Microsoft Fabric Pricing & Capacity Planning Guide

Microsoft Fabric pricing starts at about $262.80/month for the entry-level F2 capacity and climbs the doubling ladder to $8,409.60/month at F64 — the tier where your report viewers stop needing paid Power BI licenses (all figures pay-as-you-go, as of June 2026). Capacity planning is the work of finding the smallest F-SKU that keeps your 24-hour smoothed compute under 100% while clearing that F64 viewer-licensing line if you have enough readers. This guide gives you the full price table, the exact mechanics behind the bill, and a decision flow you can run today to land on a SKU.

Fabric bills one thing for compute: the Capacity Unit (CU). Every workload — Data Factory pipelines, Spark notebooks, Warehouse SQL, semantic-model refreshes, Power BI report renders — draws from the same CU pool of one F-SKU. The list price is fixed at $0.18 per CU-hour (Azure pricing, checked June 2026), and an F-SKU's number is its CU/second baseline: F64 = 64 CUs. So the monthly PAYG cost is mechanical: CUs × $0.18 × 730 hours. Storage is decoupled and billed separately — OneLake works like Azure Data Lake Storage Gen2, pay-as-you-go per GB stored with no transaction charge (OneLake capacity consumption, Microsoft Learn, checked June 2026).

Two tensions decide whether your bill is fair or wasteful, and this guide is organized around them: the F64 licensing cliff (buy a bigger capacity, or buy per-user licenses?) and the reserved-capacity trap (commit for a discount, or stay flexible and pay more?). Get those two right and the SKU number almost picks itself.

The canonical F-SKU price table (June 2026)

This is the single table to bookmark. Every figure is computed from the published rate — PAYG = CUs × $0.18 × 730; reserved = PAYG × 0.5949, the factor Microsoft Learn states as reserved capacity costing "about 59% of pay-as-you-go rates" (Apache Airflow compute in Fabric, checked June 2026), i.e. a 40.51% discount. Prices are USD, as of June 2026, and are list/retail before any enterprise agreement discount.

Reserved figures are estimates rounded from the 0.5949 factor; check the Azure portal reservation quote for your region and subscription, since reservations are region- and subscription-bound. The doubling is exact: every step up the ladder is precisely 2× the CUs and 2× the cost, which makes "do I jump a tier?" a clean yes/no rather than a negotiation. For the mechanics of a single CU, see what a Fabric capacity unit actually is; for a deeper read on the SKU ladder itself, see the F-SKU pricing breakdown.

Free download: the printable Fabric pricing cheat-sheet (PDF) — the full F-SKU table plus the licensing thresholds on one page. No email required.

The F64 licensing cliff — the first thing to size for

The biggest single decision in Fabric pricing is not which SKU runs your pipelines fast enough. It's whether you cross F64. At F64 and above, any user with a free Fabric license can view and interact with Power BI reports hosted on that capacity. Below F64 — F2 through F32 — every viewer must hold a Power BI Pro ($14/user/month) or Premium Per User ($24/user/month) license (Understand Microsoft Fabric Licenses, Microsoft Learn, checked June 2026: "Viewing Power BI content requires a Pro or PPU license when the F SKU is below F64.").

That creates a cliff, not a slope. Run the arithmetic at June-2026 prices:

• An F32 costs $4,204.80/month PAYG. If it serves 250 report viewers, you also pay 250 × $14 Pro = $3,500/month in licenses → $7,704.80/month all-in.
• An F64 costs $8,409.60/month PAYG — and those same 250 viewers cost $0 in Power BI licenses.

So the moment your viewer count pushes the F32-plus-Pro total past the F64 line, the bigger capacity is the cheaper option, and it gives you more compute. The crossover where F64's free viewers beat "smaller SKU + Pro" sits at roughly ceil((F64 − smaller-SKU) ÷ $14 viewers): against an F32, that's ceil((8,409.60 − 4,204.80) ÷ 14) = 301 Pro viewers; against an F2 it's about 582. Below those counts the small SKU plus Pro is cheaper; above them, F64 wins outright. One caveat the cliff doesn't cover: Pro/PPU is only required for Power BI report consumption — building or running non-Power BI Fabric items (lakehouses, warehouses, notebooks) needs an F capacity regardless of viewer licensing, and PPU alone never provisions Fabric compute (Microsoft Learn, checked June 2026). The full crossover math lives in the F64 licensing threshold deep-dive.

