Power BI is one of the easier Microsoft products to start using and one of the harder ones to budget for. A single person can build a working dashboard in an afternoon at no cost. The moment that person wants a colleague to look at it, the bill starts, and the shape of that bill depends on decisions most buyers do not know they are making.
We quote no prices here, because Microsoft has changed them and will change them again. Current figures live on the Power BI pricing page and the Microsoft Fabric pricing page. Everything below is the mechanism, which is more stable and more useful when you are planning.
The three per-user licenses, and the sharing rule
What this costs you: getting this wrong is the difference between one paid seat and a paid seat for every person who opens a report.
There are three per-user licenses, documented in Understand Microsoft Fabric licenses.
Power BI Free lets one person connect to data, build reports, and view them in their own personal workspace. It does not let them share. Microsoft states it plainly in Power BI service features by license type: users with free licenses "can't use the Power BI sharing or collaborating features with others."
This is the most common surprise. Power BI Desktop, the authoring tool, is a free download, so the pilot looks free. Distributing the result to eight colleagues is a separate purchase.
Power BI Pro is the working seat. It lets a person publish content into shared workspaces, share it, and view what other Pro users publish. Both sides need it: the author to publish, and each viewer to open the report, unless capacity licensing changes that rule. For a company of any size, Pro seats for everyone who touches a report is the default starting shape.
Power BI Premium Per User includes everything Pro does, plus most features that used to require a large capacity purchase. Two matter more than the rest: a semantic model (the data behind a report) can reach 100 GB instead of the default 1 GB, and it can refresh up to 48 times a day instead of eight. Microsoft's Premium Per User documentation has the full comparison.
The catch is symmetrical. Content in a Premium Per User workspace can only be opened by other Premium Per User license holders. If four analysts need the bigger models and 30 people need to read the output, you are either licensing all 34 at the higher tier or splitting the content across workspaces.
| Free | Pro | Premium Per User | |
|---|---|---|---|
| Build reports for yourself | Yes | Yes | Yes |
| Publish and share | No | Yes | Yes |
| Who can view what you publish | Nobody | Pro and Premium Per User holders | Premium Per User holders only |
| Scheduled refreshes per day | 8 | 8 | 48 |
| Data model size limit | 1 GB | 1 GB | 100 GB |
What your Microsoft 365 plan includes
What this costs you: assuming Power BI Pro is bundled, and finding at rollout that it is not, is a mid-project budget change.
Microsoft 365 E5 includes a Power BI Pro license for each user, and Microsoft's licensing guide for organizations says so directly. The E5 family is the exception rather than the rule. Business Basic, Business Standard, Business Premium, E1, and E3 do not include it, and most small and midsize companies are on one of those.
Before planning around a bundle, check the subscription itself. In the Microsoft 365 admin center, go to Billing, then Licenses, and look for Power BI Pro as its own line.
Capacity licensing: renting compute instead of buying seats
What this costs you: above a certain number of viewers, per-seat licensing stops being the cheaper answer, and the switch is not automatic. You have to make it.
Per-user licensing is a seat count. Capacity licensing works differently. You rent a fixed amount of computing power, measured in capacity units, and everything your organization runs on that capacity draws from the same pool. The number of people using it does not change the price. Running out of headroom does.
Microsoft sells capacity in fixed sizes named F2, F4, F8, F16, F32, F64, F128, and upward to F8192. The number is the capacity units, so F64 has 64 of them. You buy the size in the Azure portal or through a partner and assign workspaces to it.
Power BI Premium per capacity, the older P1 through P5 sizes, is being retired for new purchases, and Microsoft's guidance is to buy Fabric capacity instead. Enterprise Agreement customers can generally run out their term, but this is not a platform to start on.
The F64 rule that decides whether you pay per viewer
This rule determines the whole cost shape, so it is worth stating precisely.
On a capacity of F64 or larger, a person with only a Free license, holding the Viewer role on the workspace, can open Power BI content hosted there. Microsoft's wording: "To view Power BI content with a Microsoft Fabric free per user license, your capacity must reside on an F64 or larger SKU, and you need to have a viewer role on the workspace." ("SKU" there is Microsoft's word for a purchasable size.)
