Fabric capacities are on sale: how to think about F SKUs, pausing and Premium

As of today, Microsoft Fabric (preview) capacities can be bought in the Azure portal on a pay-as-you-go basis. We now know how Fabric is billed, and the answer is simpler than most of us feared: you pay for compute you provision and for storage you use.

Cost is the first question every customer asks about a new platform. Here is how I read the model, and what I would do with it now.

One capacity, every workload

A Fabric capacity is the compute for everything in Fabric: Data Factory, Data Engineering, Data Science, Data Warehouse, Real-Time Analytics and Power BI. You do not split it up front between workloads. A single capacity runs them all concurrently and can be shared by many users and projects, with no limit on the number of workspaces or creators.

This is the real shift. Instead of sizing separate compute for each service, each with its own bill, Fabric has one meter for compute.

F SKUs and capacity units

Capacities come as F SKUs from F2 to F2048. The number is the count of Capacity Units (CU), so an F2 has 2 CU and an F64 has 64 CU.

Prices differ by region. Microsoft uses US West 2 as its example: $0.18 per CU per hour, which makes an F2 $0.36 per hour, or $262.80 per month if it runs around the clock (730 hours). An F64 at the same rate is $11.52 per hour. Check your own region in the Azure portal before quoting anything.

Billing is per second with a one-minute minimum. You can scale up and down, and you can pause a capacity when it is not in use.

OneLake is billed separately

The second line on the bill is OneLake storage, charged for the data you store. Microsoft describes the price as comparable to Azure Data Lake Storage, with $0.023 per GB per month in US West 2 as the example. Cross-region data transfer charges can apply.

The region you choose when you provision a capacity also decides where data for its workspaces is stored and where compute runs. For Danish and other EU customers, that makes the region choice a data residency decision, not just a price one. You can provision several capacities in one tenant, for different regions or business units.

The relation to Power BI Premium and Pro

If you already run Power BI Premium, you do not need to buy anything to try Fabric. All Premium capacities can power the Fabric workloads once the Fabric preview switch is turned on. The mapping is direct: a P1 has the same power as an F64, a P2 the same as an F128, and so on.

Microsoft is open that F SKU prices in the portal are higher than the equivalent Premium capacities. The reason is flexibility: pay-as-you-go means no commitment, scaling at will and pausing. An Azure Reservation model for Fabric capacities is promised later this year, which should bring prices in line with a one-year Premium P subscription.

Pro licences still matter, and this is the part I expect to cause the most confusion:

• Power BI report authors need a Pro licence, as with Premium today.
• On capacities smaller than F64, report consumers also need a Pro licence.
• From F64 (the P1 equivalent) and up, report consumers do not need a Pro licence.
• Non-Power BI work, such as pipelines, warehouses, notebooks and capacity management, does not require Pro.

Direct Lake and Dataflows Gen2 are only available on a capacity, Fabric or Premium.

How to think about sizing

Microsoft says plainly that there is no formula yet for estimating the capacity you need. The advice is to put a capacity to use and measure the load. A planning calculator is promised for later this year.

There is a useful window for that measuring. Until August 1, Fabric workloads other than Power BI do not count against the limits of a purchased capacity, and the capacity metrics app shows preview usage separately from paid usage. You can see what your workload would consume without paying for it.

My suggested approach:

1. Start with the Fabric trial if your tenant allows it, or buy a small F SKU to test.
2. Run a realistic slice of your workload, not a demo: one real ingestion, one real transformation, one real report.
3. Watch the metrics app and note the peaks, not just the average.
4. Pause the capacity outside working hours, and decide if your pattern allows that in production.
5. Only then compare the pay-as-you-go cost with an existing or planned Premium P1.

Pausing is the lever most people will underestimate. A development capacity running only in working hours costs a fraction of one running 730 hours a month. A capacity serving readers around the clock cannot be paused.

What this means for you

For small teams, the F2 entry point is the news. Microsoft positions it as under $9 a day in the US example, with a clear warning not to expect it to run large jobs or serve many users. For a department or a proof of concept, that is a low barrier.

For Premium customers, nothing changes on the bill today. Turn on the preview switch, measure, and wait for reservation pricing before planning any move to F SKUs.

For anyone below F64, remember the Pro licences for report viewers. The capacity price alone is not the full cost of a Power BI rollout.

Takeaway

After two decades in data, I have seen many pricing models that looked simple until the first invoice. This one is simple: compute plus storage, sized in CUs, billed per second, pausable. The hard part is sizing, and the free preview window is the time to do it.

What workload would you put on a test capacity first, to learn what your F SKU really needs to be?

Sources

• Announcing Microsoft Fabric capacities are available for purchase (Microsoft Fabric blog)
• Microsoft Fabric licenses and capacity (Microsoft Learn)