Sizing and managing Fabric capacity

Capacity units and the F-SKU ladder

Fabric capacity is measured in capacity units (CU) — a normalised unit of compute shared across all Fabric workloads (Power BI, data engineering, warehousing, real-time). You buy an F-SKU, and the number is the CU allocation:

• The ladder runs F2, F4, F8, F16, F32, F64, F128 … up to F2048 — each step roughly doubles CU (and cost).
• F-SKUs are the successor to Power BI Premium P-SKUs (a P1 corresponds roughly to F64) and are the path for new deployments.
• Capacity is bought pay-as-you-go (billed hourly, pausable) or reserved (a one-year commitment at a discount).

The F64 threshold (the sizing hinge)

F64 is not just a size — it's the licensing threshold: on F64+ capacity, Free users can view content (Viewer role) with no per-user Pro licence. Below F64, every viewer still needs Pro. So sizing is entangled with licensing: a broad read audience often justifies jumping to F64 purely to drop per-viewer Pro costs (the licensing concept's break-even), independent of raw compute need.

Pausing and scaling

Pay-as-you-go capacity can be paused (billing stops) — valuable for dev/test capacities outside working hours — and scaled up/down between SKUs as demand changes. Reserved capacity trades that flexibility for a lower rate. Pausing a capacity makes its content unavailable, so it's for non-production or scheduled-downtime scenarios, not the audience's F64.

Failure modes

• Oversized 'to be safe'. Paying for F128 headroom that sits idle — the most common waste.
• Undersized under load. An F-SKU too small for the workload throttles users (the throttling concept) — reports slow or queue.
• F64 bought for compute, ignored for licensing (or vice versa) — the two reasons to sit at F64 get conflated and the decision is never actually made.
• Pausing production capacity — content goes dark; pause is for dev/test only.
• One capacity for everything — a noisy workload starves the rest (the monitoring concept).

What to watch (proofs)

• Capacity inventory: Get-PowerBICapacity and the admin/capacities REST endpoint — every capacity, its SKU/CU size, state (active/paused) and admins.
• Sizing vs usage: the capacity metrics app shows CU consumption against the SKU ceiling — the evidence for up/downsizing.
• F64 eligibility map: which workspaces sit on F64+ (free-viewer eligible) vs below — reconciled with the licence census (Get-MgUserLicenseDetail) for the cost decision.
• Idle/paused capacities: capacities running with low utilisation, or paused capacities still assigned to production workspaces.
• Workspace→capacity assignment: Get-PowerBIWorkspace -Scope Organization mapped to capacities — the cost-attribution table.