Below F64, that does not apply. On F2 through F32, every person who views Power BI content needs a Pro or Premium Per User license, exactly as they would with no capacity at all.
Two consequences follow. Buying a small capacity does not reduce your seat count; it buys performance and Fabric features, and nothing else. And even on F64, anyone who creates or edits content outside their own personal workspace still needs a paid seat. The free-viewer rule covers viewing only.
Where the crossover sits
Take the monthly cost of an F64 from the Fabric pricing page and divide it by the monthly cost of one Power BI Pro seat. That quotient is roughly the viewer count at which capacity becomes the cheaper answer, and it typically lands in the hundreds. Add back the Pro seats you still need for authors, and subtract the discount if you commit to a reservation. Use today's numbers rather than a figure quoted last year, because both sides of the ratio have moved.
Power BI Embedded
Reports inside a product you sell to your own customers follow a separate path, covered in capacity and SKUs in Power BI embedded analytics. When your application owns the data and signs in on its users' behalf, your end customers need no Power BI licenses. When they sign in with their own Microsoft Entra accounts, the F64 rule applies as it does everywhere else.
Pause, resume, and the commitment question
What this costs you: if reporting only matters during business hours, you may be paying for nights and weekends you never use. Or you may have signed a commitment that makes pausing pointless.
Fabric capacity bought through Azure is billed by the second with no commitment, and you can pause and resume it from the Azure portal or on a schedule. While paused, compute billing stops and content becomes unavailable. Microsoft's warning is worth repeating: "Pausing a capacity can prevent Microsoft Fabric content from being available." A paused capacity is an offline reporting system, so schedule around the people who open reports, including anyone in another time zone.
The alternative is a capacity reservation, a commitment of one year or three years in exchange for a lower rate. The discount applies hourly, and Microsoft states that unused reserved hours do not carry forward, so pausing a reserved capacity saves nothing. You have already committed to that hour.
The two are alternatives rather than a combination. Pay-as-you-go plus scheduled pausing suits a business whose reporting is genuinely idle overnight and on weekends. A reservation suits one that runs continuously, or close to it. One further consequence: Power BI Report Server rights attach to F64 or larger reserved instances, not to pay-as-you-go capacity.
Microsoft Fabric in plain terms
What this costs you: Fabric is frequently proposed for reporting problems that Power BI alone solves, and buying it early adds cost and administration without changing what anyone sees.
Power BI is the reporting and dashboard layer. Microsoft Fabric is the wider analytics platform Power BI now sits inside. Alongside Power BI it includes pipelines for moving and cleaning data, a data warehouse, tooling for large datasets and data science, real-time analytics for streaming sources such as sensors or logs, and OneLake, a single storage layer they all share.
The answer for most small and midsize businesses is that you do not need Fabric on day one. If your reporting draws on a handful of systems, refreshes on a schedule, and produces dashboards people read in the morning, Power BI on Pro licenses does that. Fabric earns its place when one of these is true:
- Data has to be combined and reshaped from several systems before it is reportable, and that work currently lives in spreadsheets or a fragile set of scripts.
- Refreshes have grown long or unreliable because reports are doing work a data warehouse should do.
- You need history the source systems do not keep, such as a daily snapshot of inventory or pipeline.
- You have real-time data, and yesterday's numbers are too late.
Adding Fabric does not replace Pro seats for your authors, and it does not change the viewer rules above.
Capacity units, bursting, throttling, and outgrowing what you bought
What this costs you: capacity is the one purchase where buying too small shows up as visible slowness for staff rather than a line on an invoice.
A capacity unit is a unit of compute. Every report query, every refresh, every pipeline run draws from the pool you bought. Microsoft's throttling documentation describes behavior more forgiving than a hard cap.
