Microsoft Fabric Pricing & Licensing

Fabric capacities, SKUs and hidden cost drivers at a glance, so you can make an informed decision on whether and how Microsoft Fabric fits your organisation.

Microsoft Fabric has fundamentally changed the way companies build their data platforms, and with it the pricing model. Organisations that have so far licensed Azure Synapse, Power BI Premium or Azure Data Factory separately now face a consolidation decision: Is Fabric worth it? And what does it really cost?

The short answer: it depends on your usage profile. The long answer is on this page.

1. The Fabric Licensing Model at a Glance

Microsoft Fabric follows a capacity-based model. Instead of buying individual services, you purchase compute capacity, known as Fabric capacities, which can then be used flexibly across all Fabric workloads: Data Engineering, Data Warehouse, Real-Time Analytics, Data Factory, Power BI and more.

This model has one decisive advantage: all workloads share a common resource pool. That makes usage elastic. Lakehouse queries get more compute when Power BI is generating little load, and vice versa.

Important for planning: Microsoft Fabric replaces dedicated Azure services such as Azure Data Lake Storage Gen2, so a separate storage account is no longer strictly required. However, capacity costs do not cover storage: Microsoft OneLake storage is still billed separately per GB consumed and is part of the overall calculation.

For most workloads, the licensing model is based on pre-provisioned capacity. A fully consumption-based model for all workloads has been officially announced but is not yet generally available (see F0 below). For dynamic Spark workloads, the optional on-demand billing for Spark (formerly Autoscale Billing for Spark) is already available. This opt-in model bills Spark jobs separately and precisely by consumption without drawing on the shared Fabric capacity, which makes sense for unpredictable or sporadic Spark loads. There are two purchasing options for the base capacity:

• Fabric Capacity (monthly): A fixed monthly price for a selected SKU tier, cancellable at any time. Because Fabric bills per second, pausing the capacity strategically can reduce the effective monthly price.
• Fabric Capacity (1-year reservation): The same fixed-price logic, but with a 12-month commitment in exchange for a discount of around 41%. Ideal for production environments with stable utilisation.

Free trial: Microsoft offers a 60-day Fabric trial capacity, configured as F4 (4 CUs) or, if eligible, as F64 (64 CUs). It includes 1 TB of OneLake storage and access to all Fabric workloads. Ideal for evaluating real workloads before committing to a production licence.

New since FabCon Europe 2026: F0 and consumption-based billing. At FabCon Europe in Barcelona (late September 2026), Microsoft officially announced the new F0 capacity tier (“Fabric zero-provisioned”). F0 removes the need to provision compute capacity up front: organisations get access to Fabric and Microsoft OneLake and pay only for what they actually use (pay-as-you-go). Usage is billed in CU hours without smoothing and spending limits can be set over a rolling 24-hour window. Microsoft is also extending on-demand billing from Spark to further workloads, so individual billing categories can run on a consumption basis on F2 and larger capacities as well. Important for budgeting: F0 has been officially announced but is not yet generally available (GA). Microsoft has not yet published specific prices. On-demand rates are expected to be calculated as a multiplier on the regular pay-as-you-go rate, varying by workload. Power BI licensing under F0 is also still open; until it is clarified, plan for Power BI Pro licences as with F2 to F32. Our assessment: F0 will be particularly attractive for proofs of concept and sporadic workloads. For permanently utilised environments, reserved F SKUs are likely to remain the more cost-effective option. We will update this article as soon as pricing is confirmed.

2. Fabric Capacity SKUs: F2 to F8192

Capacities are tiered in SKU levels, from F2 (entry level) to F8192 (enterprise scale). Each tier doubles the available computing resources and therefore the parallel processing capacity. One Capacity Unit (CU) corresponds to 1 CU per second of available compute. An F2 therefore delivers 2 CUs per second and provides 1 vCore for batch jobs. Understanding this unit is essential for the job smoothing model.

All prices in EUR, region: West Europe, as of November 2026. A 1-year reservation saves around 41% and requires a 12-month commitment. Current list prices: Official Microsoft pricing page

F64 is the critical licensing threshold: Only from F64 upwards is a Power BI Pro licence no longer required for report consumers. With F2 to F32, all report recipients still need a Power BI Pro licence (~€12 per user per month). From F64 (~€9,227 PAYG / ~€5,487 reserved), users with a free licence can view reports.

At how many users does F64 pay off?

