Power BI Premium Review: Capacity Licensing and the Break-Even Point

Fix it now

Power BI Premium is two different things. Premium Per User is a per-user licence whose constraint is contagious: everyone opening the content needs it too. Capacity is a block of dedicated compute, now bought as Microsoft Fabric capacity, and from F64 up people with a free licence can view content on it. Only one of those changes what your readers pay.

1. Buy capacity if your reader population is large enough that per-seat licensing costs more, which is arithmetic you can do in ten minutes from your own quote.
2. Buy capacity if you need paginated reports, semantic models above the shared-capacity ceiling, more frequent refresh or the XMLA write endpoint.
3. Buy Premium Per User if a small self-contained analytics team is the whole audience and everyone in it will hold the licence.
4. Skip Premium Per User if the audience is the wider business. Everyone who opens the content needs the same licence, including through the XMLA endpoint, Analyze in Excel and composite models.
5. Skip capacity if nobody will monitor utilisation, because an undersized capacity degrades every report assigned to it at once rather than one.
6. If you hold a Power BI Premium P SKU today, plan the move rather than the renewal. Microsoft removed P SKUs from the purchase flow for new customers on 1 July 2024 and each subscription retires at the end of its agreement term.

Free-licence viewing starts at F64. On F2 through F32 every viewer still needs Pro or Premium Per User, so buying a small capacity for performance does not reduce your seat count.

If that settles it you can stop here. If you want the retirement timetable, the P to F mapping and the break-even method, read on.

Why it happens

Premium Per User is a per-user licence that gives one person most of the capabilities of capacity: semantic models up to 100 GB, up to 48 refreshes a day, paginated reports, deployment pipelines, the XMLA endpoint, incremental refresh, enhanced automatic page refresh and the Power BI AI features. Its constraint is that it is contagious. Microsoft states that a Premium Per User licence is required to access content in a Premium Per User workspace or app, and that this includes access through the XMLA endpoint, Analyze in Excel and composite models. It works for a self-contained analytics team and fails the moment the audience is the wider business.

Capacity is a block of dedicated compute your organisation reserves, with content assigned to it by workspace. From F64 up, people holding only a free licence and a viewer role may view Power BI content hosted there, and that property is what makes capacity a cost decision rather than a features decision. Below that, on F2 through F32, every viewer still needs Pro or Premium Per User, so a small capacity buys performance and changes nothing about your seat count.

The purchasing route has changed and the dates matter. Microsoft removed the Power BI Premium per capacity P SKUs from the purchase flow for new customers on 1 July 2024. Customers outside an Enterprise Agreement could renew until 1 January 2025. New capacity is bought as Microsoft Fabric capacity through Azure, on pay-as-you-go or reservation, and each remaining P SKU subscription retires at the end of its current agreement term. Power BI Pro and Premium Per User are not affected by any of this and are not being retired.

If you are on a P SKU, the mapping is published and simple: P1 becomes F64, P2 becomes F128, P3 becomes F256, P4 becomes F512 and P5 becomes F1024. That is the conversation to have with whoever renews your agreement, and it is worth having early, because the F-SKU billing model and commitment options are not the same shape as the ones you are used to.

What the capacity actually buys, beyond free-licence viewing, is a set of things Pro cannot do at all: substantially larger semantic models, with a single model on an F64 capacity limited to 25 GB and larger sizes above that; more frequent scheduled refresh and incremental refresh patterns that keep large tables current without reloading everything; paginated reports; the XMLA endpoint in read and write mode; deployment pipelines; and dedicated compute so one heavy refresh no longer competes with everybody else’s shared resources.

Two things override the arithmetic entirely. If you need paginated reports, models above the shared-capacity ceiling or the XMLA write endpoint, the decision is made on capability and the cost comparison is beside the point. And if your reader population is genuinely uncertain, remember that Fabric capacity can be paused and resumed, which suits seasonal or project-shaped demand in a way that annual per-seat commitments do not, provided you plan the pause around working hours rather than switching it off casually.

Full reference

The retirement timetable for P SKUs

P SKU to Fabric SKU mapping

Premium Per User against capacity

Those three exclusions are the ones people discover late. Premium Per User is not a cheaper capacity; it is a per-user licence with most of the capacity features and none of the distribution model.

Working out the break-even point

1. Count the people who need to read Power BI content today, and estimate the figure for a year from now. Readers, not authors, drive this.
2. Count the authors and publishers. They need Pro whether or not you buy capacity, so they belong on both sides.
3. Get the monthly figure for F64, which is the smallest capacity granting free-licence viewing, and for a Pro seat, from your own quote.
4. Divide the capacity figure by the seat figure. That is the reader count at which capacity becomes the cheaper option.
5. Add a margin for the administrative time capacity needs, and for the difference between pay-as-you-go and reserved commitments.
6. Compare that number with your reader count and revisit it annually, because the answer moves with your headcount rather than with your data.

Undersizing a capacity affects everyone assigned to it at once. When demand exceeds what the capacity can serve, operations are slowed or delayed rather than failing outright, which presents as reports that feel sluggish for reasons nobody can pinpoint. Watch the utilisation reporting from the first week and size for your busiest hour, not your average one.

If you are on a P SKU now

• Find your agreement end date. That is the deadline, and it is specific to you rather than a single published cut-off.
• Map your SKU to its Fabric equivalent using the published table and price it before the renewal conversation rather than during it.
• Confirm nothing in your estate depends on a capability that behaves differently between the two, particularly around embedding and multi-geo.
• Keep your Pro seats. Authors need them either way, and they are unaffected by the retirement.
• Sovereign cloud customers are treated as an exception in Microsoft’s retirement notice, so confirm your own position rather than assuming the general rule applies.

Questions people ask about this