Bursting lets an operation temporarily use more compute than you bought, so a heavy job finishes quickly rather than crawling. Smoothing then spreads that cost across time so the spike is not penalized: interactive work such as clicking through a report is smoothed over minutes, and background work such as an overnight refresh across 24 hours. In practice you can size for average load rather than your worst moment.
Throttling is what happens when the average itself is too high, and it arrives in stages. A grace window first lets you borrow about 10 minutes of future compute with no penalty. Past that, new interactive requests get a 20-second delay, then are rejected while background jobs continue, and at the extreme everything is rejected until the debt is paid down. Staff see reports that are slow, then reports that return an error asking them to try again later.
If you outgrow the capacity, the options are straightforward. Scale it up from the Azure portal at any time and back down later. Pause and resume, which clears the debt immediately. Enable capacity overage billing, which prevents throttling but charges the overage at three times the normal rate. Or wait, because capacities recover on their own.
Scaling crosses a licensing boundary in one direction that matters: moving from F32 to F64 changes who needs a paid seat, and moving back down changes it again, cutting off every free viewer. Watch usage with the Fabric Capacity Metrics app, which is free and shows throttling before staff report it as a complaint.
OneLake storage is a separate line
What this costs you: a storage charge that grows quietly, keeps growing while your capacity is paused, and is not covered by a reservation.
If you use Fabric, data stored in OneLake is billed separately from compute. Microsoft's OneLake consumption documentation is direct: "OneLake storage is billed at a pay-as-you-go rate per GB of data used and doesn't consume Fabric Capacity Units."
Four details matter before the first invoice. Reading and writing data does consume capacity units, so transactions come out of compute while storage is its own charge. Soft-deleted data is billed at the same rate as live data. Storage keeps billing while a capacity is paused. And a reservation covers capacity usage only, not storage.
Refresh limits and model size limits
What this costs you: these two numbers push more companies up a tier than any feature does.
Microsoft's data refresh documentation sets the ceiling: eight scheduled refreshes a day on shared capacity, which is where Pro content lives, and up to 48 a day on Premium Per User or Fabric capacity. Refreshes on shared capacity must also finish within two hours, against five hours on a capacity.
Eight a day sounds generous until you write it out. Refreshing hourly across a nine-hour working day is nine refreshes, and you are over. Operations teams watching today's orders hit this quickly.
Model size follows the same pattern. The default limit is 1 GB, per Microsoft's large semantic models documentation. Premium Per User raises that to 100 GB. On Fabric capacity the ceiling is tied to the capacity size: 3 GB on the smallest capacities, 10 GB at F32, 25 GB at F64, and higher above that, per the capacity limits table.
A well-built model summarizing a few years of transactions usually fits under 1 GB. One that loads every row of every table does not. Before buying a tier to solve a size problem, have someone check whether the model needs to be that large.
Sharing with people outside your company
What this costs you: external viewers are licensed, and the rules mirror the internal ones.
To share content outside your organization, the person sharing needs Pro or Premium Per User. The external guest needs either their own Pro or Premium Per User license, or a Free license if your content sits on F64 or larger capacity. Microsoft's licensing guide adds a caution worth heeding: check with your Microsoft representative before assigning paid licenses to guest users, to confirm you are within the terms of your agreement.
Publishing to a genuinely public audience is a different mechanism with different data-exposure implications, and it deserves its own conversation first.
Report Server and on-premises reporting
What this costs you: if you have a regulatory or connectivity reason to keep reports on your own servers, the rights come attached to purchases you may not have made.
Power BI Report Server runs reports on hardware you control. Per Microsoft's Report Server overview, the rights come from one of two places: a Fabric F64 or larger reserved instance, or SQL Server licensing. With SQL Server 2025, Standard or Enterprise core licenses carry the right. For earlier versions it applies only to Enterprise Edition core licenses with active Software Assurance, and it ends when that coverage ends.
Microsoft also states that Report Server is not included with any Fabric capacity below an F64 reserved instance. Publishing reports to Report Server requires a Pro license; viewing them does not.
Power BI inside Dynamics 365, Microsoft 365, and your own apps
What this costs you: an embedded report is still a licensed report, whichever window it appears in.