Rule of thumb: as soon as the total cost of Power BI Pro licences exceeds the price difference to F64, switching makes sense. In practice, the break-even point usually lies between around 250 and 500 users, depending on the capacity tier currently in use, administrative effort and growth plans.

3. Monthly Costs in Practice

The capacity licence alone is only part of the actual operating costs. For a realistic calculation, you need to consider the following cost blocks:

• Fabric Capacity: ~€2,307 per month / ~€1,372 per month with reservation (F16, region: West Europe).
• OneLake Storage: Hot ~€0.021/GB; BCDR ~€0.04/GB; Cool/Cold cheaper, per month.
• Data transfer: Variable egress costs depending on region.
• Power BI Pro licences: ~€12 per user per month, mandatory for report consumers with F2 to F32, no longer required from F64.

A company with 50 to 200 data users, 5 to 10 TB of data and regular ETL pipelines typically ends up with total costs of between €2,200 and €4,500 per month.

Microsoft OneLake Storage Tiers in Detail

Since May 2026, Microsoft has distinguished three official storage tiers in OneLake: Hot (standard), Cool (infrequent access) and Cold (archive). Important note: these tiers apply exclusively to file storage, meaning raw files in the Files section of a lakehouse. Lakehouse tables, warehouse tables and semantic models are always billed at the Hot tier price (~€0.021/GB/month), regardless of how rarely they are accessed. This is often where the real hidden storage costs arise.

A frequently underestimated factor: by default, Delta tables retain all versions from the last 30 days. With daily loads, this means up to 30 data versions stored simultaneously, so the actual storage requirement can be a multiple of the underlying data volume. Defining VACUUM strategies and retention policies early on is therefore recommended. Geo-redundant protection (BCDR) costs an additional ~€0.04/GB/month.

4. Hidden Cost Drivers: What to Watch Out For

Spark workloads and compute hunger

Data engineering jobs on Spark can consume enormous amounts of CUs in a short time. Without job monitoring and throttling rules, capacities quickly overflow. Even more critical, however, are Real-Time Intelligence workloads (Eventhouse, KQL databases) and Workspace Monitoring: these run continuously in the background and can generate permanently high CU loads, often unnoticed.

External data sources and integration overhead

For many companies, the starting point is not an existing lakehouse architecture but a classic SQL database or ERP system. Connecting these sources via pipelines or mirroring incurs network costs (egress) and, depending on the integration approach, additional CU load. Consolidating all data in Microsoft OneLake is the goal, but the path there should be planned step by step.

Bursting and smoothing

Fabric allows short-term bursting of up to 12× the purchased capacity, so an F2 can temporarily operate like an F24. This is balanced out by a smoothing model with two windows: background jobs (pipelines, stored procedures, scheduled notebooks) are smoothed over a 24-hour window, while interactive jobs (lakehouse selects, interactive sessions) are smoothed over a 5-minute window. This behaviour can be controlled through capacity settings (throttling thresholds, surge protection), which should be part of the initial setup.

Practical tip: The provisioned compute capacity itself does not incur unexpected additional costs: you pay the fixed SKU price. Storage, mirroring and optional on-demand billing are charged separately. The real risk is throttling or overage: if more than 12 times the purchased capacity is consumed within a 24-hour window, the capacity is fully throttled and no jobs will run until the window resets. Set up alerting and throttling thresholds from day one. The Fabric Capacity Metrics App provides granular per-item reporting and is the central tool for capacity monitoring.

Lack of item governance

An empty workspace without active workloads does not generate any CU costs; compute is only consumed by actual item activity. The governance challenge therefore lies less in the number of workspaces than in the uncontrolled jobs, automations and background processes running within them.

Mirroring storage and free allowance

Fabric provides free mirroring storage equal to the purchased CU count in terabytes (e.g. F16 = 16 TB free). If the capacity is paused, all mirroring storage becomes chargeable immediately, with no grace period. Anyone who pauses capacities overnight should factor in this hidden cost effect in advance when working with large, mirrored datasets.

5. Fabric vs. Azure Synapse: Which Is Cheaper?

The most common question from existing customers: is migrating from Azure Synapse Analytics to Microsoft Fabric worth it?

The calculation often tips in favour of Fabric as soon as Power BI Premium or multiple Azure Data Factory pipelines are in use. In most cases, break-even is reached in under 6 months.

