A finance team uses Power BI Desktop, Microsoft’s free Windows app, to build its reports. Each report pulls in exports from the business software and an Excel budget file. At month end, two reports show two different margins for the same period. And once the reports go to management, there’s a licensing question too.
It takes two decisions to fix this. Where the data and the calculations live decides whether the figures agree. The licence that each author and each reader holds decides what sharing costs you. The prices below are the ones Microsoft published on 14 October 2026, excluding VAT.
The limits of a report built on files
In Power BI Desktop, every report carries its own Power Query steps. Power Query is the tool that you use to clean and combine the data before any calculation runs. If two authors exclude credit notes in different ways, they’ll get two revenue figures. But neither report tells you that the figures disagree.
The source file also makes refreshes fragile. A refresh is when the Power BI service reads the data again on a schedule. For workbooks stored in OneDrive or SharePoint, Power BI follows the file’s internal ID. So if you replace a workbook with a copy, the copy gets a new ID and the refresh stops, as Microsoft’s documentation explains.
In a Pro workspace, which is the shared folder of the Power BI service, each model is limited to eight scheduled refreshes a day and to 1 GB. Microsoft itself advises you to build a data warehouse with periodic loads once Power Query reaches its limits. That’s especially true for large volumes and for keeping a history of changes.
A data warehouse and one shared semantic model
The data warehouse is a database that’s set aside for analysis. Scheduled loads copy sales and budgets into it. Then they compare its totals with the totals in the sources.
On top of the warehouse, one semantic model describes how the tables relate and holds the measures. Measures are the calculation formulas, written in DAX, Power BI’s formula language. Reports connect to that published model instead of carrying their own copy of the data. So when you correct the margin formula, every report changes at once.
The diagram follows the data of a company that runs its warehouse on SQL Server in its own network.
The on-premises data gateway is Windows software that you install on the company network. It only opens outbound connections to Microsoft, so the database isn’t exposed to the internet. Microsoft keeps the gateway for sources that the Power BI service can’t reach directly over the internet. If the warehouse lives in Azure SQL Database, Microsoft’s online database service, you don’t need one.
The warehouse can use the SQL Server that already runs the business software. Azure SQL Database in the Switzerland North region saves you from running a server at all. Its S2 tier, with 250 GB of storage, costs CHF 67.34 a month. Microsoft puts the cheaper S0 and S1 tiers on disks that are meant for testing. And to use a Fabric warehouse, you need a Fabric capacity, which comes up with the licences below.
By default, the model loads a copy of the data into memory. In this mode, called Import, every refresh replaces the copy. DirectQuery mode queries the warehouse every time a chart loads. It suits figures that have to follow the source to the minute, but you’ll wait longer for each report to respond. Direct Lake mode reads tables straight from OneLake, the storage layer of Fabric. So it needs a Fabric capacity too.
The star schema behind the measures
The semantic model gets simpler when the warehouse hands it a star schema. Microsoft separates two kinds of tables in it. Dimension tables describe the things that the business deals with, like customers or products. They’re used to filter and to group. Fact tables record events, like invoice lines. They hold the amounts that you want to add up.
Each filter on a dimension follows a relationship to the fact table. So the margin formula only has to add up amounts. It doesn’t have to fix the structure of the exports.
The history of changes belongs in the warehouse too. When a customer moves region in September, their earlier sales stay in the old region, but only if the customer table keeps one version per period. Microsoft notes that Power Query can’t produce those versions. It takes a process in the warehouse to detect the change and store it. The article on data platforms shows how to build that history with dbt.
Licences for authors and readers
Power BI separates the author, who publishes a report, from the reader, who opens it. In a Pro workspace, both of them need to hold a paid licence as soon as you share the report.
Microsoft offers three per-user licences.
- The free licence, for content that people create for their own use
- Power BI Pro, at CHF 11.30 per user a month, which is included in Microsoft 365 E5
- Premium Per User, at CHF 19.40 a month, with 48 refreshes a day and models over 1 GB
With a Fabric capacity, you rent computing power that’s reserved for your organisation. In the Switzerland North region, Microsoft charges CHF 0.1914 per capacity unit per hour. The smallest capacity is F2. It has two units and costs CHF 279.44 a month with no commitment. If you reserve it for a year, the price drops by 40.5%.
F64 is the size that changes the licence count. Below it, every reader still needs Pro or Premium Per User. From F64 up, it’s enough for readers to have the free licence and the viewer role. Authors still need Pro either way.
But a small capacity also helps you to get past the limit of eight refreshes. On a reserved F2 capacity, each model can take 48 scheduled refreshes a day. Every reader still has to hold a Pro licence. From about twenty people, this combination costs less than Premium Per User for everyone. But F2 doesn’t have much computing power. Microsoft caps the memory of each model at 3 GB on that size, refresh included, so it only works for small models.
Premium Per User doesn’t come with a Fabric capacity. So if you want a Fabric warehouse, you’ll need an F capacity in every case.
The table compares monthly costs in Swiss francs for five authors, by the total number of people. Reserving an F64 capacity comes to CHF 5,316.56 a month.
| People, including five authors | Pro for everyone | Premium Per User for everyone | Reserved F2 and Pro for everyone | Reserved F64 and Pro for the authors |
|---|---|---|---|---|
| 20 | 226.00 | 388.00 | 392.14 | 5,373.06 |
| 50 | 565.00 | 970.00 | 731.14 | 5,373.06 |
| 200 | 2,260.00 | 3,880.00 | 2,426.14 | 5,373.06 |
| 500 | 5,650.00 | 9,700.00 | 5,816.14 | 5,373.06 |
F64 becomes the cheapest option from about 470 readers on Pro and from about 270 readers on Premium Per User. Below those thresholds, per-user licences cost less, even if you have two hundred people.
Data location and the on-site option
Microsoft stores Power BI data in the home region of the organisation’s Microsoft 365 environment, which is set when that environment is created. Switzerland North and Switzerland West both offer Power BI and every Fabric workload, according to the list that Microsoft published on 25 September 2026. You can also put a capacity in a region other than your home region.
If you keep everything on site, there’s Power BI Report Server, a report server that you install on your own premises. It’s licensed through SQL Server core licences or through a reserved F64 capacity. You need a Pro licence to publish Power BI reports to the server, but you don’t need one to read them.
You only need to take three measurements before you choose a licence.
- The size of the models, against the 1 GB limit of Pro
- The number of refreshes needed each day, against the eight that Pro allows
- The number of readers, against the thresholds of about 270 and 470 people
The article Reliable reporting without replacing every tool covers how to define the indicators themselves.
Sources and reference date
- Microsoft | Power BI prices for Switzerland, French page
- Microsoft | Microsoft Fabric pricing
- Microsoft | Azure SQL Database pricing
- Microsoft | Fabric licences and capacities
- Microsoft | Power BI service features by licence type
- Microsoft | semantic model limits by capacity
- Microsoft | data refresh in Power BI
- Microsoft | star schema and Power BI
- Microsoft | table storage modes
- Microsoft | on-premises data gateways
- Microsoft | Azure SQL Database DTU limits
- Microsoft | Fabric region availability
- Microsoft | Fabric home region
- Microsoft | Power BI Report Server
Prices were read on 14 October 2026, in Swiss francs and excluding VAT. Pro and Premium Per User assume an annual commitment. The Fabric and Azure SQL Database prices come from Azure’s public retail price interface for the Switzerland North region, at 730 hours a month. The reader thresholds assume five authors and move with these prices.