Microsoft Fabric licensing in 2026 is a capacity-based commercial model built around F-SKU Capacity Units (CUs), ranging from F2 (2 CUs, ~$262/month) to F2048 (2048 CUs, ~$268K/month). Pay-as-you-go capacity is billed per CU-hour with the option to pause and resume; reserved capacity at 1-year commitment unlocks a 41% discount over PAYG. Fabric capacity is the single shared compute resource for Power BI, Data Factory, Data Engineering (Spark), Data Warehouse, Real-Time Intelligence, Data Science, and Copilot in Fabric — meaning right-sizing depends on understanding the aggregate workload pattern, not individual product utilisation. The Power BI Premium P-SKU end-of-sale in October 2024 and the formal retirement during 2026 makes Fabric migration mandatory for existing Power BI Premium customers. The buyer-side strategy: instrument capacity utilisation for 60–90 days, reserve at the 70th-percentile sustained capacity (not peak), use bursting headroom for spikes, and structure F-SKU commitment with the EA renewal cycle so Fabric does not negotiate separately from M365.
Microsoft Fabric F-SKU capacity structure
Microsoft Fabric is licensed as a capacity-based commercial model in 2026. The F-SKU family ranges from F2 (2 Capacity Units / CUs, approximately $262/month PAYG) through F4, F8, F16, F32, F64, F128, F256, F512, F1024, to F2048 (2048 CUs, approximately $268,000/month PAYG). Each F-SKU doubles the CU allocation, with linear pricing scaling. The Capacity Unit is the abstracted unit of compute — one CU roughly corresponds to a defined throughput of mixed-workload compute across Spark, SQL, and Power BI rendering.
The reservation discount: a 1-year reserved-capacity commitment unlocks approximately a 41% discount versus PAYG. There is no 3-year reservation tier for Fabric capacity (in contrast to Azure compute Reserved Instances). The reservation is per-F-SKU per-region, with the customer commitment locked for the 1-year term.
The procurement implication: Fabric capacity pricing is more elastic than the legacy Power BI Premium P-SKU model it replaces. Capacity can be paused (suspending billing entirely) and resumed, scaled up or down on demand, and reserved at 41% discount for predictable baseload. Read the P-to-F migration playbook for the structural transition mechanics.
Shared capacity across the Fabric workloads
The critical conceptual shift from Power BI Premium P-SKU to Fabric F-SKU: capacity is shared across all Fabric workloads. The same F64 capacity pool serves Power BI reports, Data Factory pipelines, Spark notebooks, SQL warehouse queries, Real-Time Intelligence stream processing, Data Science notebooks, and Copilot in Fabric inference. Capacity right-sizing therefore depends on understanding the aggregate workload pattern, not on sizing for any individual product.
This is procurement-relevant: Power BI Premium customers transitioning to Fabric who size their F-SKU against historical Power BI utilisation will systematically undersize, because Power BI was the only workload on Premium. Adding Data Engineering, Warehouse, and Real-Time Intelligence to the same capacity drives utilisation upward. Conversely, Power BI Premium customers without active plans to deploy other Fabric workloads will systematically oversize if they assume Fabric requires substantially more capacity than Premium provided.
The instrumentation discipline: measure aggregate CU consumption across all in-scope Fabric workloads for 60–90 days before committing to a reserved capacity tier. The Fabric Capacity Metrics App provides the necessary telemetry. Commit at the 70th-percentile sustained capacity, not at the 95th-percentile peak.
Power BI Pro and PPU under Fabric
Fabric does not eliminate Power BI Pro ($14/user/month) or Power BI Premium per User (PPU, $24/user/month). These per-user SKUs remain the licensing path for individual report authors and consumers on Fabric content. The mechanics:
- Free Power BI users. Can view content hosted on Fabric capacity if the capacity is F64 or larger (the free-user threshold).
- Power BI Pro users. Required for content authoring on Fabric capacity below F64, and for content sharing in personal workspaces.
- Power BI Premium per User (PPU). Required for individual access to premium-tier Power BI features outside the Fabric capacity context.
