Microsoft Fabric pricing looks simple on the first page of the quote: pick a capacity size, pay by the hour, done. The trouble starts in month two, when the bill shows OneLake storage, Power BI seats, a few hours of overage, and a capacity that ran all weekend because nobody paused it.
Most of those charges are predictable once you know where they come from. The rest come from two choices teams make early and rarely revisit: how big the capacity is, and whether it ever turns off.
Microsoft publishes every one of those line items, though they’re spread across a pricing page, licensing docs, and a handful of footnotes. We’ve pulled them into one place, with every rate taken from Microsoft’s own pages so you can check the math against your region.
Microsoft Fabric pricing: What each F-SKU costs
Microsoft Fabric costs $0.18 per CU hour on pay-as-you-go in East US 2, which puts an F2 (2 CUs) at $262.80 a month and an F64 at $8,409.60 if they never pause.
Fabric bills on three layers: the capacity (an F SKU compute pool every workload draws from), OneLake storage billed per GB on its own, and Power BI licenses for the people who build and, below F64, view reports. Capacity is usually the biggest number, but storage and licenses decide whether your estimate holds up.
The F SKUs most teams consider run from F2 to F2048. Monthly figures assume 730 hours, matching the Fabric pricing page for East US 2 as of October 2026.
Two larger SKUs exist too: F4096 at $538,214.40 a month and F8192 at $1,076,428.80 on pay-as-you-go. Price scales in a straight line, so every step up doubles the bill.
Rates differ by region. East US, West US 2, and Central US match East US 2 at $0.18 per CU hour, while West Europe works out to $0.22, about 22% more. Set your region in the Azure pricing calculator before you compare quotes.
Key characteristics of Fabric billing
- One compute pool: Pipelines, notebooks, warehouse queries, and Power BI reports all draw from the same capacity units. There’s no separate meter per engine.
- Per-second billing: Azure bills F capacities per second with no commitment and a one-minute minimum.
- Pause and resize: You can pause an F SKU or move it up or down a size. Pausing stops compute charges, but storage keeps billing.
- Regional rates: An F2 costs $262.80 a month in East US 2 and $321.20 in West Europe, about 22% more.
The F64 threshold: Fabric capacity vs. Power BI per-user licenses
F64 is where Fabric’s cost structure changes. On any F SKU smaller than F64, each person viewing Power BI content needs a Pro, PPU, or individual trial license. On F64 and above, people with only a free Fabric license and a viewer role can open reports.
People who build and share Power BI content need a paid license at every capacity size. Microsoft lists Power BI Pro at $14 and PPU at $24 per user per month, paid yearly. Pro is also included in Microsoft 365 E5 and Office 365 E5, so some of your users may already have it.
That makes the jump from F32 to F64 a licensing decision as much as a compute one. For report viewers, the math flips at these counts:
Past those counts, F64 costs less than buying Pro seats for everyone who only reads reports. Below them, a smaller capacity plus Pro licenses usually wins, as long as the smaller SKU has enough compute for your workloads. Viewers who already hold Pro through Microsoft 365 E5 don’t count toward the break-even.
F64 reservations carry one more perk for teams that still report on premises. Power BI Report Server rights come with Fabric F64+ reserved instances, up to the number of cores included in the capacity. The other route is SQL Server core licensing, so check whether you already own those rights before counting this perk.
OneLake storage pricing
OneLake storage bills per GB per month at pay-as-you-go rates, separate from capacity. Microsoft prorates it through the month, so 1 TB kept all month bills as roughly 33 GB a day.
A few rules sit behind those rates:
- Cache is its own meter: It covers KQL cache storage and Data Activator retained data, and it isn’t affected by tiering or BCDR.
- Power BI storage is mostly free: Data in OneLake for Power BI import semantic models is included in Power BI licensing.
- Mirroring comes with free storage: Each capacity gets 1 TB of free mirroring storage per CU, so an F64 gets 64 TB for mirrored replicas. The free trial doesn’t include it.
- Cool and cold tiers cost compute to read: Both charge a per-GB retrieval fee in capacity units, 200 CU seconds per GB for cool and 600 for cold.