Microsoft Dynamics 365 Business Central can display Power BI reports inside its own pages, as can Microsoft Teams, SharePoint, and PowerPoint. In every case the viewer needs a Power BI license under the same rules: Pro or Premium Per User, unless the content sits on F64 or larger capacity, where a Free license and the Viewer role are enough. A Dynamics 365 license does not include Power BI.
Buying: commit annually, direct or through a partner
What this costs you: the commitment you pick determines whether you can shed seats when headcount changes.
Pro and Premium Per User seats are bought in the Microsoft 365 admin center, where you choose your term and billing frequency at checkout. Commit annually and pay annually. The annual term costs less per seat, one invoice a year is less administrative work than 12, and your rate is fixed for the term. The trade is that the seat count is locked until renewal, so commit to the analysts and report consumers you are confident about and add seats above that line as they appear. Microsoft covers adding licenses, generally straightforward mid-term, and removing them, which depends on where you are in the term. Fabric capacity is bought in Azure, billed by the second, with reservations for one or three years that can be returned at a prorated credit, subject to a rolling cap on cancellations.
You can also buy both through a Cloud Solution Provider, a Microsoft-authorized partner that provisions and bills the subscription for you. Microsoft's buy a Fabric subscription page confirms this changes nothing about how the product works. It changes who you call, who consolidates the invoice, and in some cases whether staff can sign themselves up for licenses, since partner-managed tenants often have self-service purchasing turned off.
Gotchas that catch non-technical buyers
- Free does not share. The authoring tool costs nothing, so the pilot looks free. Distribution is the purchase.
- A small capacity does not remove seat costs. F2 through F32 buy performance and Fabric features. Every viewer still needs a paid seat.
- F64 does not remove seat costs for authors. Anyone building or editing outside their own workspace still needs Pro or Premium Per User.
- Premium Per User is contagious. Its content can only be opened by other Premium Per User holders, including people who only read it.
- Scaling below F64 cuts off free viewers. A cost-saving change in the Azure portal can remove access for most of the company.
- Pausing a reserved capacity saves nothing. The reservation bills whether the capacity runs or not.
- OneLake storage keeps billing while paused, and a reservation does not cover it.
- Report Server rights need a reserved F64 or larger, or the right SQL Server licensing. Pay-as-you-go capacity does not carry them.
- The 60-day Fabric trial behaves like an F64. Free-licensed colleagues can view content during it. If the trial ends on a smaller capacity, they cannot.
- Microsoft 365 E5 includes Pro; the plans most midsize companies run do not.
A short decision guide
A handful of viewers, under 10 people. Buy Pro seats for everyone who authors or reads, and skip capacity entirely. If one analyst needs bigger models or more frequent refreshes, check whether the model can be trimmed before moving anyone to Premium Per User, since that tier pulls their readers along too.
Dozens of viewers, roughly 10 to 100. Pro seats are usually still cheaper here. Run the crossover arithmetic anyway, since the ratio moves with pricing. Consider capacity only if you need Fabric itself, and remember that below F64 your viewers still need seats. If refresh frequency is the constraint rather than viewer count, Premium Per User for the authors may solve it alone.
Hundreds of viewers. Here capacity changes the shape of the bill: F64 or larger, Pro seats for authors only, Free licenses and the Viewer role for everyone who reads. Choose deliberately between a reservation and pay-as-you-go with scheduled pausing, and size against measured usage rather than a guess, using the metrics app during the trial.
In all three cases, size capacity to the work rather than to the headcount. Two hundred people opening one small dashboard each morning is a lighter load than four analysts running heavy refreshes all day.
Where to take this next
The useful conversation starts with your report inventory: which numbers people open, how fresh they need to be, and how many colleagues read versus build.
Bring us that list along with your current Microsoft 365 plan, and we can map it to a licensing shape and tell you where the crossover sits for your headcount. If the answer is that Pro seats are sufficient and Fabric would be premature, that is a short conversation, and a useful one.