Plain-language explanations of the terms used in Licensing Options, Pricing, and the Decision Tree. Every statement was checked against Microsoft's documentation on the date above.
Pro and Premium Per User compared
Power BI Pro already does most of what a finance or operations team asks for. You can publish reports, share them, schedule the data to refresh up to eight times a day, produce print-ready documents such as invoices, and connect Excel to the Power BI data so you can build PivotTables against it (Microsoft calls this Analyze in Excel; you need Build permission on the dataset).
Premium Per User (PPU) adds the following to Pro:
- Bigger data. Each dataset can be up to 100 GB once loaded, instead of 1 GB.
- Fresher data. Up to 48 automatic refreshes a day instead of 8, so roughly every 30 minutes instead of every 3 hours.
- Tools for BI developers. A connection point that lets specialist modelling software such as Tabular Editor or SQL Server Management Studio open and edit the data model (Microsoft calls this the XMLA endpoint), plus separate development, test, and production copies of a workspace (deployment pipelines).
You need PPU when your data is too big for Pro, when your numbers must be fresher than every three hours, or when a BI developer on your team uses those tools. Everyone who opens that team's reports needs PPU too, so the whole audience is priced.
Copilot and how it is billed
Copilot in Power BI lets you ask a report a question in plain English and get an answer with a chart, get a written summary of a report page, ask for a first draft of a report page from a sentence, and get help writing a formula.
There is no per-person Copilot licence for Power BI. Instead, each request uses some of the Fabric capacity you already pay for. Microsoft meters it in tokens, the units its language models read and write. A token is a fragment of a word, and 1,000 tokens is roughly 750 words of English. Microsoft's published rates are 100 capacity-unit-seconds per 1,000 tokens of question and supporting context, 10 per 1,000 tokens of context Copilot has recently seen and can reuse, and 400 per 1,000 tokens of answer.
Microsoft's own illustration assumes a request of 2,000 input tokens and 500 output tokens, which works out to 400 capacity-unit-seconds. On that assumption, and only if nothing else were running, an F2 (two capacity units around the clock, about 2,880 capacity-unit-minutes a day) could absorb roughly 430 such requests a day, and an F64 about 13,800. Real requests vary in size, and reports, refreshes, and everything else on the capacity draw from the same pool, so treat those figures as an illustration of scale rather than a throughput you can count on.
Copilot is classed as background work, so Fabric spreads each request's consumption across the following 24 hours instead of charging it all at once. That smoothing softens bursts. Copilot still uses capacity that reports would otherwise use, and if total usage exceeds the capacity's size for long enough, Fabric slows and then refuses new work on that capacity, as described under "Fabric capacity sizes and billing" below.
Copilot has no separate bill and no per-person licence, but its usage counts against your capacity. Heavy use can mean a larger capacity or, if you turn it on, the optional overage charge.
Lakehouses, warehouses, notebooks, and pipelines
These are the data-engineering side of Fabric, for organizations whose data preparation has outgrown what Power BI does on its own.
- A lakehouse is cloud storage that holds tables of raw and prepared data from many systems.
- A warehouse is the same idea arranged as a SQL database tuned for reporting.
- A notebook is a page where a data engineer writes code to clean and combine data.
- A pipeline is a scheduled job that moves data in.
You want them when you need one central copy of data from several systems, when you want to keep history that the source systems discard, or when your transformations are too heavy for Power BI's built-in query editor. Most small and mid-sized organizations do not need any of this. You probably need them only if you have, or plan to hire, a data engineer.
Paginated reports
Any Power BI report can be printed or exported to PDF on any licence. A paginated report is a different kind of report, built in a separate free tool called Power BI Report Builder, and designed for documents: a fixed layout that flows across pages, with one copy per customer or employee. Invoices, statements, pay slips, and a 300-page detail listing are typical uses. Pro publishes them.
Fabric capacity sizes and billing
You choose a size: F2, F4, F8, F16, F32, F64, and larger. Each is a fixed amount of computing power, billed per second while it runs, at that size's rate. There is no spend range for the capacity itself: an F2 bills as an F2. Storage and the optional overage feature described below bill separately.
You can pause it, which stops the billing, or change its size at any time. A reservation commits to a size for one year in exchange for about 40 percent off, and cannot be paused.
If you ask more of a capacity than its size allows, Fabric first allows short bursts above the capacity's size and spreads the cost over the following minutes or hours. If you are still over, it slows down in stages: after ten minutes' worth of borrowed capacity, new report queries wait 20 seconds; after an hour's worth, they are refused; after a day's worth, everything is refused until the capacity catches up. Nothing extra is billed.
To avoid the slowdown, move up a size, which takes effect immediately, or switch on Microsoft's capacity overage feature, which bills the excess at three times the pay-as-you-go rate instead of slowing down, with an optional daily cap. That feature is in preview and is off unless you turn it on.
Developer tools
Deployment pipelines (development, test, and production copies of a workspace), source control through Git, the XMLA connection for external modelling software, automation through Microsoft's programming interface, and notebooks. These matter when reports are important enough to test before release, or when several people edit the same data model. A small team with one report builder rarely needs them.
DirectQuery
Normally Power BI copies your data into itself on a schedule. With DirectQuery it copies nothing: each time someone opens or clicks a report, Power BI asks your database directly. The data is always current, the 1 GB limit does not apply, and the refresh count does not apply. It works on Pro and on the Free licence.
DirectQuery reports are slower, and every click puts load on your database. The report is only as fast as your database. Microsoft says under five seconds per query is good and over thirty is poor, and a query that takes four minutes times out. Every viewer's clicks reach your server. A single query can return at most a million rows from a cloud source.
Reporting on SQL Server or PostgreSQL running on your own servers
With DirectQuery. Yes, this works for both, through the on-premises data gateway, which is free software installed on a Windows machine on your network. The DirectQuery trade-offs above apply, and your own server carries the query load.
The Fabric equivalent. Copy the data into Fabric storage and report from it with Direct Lake, which reads the stored tables directly at near-Import speed and needs no scheduled refresh. Direct Lake requires an F capacity of any size, with limits by size (an F2 handles a model up to 10 GB on disk). The data has to arrive in Fabric by one of two routes:
- Mirroring, which keeps a live copy in sync automatically. Microsoft supports this for SQL Server 2016 through 2025 running on your own servers, through the gateway. The replication itself is free, and Microsoft includes one terabyte of mirrored storage for every capacity unit you buy. PostgreSQL mirroring is available only for Microsoft's hosted Azure Database for PostgreSQL; PostgreSQL on your own server is not on Microsoft's list.
- A pipeline, which copies the data on a schedule. This works for any database the gateway can reach, including on-premises PostgreSQL.
What it costs. At Canadian list prices, an F2 is CAD 241 a month reserved or CAD 405 pay-as-you-go. Storage beyond the free mirroring allowance is about CAD 0.035 per GB per month. Publishers need Pro, and viewers need Pro as well on any capacity below F64.
Prices on this page are Microsoft list prices as of 22 September 2026, in Canadian dollars for the Canada Central region, and change without notice. Talk to us before relying on a figure.
Sources: Microsoft Learn pages on Analyze in Excel, Copilot consumption and billing, capacity throttling and smoothing, capacity overage, DirectQuery, Direct Lake, mirroring, and SQL Server mirroring, all read 22 September 2026. Links are listed in the licensing reference and pricing reference.