Note on Power BI Premium P-SKUs: Microsoft is gradually retiring the Power BI Premium per-capacity SKUs (P-SKUs). New and existing customers should move to Fabric F-SKUs, which offer the same functionality, are more flexible and can be purchased via Azure. Organisations still running active P-SKUs under an Enterprise Agreement can continue to use them but should plan a migration strategy.

Fabric migration calculator: Answer a few questions about your data platform in 8 short steps and get your estimated migration duration, risks and recommendations in about 2 minutes. Go to the migration calculator →

6. Typical Total Costs by Project Phase

The capacity table alone only answers part of the real budget question. Based on our project experience, three typical phases can be distinguished:

• Proof of concept: F2 to F4, ~€288–577 capacity per month plus 7–10 days of consulting.
• Scaling: F8 to F32, implementation of approx. 20–80 consulting days depending on the use case.
• Ongoing operations: €500–8,000 depending on workload growth, plus 0.5–1 FTE for support.

In the proof-of-concept phase, a small F2 or F4 capacity is sufficient to test the platform with real data sources. In ongoing operations, capacity costs typically range between €500 and €8,000 per month. From a free initial consultation to the migration assessment workshop: see our entry packages at a glance.

Microsoft funding to get started: Through the Frontier Accelerate for Azure programme, Microsoft co-funds assessments, proofs of concept and migration projects. Only specialised Microsoft partners can apply. We check whether your project is eligible and handle the entire application for you.

Check your eligibility now →

7. 6 Levers to Reduce Costs in Microsoft Fabric

• Use capacity scheduling: Pausing a capacity saves money, but only if no jobs are actually running during the pause and the smoothing profile fits. Very effective for batch environments without overnight pipelines. Note: a 1-year reservation is charged regardless of scheduling status.
• Consider reserved instances: For production environments with stable utilisation, annual reservations offer a discount of around 41%. Important: reservation and scheduling are mutually exclusive, so a reservation only pays off for environments that run continuously.
• Optimise Spark session configuration: Oversized Spark clusters are a common cost driver. Define standard profiles for each workload type.
• Delta tables and partitioning: Optimised table formats significantly reduce scan effort and thus CU consumption.
• Establish workspace governance: Clear rules for workspace lifecycle and resource allocation prevent costs from creeping up.
• Set up the Metrics App: The Fabric Capacity Metrics App provides granular per-item reporting, a prerequisite for any optimisation.

8. Conclusion & Next Steps

For many companies, Microsoft Fabric offers a more cost-effective alternative to a fragmented landscape of Azure services, provided the licensing model is sized and operated correctly. The most common mistakes are not made at the point of purchase but during operations: uncontrolled storage growth due to Delta retention, real-time workloads without monitoring and missing throttling configuration.

A well-founded Fabric rollout therefore starts with a clear assessment of your current workloads and cost structures and ends with an operating model that builds cost control in from the outset.

Evaluating Microsoft Fabric? oneLake GmbH is a Microsoft Solutions Partner for Data & AI and supports companies across the DACH region from cost calculation through migration to production operations. We help you find the economically right entry point. See how this works in practice in our Dentsu reference project.

FAQ: Quick Answers

How much does Microsoft Fabric cost per month?

Entry level starts at around €288 per month (F2, monthly). For production workloads, F16 (~€2,307) or F32 (~€4,613) is more realistic. Annual reservations save around 41%.

What is the difference between Fabric and Power BI Premium?

Power BI Premium (P-SKUs) only covers Power BI workloads and is being gradually retired by Microsoft. Microsoft Fabric (F-SKUs) is the official successor: F-SKUs combine Power BI, Data Engineering, Data Warehousing and Real-Time Analytics in a single shared capacity model and can be purchased via Azure.

Can I bill Microsoft Fabric against my Microsoft Azure Consumption Commitment (MACC)?

Yes. Fabric capacity is an Azure SKU and therefore MACC-eligible. Companies with an existing Azure Consumption Commitment can offset Fabric costs directly against it. This simplifies budget planning considerably, as no separate billing item is created.

What hidden costs are there with Microsoft Fabric?

Common cost drivers: uncontrolled real-time workloads (Eventhouse), Delta table retention (default 30 days = up to 30× storage requirement) and missing throttling configuration.

When does an annual reservation pay off?

As soon as a capacity runs continuously, since reservation and scheduling are mutually exclusive. Start with monthly billing, monitor for 4–6 weeks, then decide.