How the bill is actually computed: CUs, smoothing, and throttling

The price table tells you the ceiling. Smoothing decides whether you ever hit it. Fabric doesn't bill you the instant a query spikes — it spreads consumption over a rolling window so short, intense jobs don't blow past your SKU. Interactive operations (report renders, DAX queries) are smoothed over 10 minutes; background operations (refreshes, Spark jobs, Warehouse loads) are smoothed over 24 hours (Understand your Fabric capacity throttling, Microsoft Learn, checked June 2026). This is why a tiny F2 can run a heavy nightly pipeline: the load is borrowed against quiet hours later in the day.

When smoothed demand exceeds the SKU's allowance, Fabric doesn't crash — it throttles in stages keyed to future-capacity time windows, not a utilization percentage:

Source: Metrics app calculations, Microsoft Learn (checked June 2026). Note what these thresholds are measured against: the metrics app expresses throttling as a percentage of future capacity — a background-rejection reading of 250% means you consumed 2.5× your daily CU allowance, not that a CPU hit 250%. That distinction matters because it tells you smoothing has already run out of future hours to borrow against.

The practical takeaway for sizing: you are not sizing to peak CPU, you are sizing so your 24-hour smoothed background usage stays under 100% and your 10/60-minute interactive smoothing rarely crosses the line. A capacity that touches 95% smoothed for a few minutes a day is correctly sized; one that sits at 130% smoothed background is throttling users and needs the next tier up — or a more efficient pipeline. Read that from the Capacity Metrics app — it holds 14 days of compute detail, which is your sizing evidence, so capture a representative two-week window before you commit to a SKU.

Reserved vs PAYG — and the trap to avoid

Once the SKU is chosen, you pick a billing model, and this is where money quietly leaks. The two honest options:

• Pay-as-you-go — billed per second the capacity is active; you can pause it when idle. Full list price.
• One-year reservation — about 40.5% cheaper (the 0.5949 factor), but it's a billing commitment, not an on-demand resource. You pay for the reserved baseline whether or not the capacity runs.

The decision is a utilization question. A reservation wins when your capacity runs at a high, steady baseline — roughly above 60% of the hours in the year. If your workload is spiky or only runs in scheduled windows (a nightly transform, business-hours-only BI), PAYG plus disciplined pausing can beat a reservation even without the discount. Compare the two paths in reserved vs PAYG for Fabric.

The named trap: the three-year reservation. In most Azure services, a three-year commitment buys a big step-down over one-year. Fabric is the exception — the three-year discount is only marginally deeper than the one-year, so locking three years of capital and architecture for a sliver of extra savings is structurally a bad trade. Default to the one-year reservation for steady baselines; buy it in smaller increments you can scale down rather than one monolithic block.

There's a second, sharper trap that lives at the seam between these models — the pause trap. Pausing a capacity sums all remaining cumulative overages and smoothed operations and writes them to your Azure bill immediately, at full PAYG rates (Pause and resume your Fabric capacity, Microsoft Learn, checked June 2026). Pausing to clear throttling therefore bills the very compute you smoothed — and on a reserved capacity you pay the reservation and the PAYG overage. We never recommend pausing to escape throttling; size up or optimize the query instead. The full pause-trap teardown lives in our cost pillar, reduce Microsoft Fabric costs.

The tier-selection decision flow

Here is the procedure we run on a buyer's numbers. Follow it in order — each step gates the next.

