Microsoft Fabric Pricing Calculator
Capacity & Cost Estimator

Microsoft Fabric Capacity Pricing: How CUs and F-SKUs Work

The Capacity Unit is the billing currency for all Fabric workloads. Every action — running a SQL query, executing a notebook, refreshing a dataflow, ingesting an event stream — consumes CUs from the pool assigned to your F-SKU. The pool is shared across all workloads assigned to that capacity.

F-SKUs range from F2 (2 CUs) to F2048 (2,048 CUs). Each step up doubles available compute. The key insight: you don’t need to size for your absolute peak because smoothing and bursting handle short spikes. Size for your 24-hour average load, not the 15-minute maximum. See our Fabric Capacity Optimization guide for the full smoothing mechanics.

The Dataflow Gen2 CU Burn Rate — the Most Misunderstood Cost

Dataflow Gen2 is frequently the largest driver of unexpected Fabric bills. Per Microsoft Learn’s official Dataflow Gen2 pricing page, the consumption rates are:

For a 30-minute ETL job running 10 times per day: in non-CI/CD mode this costs 16 × 1,800 × 10 = 288,000 CU-seconds per day. In CI/CD tiered mode: (12 × 600) + (1.5 × 1,200) = 9,000 CU-seconds per run × 10 = 90,000 CU-seconds per day. That is a 68% reduction in Dataflow CU cost from a single configuration change. See our full Dataflow Gen2 Production Guide for Fast Copy optimization and staging bypass strategies.

OneLake Storage Pricing for Microsoft Fabric

Per Microsoft Learn’s OneLake consumption documentation, OneLake storage follows Azure Data Lake Storage Gen2 hot tier pricing — approximately $0.023/GB/month ($23/TB/month) in US baseline regions. This is billed separately from your F-SKU and scales directly with data volume.

Smoothing — Why Your Peak CU Usage Doesn’t Determine Your SKU

Smoothing is one of the most misunderstood features of Microsoft Fabric pricing. When a workload bursts beyond your SKU’s base CU allocation, Fabric allows it to complete — then spreads the CU debt across a future window rather than throttling immediately. Interactive operations smooth over a minimum of 5 minutes. Background operations (scheduled refreshes, pipelines, warehouse operations) smooth over 24 hours.

This means sizing for the 24-hour average load — not the 5-minute peak — is correct. A 10-minute Spark job that requires 128 CUs on an F64 capacity uses smoothing to complete without throttling. If you consistently exceed your 24-hour average, throttling begins progressively. For how to respond to active throttling, see the Capacity Optimization guide.

F64 — The Financial Pivot Point in Microsoft Fabric Pricing

F64 is the capacity size that changes the licensing math for most enterprise organizations. Two things happen at F64:

• Power BI Pro licences become optional for viewers: Users with appropriate workspace roles can consume Power BI reports without a $14/month Pro licence. For organizations with 350+ report viewers, F64 Reserved ($5,003/month) is nearly always cheaper than paying per-user licences even if the compute requirement alone would be satisfied by F32.
• F64 = Power BI Premium P1 equivalent: F64 on 1-year reserved pricing ($5,003/month US East) is the direct functional replacement for the legacy P1 SKU at the same price point — but includes all Fabric workloads (Data Engineering, Data Factory, Real-Time Intelligence) that P1 never covered.

PAYG vs Reserved — When Each Makes Sense for Fabric Capacity

Per Microsoft Learn’s reservation documentation, a 1-year Reserved commitment saves approximately 41% versus PAYG rates. The decision rule is straightforward:

• Development capacities: PAYG. Pause nights, weekends, and when not actively in use. An F8 PAYG running 8 business hours/day costs roughly one-third of running 24/7.
• Production capacities: Reserved once your baseline load is established (minimum 30 days of PAYG usage data). The 41% saving is substantial — on F64, that’s approximately $3,400/month versus PAYG.
• Burst capacity for known events: PAYG on a temporary larger SKU for year-end processing, marketing campaigns, or Black Friday workloads. Turn on for 3 days, pay for 72 hours.

Regional Pricing: India, Europe, and the Americas

Azure infrastructure costs vary by region. The US East baseline ($0.18/CU-hour PAYG) is the lowest rate. India regions (Central India, South India) carry a ~33% premium at $0.24/CU-hour. Europe ranges from $0.191/CU-hour (UK South) to $0.23/CU-hour (Sweden Central, Brazil South). Switzerland North sits at $0.22/CU-hour.

For organizations with GDPR or data residency requirements that mandate Europe North (Ireland) or Germany West Central, the price premium versus US East is typically 7–17%. This is usually the right trade-off for compliance — the cost of a data residency incident far exceeds the regional premium. For governance architecture that works within these constraints, see our Fabric Production Stability guide.