Reads and writes to OneLake also consume capacity units, so a busy lakehouse costs storage dollars and compute. Shortcuts change who pays: transactions count against the capacity where the shortcut lives, while the capacity that holds the data pays for storage.
Pay-as-you-go vs. reserved capacity: The break-even point
A reservation is cheaper per hour, but you pay for it whether the capacity runs or not. Pay-as-you-go costs more per hour and stops billing compute when you pause.
Microsoft applies the reservation discount hourly. If no workload uses those CUs in a given hour, the discount for that hour is gone and doesn’t carry over. Reservations are bought in single-CU increments for a region, so a 64-CU reservation under an F128 discounts half the capacity and bills the other half at pay-as-you-go.
The break-even is the same at every size. An F64 reservation costs $5,002.67 a month, and F64 pay-as-you-go costs $11.52 an hour. Run it more than about 434 hours a month (roughly 14 hours a day, every day), and the reservation wins. Run it 12 hours on weekdays only and pay-as-you-go with a pause schedule costs less.
Reservations don’t renew unless you turn on automatic renewal, and cancellations across a billing scope are capped at $50,000 in a rolling 12-month window.
How smoothing and throttling shape what you pay
Fabric smooths usage over time, so a brief spike past your capacity costs nothing extra. Interactive operations (a user opening a report) are spread over at least five minutes, and background operations (a scheduled refresh or pipeline) over 24 hours.
That smoothing lets a small capacity absorb occasional heavy jobs. When smoothed usage keeps running ahead of your capacity, Fabric throttles jobs in stages (your bill stays the same, and your users wait):
- Up to 10 minutes of future capacity used: No throttling.
- 10 to 60 minutes: Interactive jobs are delayed by 20 seconds.
- 60 minutes to 24 hours: New interactive jobs are rejected.
- Over 24 hours: All new requests are rejected until the capacity catches up.
Throttling is how an undersized capacity "saves" money. Users feel it as slow or failed reports, which is usually the moment someone asks for a bigger SKU.
What Fabric costs per month: two scenarios
These two sample bills use East US 2 rates. The usage figures are assumptions for illustration, so swap in your own numbers.
Scenario A: F16 on pay-as-you-go, paused outside working hours
A small analytics group runs Fabric during business hours. Eight people build reports, and 30 only view them.
Leave the same F16 running 24/7, and the total climbs to about $2,658. A reservation brings the always-on version down to about $1,806, which is still about $490 more than the pause schedule.
Pausing has a catch, though. Reports on a paused capacity aren’t available until it resumes, and any smoothed usage still owed gets added to your bill at the moment you pause.
Scenario B: F64 reserved, running 24/7
A larger organization runs scheduled pipelines overnight and serves Power BI reports to 400 people across time zones. Twenty people build content.
The same setup on pay-as-you-go costs about $8,925. Dropping to an F32 reservation looks cheaper until you add Pro licenses for all 420 users: about $8,617 a month, with half the compute.
In both scenarios, the schedule and the viewer count moved the bill far more than storage did. Small teams save the most by pausing, while large reader audiences come out ahead by crossing F64 and reserving.
Hidden costs on a Fabric bill
The capacity line gets all the attention in a quote. These are the charges that show up later.
Capacity overage at three times the pay-as-you-go rate
Capacity overage, which Microsoft turns on by default for new capacities, pays for usage beyond your SKU so Fabric doesn’t throttle. It bills at three times the pay-as-you-go rate, which works out to $0.54 per CU hour in East US 2.
Reservation discounts don’t apply to overage, even on a reserved capacity. You set a rolling 24-hour cap, but Fabric checks it every five minutes, so the bill can go slightly past your limit.
On-demand Spark with no reservation discount
Autoscale Billing for Spark moves Spark jobs off your shared capacity onto serverless compute billed only while jobs run. That keeps heavy notebooks from starving Power BI.
The tradeoff is price. These jobs bill at standard regional pay-as-you-go CU rates, and reserved capacity discounts don’t apply. You still need a base capacity for everything else.
