operations in enabled billing categories outside the shared capacity limit while other operations continue to use the provisioned capacity.

Example scenario

Contoso has an F64 capacity that supports Data Warehouse and other Fabric operations.

  • Steady state: All workloads run on the F64 paid capacity, within the 64-CU limit.
  • Sales promotion: Data Warehouse demand increases during a promotion. Contoso enables on-demand billing for the Data Warehouse billing category. Other operations continue to draw from the capacity's 64-CU limit.
  • Billing impact: Contoso continues to pay for the F64 capacity. Data Warehouse consumption is billed separately through on-demand billing and appears as a separate line item on the Azure bill.
  • After the promotion: Contoso disables on-demand billing for Data Warehouse. New Data Warehouse operations return to the F64 capacity.

Considerations and limitations

  • On-demand billing isn't available for trial capacities.
  • When you pause a capacity, its on-demand billing categories also pause.
  • You enable on-demand billing for a category at the capacity level, not for individual workspaces. Data Warehouse provides additional workspace-level compute controls.
  • Storage-related charges are billed separately and aren't affected by capacity pause or resume.
  • On-demand billing limits aren't currently enforced for VNet data gateway operations.