GUIDE / FABRIC COST MODEL
Microsoft Fabric pricing, explained properly.
Fabric's bill is produced by a small number of mechanisms — capacity units, two separate meters, smoothing, and a licensing threshold. Understand those and you can build a number you can defend. This guide explains the model; your actual rates belong in the scenario calculator with current list prices beside you.
One pool of compute: capacity units and F-SKUs
Fabric sells compute as a capacity — a pooled allocation measured in capacity units (CUs), purchased in sizes named F2, F4, F8 … up the doubling ladder. Every workload in the platform — pipelines, Spark, warehouse queries, real-time, Power BI — draws from that one pool. Price scales roughly linearly with size: each step up the ladder buys twice the CUs for about twice the rate.
The planning consequence: you are not sizing a database and a Spark cluster and a BI server separately. You are sizing one shared pool against the sum of everything that will run on it, including each other's bad days.
Two meters, not one
The capacity covers compute only. OneLake storage bills separately, per GB per month, and it never pauses. Early on it looks like a rounding error; three years of CDC feeds, snapshots and experiment tables later it is a real line item. Model storage with a growth rate, not a constant.
Watch for secondary meters in your own architecture too — cross-region egress and mirroring-related storage can appear depending on design. The honest model lists every meter it includes and names what it excludes.
Pay-as-you-go vs reservation: an operating-hours question
Pay-as-you-go bills by the hour and — crucially — can be paused. A development capacity paused nights and weekends runs a fraction of the hours of one left on 24×7. Reservations commit to a term in exchange for a substantially lower hourly-equivalent rate, but they bill every hour whether or not you use them.
The crossover is pure arithmetic on your real operating schedule, which is exactly why guessing is unnecessary: put your schedule into the calculator and read the answer. Measure first on pay-as-you-go; reserve when the usage pattern has proven itself.
Smoothing: why bursts are fine and sustained load is not
Fabric doesn't bill or throttle on instantaneous usage. Consumption is smoothed over time windows, so a short heavy job can "borrow" headroom and a modest capacity absorbs spiky interactive work gracefully. The flip side: sustained demand above your size leads to throttling and queued jobs — the system protects itself, and your refresh windows pay the price.
Model the shape of your day: when pipelines run, when analysts pile in, what overlaps. A capacity that fits your smoothed profile beats one sized to your scariest five minutes.
The licensing threshold that changes the real total
Compute is only half the bill in BI-heavy shops. Power BI consumption interacts with capacity size: at the larger tiers (the F64 boundary in current terms), report viewers can be covered by the capacity itself, while below it viewer access needs per-user licensing. For an organization with hundreds of report consumers, that interplay can dominate the comparison between two capacity sizes.
Because licensing terms change, treat the principle as permanent and the specifics as something you verify on Microsoft's current licensing pages the week you decide — then put both compute and licensing lines into the same model so the comparison is honest.
Put your own numbers under this model
The free StackSpend Lab calculator builds the scenario with your rates, your schedule and your growth assumptions — every formula visible, nothing uploaded.
Open the scenario calculatorQuick answers
Is OneLake storage included in the capacity?
No — compute and storage are separate meters. Budget storage per GB-month with a growth rate; it accrues even when capacity is paused.
Pay-as-you-go or reservation?
Pay-as-you-go pauses; reservations discount but never sleep. Your real operating hours decide — measure first, commit second.
Why does everyone talk about F64?
It's where viewer licensing economics change in current terms. Model compute and Power BI licensing together and verify the current threshold before deciding.
Will a small capacity survive spiky usage?
Bursts, yes — smoothing absorbs them. Sustained overload, no — jobs throttle. Size to your smoothed daily profile.
Next: Seven Fabric capacity sizing mistakes (and the honest fix) · For a defensible, team-ready model of your environment, see the fixed-price spend model engagement.