Storage for deleted workspaces
Deleting a workspace doesn’t stop its storage charge right away. You keep paying for OneLake storage of a deleted workspace during its retention period, which admins can set from 7 to 90 days (seven is the default).
Soft-deleted files and items work the same way. They bill at the same rate as active data until you permanently delete them or the retention period runs out.
BCDR storage and writes
Turning on disaster recovery geo-replicates OneLake data, and hot storage jumps from $0.023 to $0.0414 per GB, about 80% more. Writes cost more compute too: a hot-tier BCDR write uses 3,056 CU seconds per 10,000 operations versus 1,626 without it.
BCDR also only covers data in OneLake. The Azure databases and storage accounts feeding Fabric need their own backups, which we cover below.
Mirroring storage when the capacity is paused
Mirrored replicas are free up to your SKU’s limit, but only while the capacity runs. OneLake starts billing that storage when the capacity is paused, so a nightly pause schedule can quietly turn free storage into a monthly charge.
Spark operations now counted as OneLake usage
Starting October 15, 2026, Fabric includes certain Spark operations in OneLake CU consumption that it didn’t report before. Rates aren’t changing, but Spark-heavy workloads may see higher capacity usage, which can push a capacity closer to throttling.
Networking charges on the way
Networking billing is coming, with at least 90 days’ notice before it starts. It will cover data transfer based on the source and destination of each storage access, so cross-region and multi-cloud shortcuts are worth tracking now.
How to size Fabric capacity and keep costs down
These habits keep a Fabric bill close to the estimate:
- Start small and measure: Run real workloads for a few weeks (a Fabric trial works if your organization can still start one), then read the Capacity Metrics app to see actual CU usage, throttling, and storage per workspace.
- Use the SKU estimator for a first guess: Microsoft’s Fabric SKU Estimator (in preview) recommends a size from data volume, daily batch cycles, and table counts. Treat it as a starting point.
- Pause on a schedule: For pay-as-you-go capacities used during working hours, pausing nights and weekends can cut compute by more than half. Pausing also ends active throttling right away, though it bills any carried-forward usage at that moment.
- Reschedule peaks before you size for them: Buying an F128 for one month-end refresh means paying F128 rates through 20 quiet days. Spread heavy background jobs across the day so smoothing can absorb them.
- Reserve only the steady baseline: Reserve the CUs you use around the clock and let bursts run at pay-as-you-go, or through overage with a tight cap (remember it bills at three times the pay-as-you-go rate).
- Split capacities by team: Separate capacities for engineering and reporting make it easier to attribute spend and keep one team’s jobs from throttling another’s, which is where cloud cost allocation begins.
- Move cold data down a tier: Archive tables you rarely read to cool or cold storage, but keep frequently queried data hot, since retrieval from those tiers costs capacity units.
Three sizing mistakes undo those savings. The first is sizing from data volume alone, which misses concurrency: 300 people opening reports at 9 a.m. can stress a capacity more than a large table. The second is sizing straight from trial results, since a trial rarely carries production users.
The third is reserving before you’ve seen two or three months of usage, which locks in the wrong size for a year or more.
Several Azure cost optimization tools track Fabric alongside the rest of your Azure spend.
Should you pay for Fabric capacity?
For teams that already run Power BI on Azure, Fabric is usually the simplest way to buy analytics. The questions below help you pick a starting point.
Fabric capacity makes sense if you:
- Already need an F32 for your workloads and serve Power BI reports to more than roughly 179 (reserved) to 300 (pay-as-you-go) viewers, where F64 costs less than Pro seats.
- Want engineering, warehousing, and reporting under one bill.
- Run predictable daytime workloads you can pause, or steady 24/7 loads worth reserving.
Look elsewhere or start smaller if you:
- Only need dashboards for a few dozen people (Power BI Pro alone may cover it).
- Run rare, spiky jobs that would leave a capacity idle most of the month.
- Can’t estimate your workloads yet. Capacity units don’t map neatly to reports or pipelines, so start on a small SKU and measure before you commit.