The F64-or-larger free-viewer threshold is the most-misunderstood economic lever in Fabric licensing. For organisations with 1,000+ free report consumers, the F64 vs F32 step (~$8,400/month PAYG, ~$4,950/month reserved) frequently pays for itself by eliminating Power BI Pro licensing for the viewer population.
Bursting, smoothing, and the throttling pattern
Fabric capacity supports bursting (consuming more than allocated CUs during short spikes) and smoothing (aggregating CU consumption over a 24-hour window for billing reconciliation). Bursting allows occasional spikes without immediate throttling; smoothing absorbs short bursts into the 24-hour average. Sustained over-consumption beyond the smoothing window triggers throttling: interactive workload slowdown first, then background workload deferral, then ultimately query rejection.
This is operationally relevant because it makes Fabric tolerant of peak/baseload patterns that would have triggered throttling on Power BI Premium P-SKU. The procurement implication: sustained baseload is what to reserve against; transient peaks can be absorbed by bursting headroom. Sizing reserved capacity to peak forecast is the most-common Fabric overcommit pattern in 2026.
Reserved capacity vs PAYG decision
The 1-year reserved capacity discount is approximately 41% — meaning reserved capacity is justified for any sustained utilisation above approximately 59% of the F-SKU tier. For most enterprise Fabric deployments past the 60-day instrumentation window, the sustained utilisation comfortably exceeds 59%, making reservation the economically dominant choice.
The exceptions: workloads with extreme variability (dev/test environments, seasonal businesses with concentrated peak windows, project-based deployments), where pause-and-resume PAYG capacity may deliver better realised economics than 1-year reservation. The pause capability is unique to Fabric (Power BI Premium P-SKU did not offer it) and is one of the cleaner commercial improvements in the migration.
EA negotiation levers for Fabric
- Anchor Fabric commitment to EA renewal cycle. Do not commit to Fabric capacity in a one-off transaction. Roll Fabric into the EA renewal so it negotiates as part of the broader Microsoft deal, not separately.
- Negotiate Fabric capacity as part of MACC drawdown. Fabric capacity consumption can flow against MACC commit in qualifying contracting structures.
- Capture Power BI Premium migration concessions. Existing P-SKU customers retiring Premium to migrate to Fabric should capture concession value in the migration trade.
- F64 free-viewer breakeven. Run the F64 vs (lower-F-SKU + Power BI Pro for viewers) analysis explicitly; the breakeven point is typically around 600–900 viewers.
- Reserve at 70th percentile, not peak. Instrument 60–90 days and reserve at the 70th-percentile sustained CU, using bursting headroom for spikes.
- Multi-region capacity allocation. Fabric reservations are per-region. Multi-region deployments need separate reservations per region; do not consolidate into a single oversized reservation.
- Pause-capable PAYG for dev/test. Use pausable PAYG capacity for environments with predictable downtime windows; do not reserve dev/test.
Anonymised case study: $480K Fabric capacity right-sizing
A 5,400-employee European retailer migrating from Power BI Premium P2 (10x P2 SKUs, ~$120K/month) to Fabric was proposed by their LSP a Fabric F128 reservation in each of two regions (~$160K/month reserved), justified as ‘1.6x P-SKU capacity to support Data Engineering and Warehouse workload expansion.’ We instrumented 90 days of actual aggregate CU consumption across the Power BI plus the early Data Engineering + Warehouse deployments. Outcome: sustained 70th-percentile consumption was 52 CUs in the primary region and 28 CUs in the secondary region. Recommended: F64 reserved primary, F32 reserved secondary, pausable F16 PAYG for dev/test. Total reserved + PAYG: ~$120K/month reserved + ~$3K/month average PAYG. Savings against the proposed reservation: $480K annualised, with bursting headroom validated against 12 months of forward forecast.
Microsoft Fabric licensing is genuinely more elastic and more procurement-friendly than the Power BI Premium model it replaces — but only if the buyer instruments before committing. Pair this analysis with the P-to-F migration playbook, the Power BI Premium retirement options, the Fabric capacity EA leverage, the Azure MACC guide, and the Azure advisory service that runs the Fabric capacity audit and reservation right-sizing.