Copilot in Fabric — What It Actually Costs

Copilot is available on any F-SKU — F64 is not a hard minimum. Per Microsoft Learn’s Copilot consumption documentation, each Copilot request consumes approximately 6.67 CU minutes as a background operation. On an F64 with 64 × 24 = 1,536 CU hours per day, you can run over 13,800 Copilot requests before exhausting capacity.

Microsoft recommends F64+ for production Copilot workloads because smaller SKUs have less headroom — Copilot competes with other workloads for the same CU pool. The recommended architecture for heavy Copilot usage is a dedicated F64 Copilot capacity, separate from your main production capacity, so Copilot consumption doesn’t affect report performance.

Direct Lake — How It Affects Your Microsoft Fabric Pricing Calculations

Direct Lake mode allows Power BI to query OneLake Delta tables without importing data — eliminating refresh cycle costs. When it works correctly, Direct Lake is far more cost-efficient than Import mode (which burns CUs on every refresh) or DirectQuery (which charges CUs on every user query). However, Direct Lake falls back to DirectQuery when semantic models exceed SKU-specific limits or use features not supported in Direct Lake mode.

A Direct Lake fallback turns what would have been an efficient cached query into a per-click CU charge. Monitoring for fallbacks is a cost-control measure, not just a performance one. For how to identify and fix fallback scenarios, see our Direct Lake Fallback guide.

How to Buy Fabric Capacity — Step by Step

1. Provision: In the Azure Portal, search for Microsoft Fabric. Create a new Fabric capacity in your target region. Select your F-SKU. Start with F2–F8 PAYG for initial measurement.
2. Measure: Run representative workloads for 30 days. Monitor daily CU utilization in the Capacity Metrics App. Track average utilization — not peak.
3. Reserve: Once average utilization is stable, purchase a 1-year reservation via the Reservations blade in Azure. The 41% saving applies from the first hour of the reservation period.
4. Assign: Move production workspaces to the reserved capacity in the Fabric Admin Portal. Keep development workspaces on a separate PAYG capacity.
5. Optimize: Use the Metrics App’s per-item operation details to identify the highest-CU operations. Fix workload inefficiencies before scaling up capacity.

Fabric vs Databricks vs Snowflake — Cost Comparison Context

Fabric’s capacity-based pricing is structurally different from both Databricks (DBU + underlying VM costs per workload) and Snowflake (credit-based per-query consumption). Fabric simplifies multi-workload environments into one F-SKU bill covering Power BI, Spark, SQL, pipelines, and real-time analytics simultaneously. For detailed architecture and cost comparisons, see our full analysis: Fabric vs Snowflake.

Methodology — How This Calculator’s Numbers Are Calculated

This calculator is transparent by design. Here is exactly what happens when you click Calculate:

SKU recommendation. Your daily workload inputs — Spark hours, Data Warehouse hours, Dataflow Gen2 runs, and an estimated Power BI interactive load — are converted to Capacity Units per day, then divided by 24 for an hourly average. That average is multiplied by your Peak Load Factor to account for uneven daily distribution.

The tool recommends the smallest F-SKU (F2 through F512) whose CU allocation covers that estimated peak. This mirrors Microsoft’s own guidance to size for sustained load rather than instantaneous peaks. Smoothing absorbs short bursts over a rolling window — a minimum of 5 minutes for interactive operations, 24 hours for background operations — rather than requiring capacity to cover every spike outright.

Microsoft’s own capacity platform evaluates consumption in 30-second windows. This calculator works from your daily and hourly averages rather than modeling that finer-grained window directly.

Compute conversion rates used:

• Spark / Notebooks: 1 Capacity Unit = 2 Spark vCores (per Microsoft Learn)
• Dataflow Gen2, CI/CD tiered mode: 12 CU/second for the first 10 minutes of a run, 1.5 CU/second after
• Dataflow Gen2, non-CI/CD mode: 16 CU/second for the full run duration
• Data Warehouse: active SQL endpoint hours, converted at the same baseline rate as Spark

Storage. Lakehouse and Warehouse storage is billed at $0.023/GB/month (~$23/TB) with no free tier. Mirrored data is free up to 1 TB per Capacity Unit purchased — an F64 capacity includes 64 TB of free mirroring storage — with any volume above that free allocation billed at the same $0.023/GB/month rate.

Pricing. Pay-As-You-Go is calculated at your selected region’s published hourly CU rate × 730 hours/month. Reserved pricing uses Microsoft’s published 1-year reserved rate per SKU, reflecting an approximate 41% discount versus PAYG. Currency conversion uses standard exchange rates at time of publication — not the London closing-spot-rate method Microsoft uses for its own monthly invoicing — so treat currency conversions as directional, not exact.