Microsoft Fabric F2 pricing: what a small team actually pays
What an F2 capacity costs in euros, why a one-year reservation is usually the wrong lever, and the Power BI licensing cliff at F64 that quietly lands on your invoice.
Everyone finds the F2 price first. Smallest capacity, a few hundred euros a month, that sounds survivable. So the pilot starts. Then the first full invoice arrives with a second line nobody mentioned in the demo, and someone has to explain to the finance lead why the cheap option grew a tail.
The tail is Power BI licensing. I will get to it, because it is the part almost nobody puts a euro number on. First, the honest sticker price.
How much does an F2 capacity actually cost?
An F2 is the smallest Fabric capacity Microsoft sells. Two capacity units, and capacity units are the only currency Fabric really understands.
In West Europe, an F2 lists at roughly €0.38 an hour on pay as you go. Left running around the clock that is about €278 a month, because Azure counts 730 hours in an average month. A one-year reservation brings it to roughly €165 a month, a discount of about 40 percent (Agilytic's Belgian and Dutch breakdown, Pexon's German F-SKU table, both checked August 2026).
Two things that price does not cover. OneLake storage is billed separately at about €0.02 per GB per month (Microsoft's OneLake consumption docs), which for a small company's entire history is rounding error. And Power BI licences for the people who read the reports, which is not rounding error at all.
One quirk worth knowing before you build a budget on any of this: Microsoft sets Fabric capacity prices in dollars and converts them to euros, so your euro number drifts month to month with the exchange rate. Open the Azure pricing page, set the region to West Europe and the currency to euros, and read today's number before you sign anything. Every figure in this post is an August 2026 snapshot, not a quote.
Why is Microsoft Fabric so expensive?
Because you are not buying usage. You are renting an engine, and the engine has a minimum size.
Three things make that feel expensive to a five person company. There is a floor: F2 is the smallest thing on the shelf, and there is no tier below it for a team of two. The clock runs on wall time, not on work done, so an idle capacity bills exactly like a busy one. And on the small sizes, per user Power BI licences sit on top of the capacity rather than inside it.
Compared to a consumption engine like BigQuery, where a quiet month is a cheap month, Fabric trades that flexibility for predictability. That trade is genuinely good for some companies and genuinely bad for others. I wrote about which side of that line a small business usually sits on in the warehouse or lakehouse post.
F2 or F4: which one should I start on?
F4 is twice the capacity units and twice the price, about €556 a month on pay as you go. That is the whole difference. Nothing unlocks at F4, you just get more engine.
In practice F2 is a build size. One person developing, small datasets, a couple of dashboards read by a handful of people. F4 suits a small team with one data engineer and a few analysts. Real production reporting for ten to twenty five people reading at once tends to want F8 (Datatako's SKU sizing guide).
So start at F2. You can resize a capacity in the Azure portal in a couple of minutes, and buying headroom you are not using is just a donation.
What happens when you push an F2 too hard?
This is the part that surprises people, because Fabric does not simply fall over. It has three behaviours stacked on top of each other, and they are all documented in Microsoft's metrics app calculations.
Bursting lets a job briefly use more capacity units than you bought, so a heavy notebook finishes instead of failing.
Smoothing then spreads the cost of that burst forward in time. Interactive work smooths over the next few minutes; scheduled background work smooths over the next 24 hours. This is why an F2 can handle a spike that looks impossible on paper.
Throttling is what happens when the smoothing runs out of future to borrow from. It arrives in stages. Once you have consumed the next 10 minutes of capacity, interactive requests get a 20 second delay. Past 60 minutes, they get rejected. Past 24 hours, background jobs get rejected too.
The practical translation: an overloaded F2 gets slow first, then rude. People report that "the dashboard is broken" long before anything actually breaks.
Does a Fabric capacity replace Power BI Pro licences?
Not until F64. This is the buying trap, and it is worth a picture.
On any capacity from F2 up to F32, everyone who opens a shared Power BI report still needs their own Power BI Pro or Premium Per User licence. The capacity buys performance, not seats. At F64 and above, people with the free licence and a viewer role can read reports with no seat of their own (Microsoft's licensing docs). The people who build reports still need a Pro seat either way.
Now the euros. Power BI Pro runs about €13 per user per month in Europe after Microsoft's April 2025 increase (Toucan Toco's European pricing breakdown; Microsoft's global list price is 14 dollars, which is €12.00 at the ECB reference rate of 1.1664 on 25 August 2026, and European list prices sit a little above the straight conversion).
An F64 reservation in West Europe is somewhere around €5,290 a month against roughly €165 for a reserved F2. That gap is about €5,125, which buys around 390 Pro seats. So the cliff only pays for itself once you have something like 400 report readers (SpendWeave reaches the same conclusion in dollars).
If you have 20 readers, that is €260 a month in seats. Buy the seats. Anyone telling a 20 person company to jump to F64 for licence relief is doing the arithmetic backwards.
Does a one-year reservation pay off?
Only if the capacity is genuinely on all the time. A reservation bills for every hour of the year whether the capacity is running or paused, and it saves about 40 percent (Microsoft's Fabric reservation docs). So the break-even is simple: a reservation wins when you would otherwise run the capacity more than about 60 percent of the hours in a month. That is roughly 14 hours a day, every day, weekends included.
Most small teams are nowhere near that. Which points at the lever nobody sells you, because nobody makes money from it: pause the thing.
Pay as you go capacity can be paused and resumed (Microsoft's pause and resume docs), and while it is paused the compute meter stops. Run an F2 for twelve hours on working days only and you are billing around 264 hours instead of 730, which puts you near €100 a month. That beats the reservation, and it beats it without a twelve month commitment.
Two cautions. Anything scheduled during the pause simply does not run, so your refresh windows have to sit inside the on hours. And if you use mirroring, the free mirrored storage allowance turns billable while the capacity is paused, so check that before you automate anything.
What should a small team actually buy?
Start on a pay as you go F2 with a pause schedule around your working hours. Buy Power BI Pro seats for the people who need them and count them honestly. Ignore reservations until the capacity has been on more than fourteen hours a day for three months in a row. Ignore F64 until you have several hundred report readers, which for most companies I work with means never.
And spend the effort you saved on the layer underneath, because a well shaped lakehouse on an F2 will outrun a messy one on an F8 every time. That shaping is what my lakehouse architecture work is, and the failure patterns it avoids are in the medallion mistakes post.
Bring me your Fabric bill and your reader count and I will tell you which of these levers is actually yours to pull. Ask any question you like in the free 30 minutes; you leave with an answer or a clear next step.