How to estimate your Microsoft Fabric bill in 5 steps
- List your workloads: Note which experiences you’ll use (pipelines, Spark, warehouse, real-time, Power BI), how many daily batch cycles you run, and your compressed data size.
- Count your Power BI users: Separate people who build content from people who only view it. Check how many already have Pro through Microsoft 365 E5.
- Pick a starting SKU: Run the SKU Estimator. If you land on F32, compare F64 against Pro seats using the break-even above.
- Price storage separately: Estimate OneLake GB by tier, decide on BCDR, and check your mirroring allowance.
- Model two schedules: Price the capacity paused outside working hours and reserved 24/7. Comparing the two shows how much the schedule alone moves your bill.
Protecting the data that feeds Fabric
A Fabric capacity pays for analytics. Protecting the systems that feed it is a separate job: the SQL Server databases on Azure VMs, the VMs themselves, and the Blob Storage accounts where raw data lands. If one of those sources is corrupted or deleted, every report built on it inherits the damage.
That’s a posture problem first. Eon brings Cloud Backup Posture Management (CBPM), the category we created, to your Azure environment. It discovers resources continuously, classifies each by data type and environment, and applies backup policy automatically, so a new storage account doesn’t sit unprotected until someone tags it.
Backups land in logically air-gapped, immutable vaults isolated from production. When something breaks, you restore the single file or the exact database records you need, with no full-environment rebuild.
Compared with native cloud backup, Eon reduces backup storage costs by 30% to 50% through global, vault-side deduplication, and pricing is usage-based per GB protected. The same workloads cost more to protect at native Azure Backup pricing.
NETGEAR saw this directly: moving a 10 TB SQL Server database to Eon cut its backup storage cost 35% and shortened recovery from 24 hours to under three.
Backup data can also cut a pipeline out of your Fabric bill. Teams can surface Eon backups in Fabric through read-only OneLake shortcuts, activating backup data in Microsoft Fabric without building another pipeline to copy historical records into OneLake.
The bottom line on Fabric costs
Microsoft Fabric pricing rewards teams that make two decisions on purpose: how big the capacity is, and when it runs. Pause it when nobody needs it, cross F64 only when your viewer count pays for it, and reserve only the baseline you’d run anyway.
Every report in Fabric depends on the Azure databases and storage accounts underneath it, and if you can’t prove those sources are protected, you don’t have coverage.
Do you know which Azure databases and storage accounts feeding your analytics are actually protected? Book a demo to see your backup posture across every subscription.
Frequently asked questions
Is Microsoft Fabric free or paid?
Microsoft Fabric is paid, with a free trial. The Fabric trial gives you an F4 or F64 capacity and 1 TB of OneLake storage for 60 days. Afterward, you have seven days to move workspaces to a paid capacity before non-Power BI items are lost.
Is Microsoft Fabric part of Microsoft 365?
No. Fabric F capacity is an Azure resource, bought and billed through your Azure subscription. Microsoft 365 E5 does include Power BI Pro, which covers the per-user license Fabric needs for Power BI creators and, below F64, viewers.
How much does an F64 capacity cost per month?
An F64 capacity costs $8,409.60 a month on pay-as-you-go in East US 2 if it runs 24/7, or $5,002.67 a month with a reservation. That’s $11.52 an hour on pay-as-you-go, before OneLake storage and Power BI licenses for content creators.
Does a paused Fabric capacity still cost money?
Yes. Pausing stops compute charges, but OneLake storage keeps billing per GB and mirrored storage loses its free allowance. Leftover smoothed usage lands on that month’s bill, and a reservation keeps costing the same while paused.
Do you need Power BI Pro with Microsoft Fabric?
Yes, for anyone who creates and shares Power BI content. Viewers also need Pro on capacities smaller than F64. On F64 and above, viewers with a free license and a viewer role can open reports.
What is the cheapest Microsoft Fabric capacity?
The cheapest Microsoft Fabric capacity is the F2, at $0.36 an hour or $262.80 a month on pay-as-you-go in East US 2, and $156.33 a month reserved. It suits development, testing, and small workloads, and everyone viewing Power BI content on it needs a Pro license.