A worked pass: a team with 300 viewers, a nightly Spark pipeline, and business-hours BI. Step 1 → 300 viewers is just under the 301-viewer crossover vs F32, so the F64-for-licensing call is essentially a coin-flip right now — park it and revisit if even one more viewer joins. Step 2 → yes, they run notebooks, so an F capacity is mandatory. Step 3 → the trial shows smoothed background load peaking at ~70% of an F16. Step 4 → 300 × $14 Pro = $4,200/mo on top of F16's $2,102.40 = $6,302.40 all-in, which is still under F64's $8,409.60 — so stay on F16 + Pro for now, but note that the F32 crossover is essentially already reached (F32 + 301 Pro viewers costs $4,204.80 + 301 × $14 = $8,418.80, just over F64). If the viewer count ticks above ~301 or compute needs push you to F32, jump to F64 instead. Step 5 → the load is windowed, so PAYG with a scheduled pause beats a reservation. Result: F16, PAYG, paused outside business hours, viewers on Pro — with an immediate re-check trigger if viewer count crosses 301. Validate your own numbers against the monthly-bill estimator walkthrough.

What to do next

1. Pull your real smoothed usage. Don't size from a vendor estimate — read 14 days from the Capacity Metrics app and find your true 24-hour smoothed peak.
2. Count your viewers and run the cliff math. If "smaller SKU + Pro" beats F64, stay small; the moment it doesn't, jump.
3. Pick the billing model from utilization, not from the discount headline. Steady → one-year reservation. Spiky → PAYG with a scheduled pause, and never pause to clear throttling debt.
4. Decide single vs split capacity. A single big capacity pools resources but shares a blast radius — one runaway query can throttle everyone. See single vs multiple capacities.
5. Budget storage separately. OneLake at ~$0.023/GB-month is its own line; model it apart from compute in OneLake storage costs.

The two enemies this guide defeats: the F64 licensing cliff, which punishes buyers who under-size into a wall of per-user Pro licenses (or over-size into capacity they don't need); and the reserved-capacity trap, which dresses a multi-year lock-in as savings. SpendWeave's stance is simple — size to your smoothed load, cross F64 only when the viewer math says so, and never trade flexibility for a discount you can't use.

If this is the kind of plain-math teardown you want monthly, subscribe to the SpendWeave Fabric cost teardowns — real capacity bills, dissected. And grab the printable F-SKU pricing cheat-sheet (PDF) — the whole table plus the licensing thresholds on one page, no email required.

Frequently asked questions

How much does Microsoft Fabric cost per month? On pay-as-you-go, an F2 capacity is about $262.80/month and an F64 is about $8,409.60/month (64 CUs × $0.18/CU-hour × 730 hours), as of June 2026. A one-year reservation costs roughly 59% of that — an F64 reserved is about $5,002.87/month — but you lose the ability to pause. OneLake storage is billed separately at about $0.023/GB-month.

What is the F64 licensing threshold in Microsoft Fabric? At F64 and above, anyone with a free Fabric license can view and interact with Power BI reports hosted on that capacity — no per-user Pro or PPU license needed. Below F64 (F2–F32), every report viewer must hold a Pro ($14/user/mo) or PPU ($24/user/mo) license. That makes F64 the breakpoint where buying a bigger capacity can be cheaper than buying viewer licenses.

Is reserved capacity cheaper than pay-as-you-go in Fabric? Yes — a one-year Fabric capacity reservation costs about 59% of the pay-as-you-go rate, roughly a 40.5% discount, per Microsoft Learn. The catch: a reservation is a billing commitment, not a paused-when-idle resource, so it only wins if your baseline utilization stays high. A three-year reservation barely beats the one-year, so it's rarely worth the lock-in.

How do I pick the right Fabric F-SKU? Start with viewer licensing: if you have more than ~64 Power BI report viewers, the free-viewer math at F64 usually beats per-user Pro licenses. Then size to load: run a 60-day F64 trial or a short PAYG burst, read the Capacity Metrics app, and pick the smallest SKU whose smoothed 24-hour CU usage stays under 100%. Only then choose reserved (steady load) vs PAYG-with-pause (spiky, schedulable load).

Does pausing a Fabric capacity save money? It can, but only on pay-as-you-go and only if you have no smoothed background debt outstanding. Pausing settles all cumulative overages and smoothed operations to your Azure bill immediately at full PAYG rates. Pausing a reserved capacity is worse — you keep paying the reservation and get billed the overage on top. We never recommend pausing to clear throttling.

Researched with AI assistance, written and fact-checked by Jonathan Flach, verified against Microsoft Learn.