Reporting & analytics articles

Fabric capacity overage: pay 3x to skip throttling, or size the capacity properly?

Fabric capacity overage is a billing switch, not a performance feature. When an F SKU capacity runs out of headroom, Microsoft Fabric normally throttles: it delays, then rejects, new work. With capacity overage on, Fabric instead pays off the excess by charging your Azure subscription at three times the pay-as-you-go rate, up to a rolling 24-hour threshold you set. It is switched on by default for new capacities. That makes it a sensible safety net for rare spikes and an expensive habit for a capacity that is simply too small — and the difference between the two is arithmetic you can do in five minutes.

What overage actually bills

Capacity overage charges only for the capacity-unit (CU) hours that would otherwise have pushed the capacity into throttling, through its own Azure meter at three times the pay-as-you-go rate. There is no standing charge: a capacity that never overloads costs nothing extra with overage switched on. Everything else about the capacity — its size, memory, speed — is unchanged.

Microsoft’s capacity overage documentation (opens in new tab) is precise about the mechanics. Overage activates at the moment the capacity would enter its first throttling stage, and what it bills is the accumulated carryforward — the smoothed usage the capacity has already committed beyond its SKU. Charges appear in Azure Cost Management under a separate meter named Capacity Overage Capacity Usage CU, which is the line to filter on when the invoice looks wrong.

The table below puts the three rates side by side per CU-hour. The pay-as-you-go and reservation figures are the East US rates we recorded from Microsoft’s rate card on 12 September 2026 for our Microsoft Fabric licensing cost guide; the overage figure is simply three times pay-as-you-go, as Microsoft defines it. Rates are regional and Microsoft’s to change.

How the CU-hour is boughtEast US rate per CU-hourRelative to pay-as-you-goWhen it applies
One-year reservationAbout $0.107 ($938 per CU per year ÷ 8,760 hours)About 0.6xEvery hour of the term, used or not
Pay-as-you-go$0.181xEvery hour the capacity is running, billed per second
Capacity overage$0.543xOnly CU-hours above the SKU that would otherwise be throttled

Not sure whether overage makes sense for your capacity? Ask us and we’ll look at your usage with you.

Ask about my capacity →

That spread is the whole story. An overage CU-hour costs five times what a reserved CU-hour does. Paying it occasionally is cheap insurance; paying it every day is buying capacity at the worst price on the menu.

Where overage sits in the throttling ladder

Overage replaces throttling, so it helps to know what throttling does. Fabric lets operations burst above the SKU and then smooths their cost forward in time: interactive operations such as report queries over a minimum of five minutes, background operations such as refreshes and pipelines over 24 hours. Microsoft’s throttling and smoothing documentation (opens in new tab) lays out the ladder:

  • Up to 10 minutes of future capacity used: built-in overage protection, no throttling and no extra charge.
  • 10 to 60 minutes: new interactive operations are delayed by 20 seconds.
  • 60 minutes to 24 hours: new interactive operations are rejected; background jobs still run.
  • More than 24 hours: everything new is rejected until idle time burns the debt down.

Capacity overage steps in at the first rung — the point where 20-second delays would begin — and pays off the carryforward so users never see them. It pays the minimum needed to keep the capacity running for the current window, not the whole future debt, so a capacity that is genuinely overloaded keeps drawing overage until it hits your threshold, and then throttles anyway.

Leave it on, or let it throttle?

Leave capacity overage on when an interruption costs more than the surcharge: a capacity serving reports to customers or executives, or one where a rejected refresh breaks something downstream. Microsoft’s own guidance is to use it for rare, unexpected spikes or small, regular ones, and to scale up instead if you are throttled regularly outside those cases.

Switch it off when throttling is tolerable and the money is not. A development or test capacity, a sandbox where analysts experiment, or an overnight batch capacity with slack in its schedule can absorb a 20-second delay or a retried job without anyone noticing. On those, throttling is free feedback that something needs optimizing, and overage just converts the feedback into an invoice.

One point that catches people out: the threshold is not a hard cap. Microsoft calls it a spending threshold. Fabric evaluates processed overage every five minutes, operations already running continue, and work for up to five minutes after the threshold is reached still bills at the overage rate. If you need a hard ceiling on spend, the only one is turning overage off. Two smaller traps: switching overage on during a heavy throttling event bills the entire accumulated carryforward at once, and scaling a capacity down while overage is on can turn the old workload into a large overage charge on the smaller SKU.

What threshold to set

The threshold is a CU-hour allowance over a rolling 24 hours, set as a percentage slider in 5 percent steps or as an absolute CU-hour value. New capacities default to 25 percent. An F2 provides 48 CU-hours a day, so 25 percent there is 12 CU-hours — at the East US rate above, a little under $6.50 of overage in any 24-hour window before throttling resumes.

Microsoft recommends keeping the threshold below one-third of the SKU’s daily CU-hours, because that is roughly where three-times overage costs the same as running the next SKU up on pay-as-you-go. A threshold higher than that is only worth it for short, severe interactive spikes that would still throttle after scaling up. In practice we start most capacities at or below the default, then move it once the Capacity Metrics app has shown a few weeks of real overage events.

Two operational details matter. The threshold consumes Fabric quota — one twenty-fourth of the threshold in CUs — so a large threshold can fail to save if the subscription’s quota is tight. And capacity overage email notifications, currently in preview, can warn you at a chosen percentage of the threshold and when the capacity enters or leaves an overage state. Turn those on; an overage you only learn about from the monthly invoice has already happened thirty times.

When upsizing the F SKU is cheaper than overage

F SKUs double at each step, so moving up one size adds as many CUs as you already have. The table below compares what a full day of default-threshold overage costs against what the next SKU up adds per day, on pay-as-you-go and on a one-year reservation. Daily CU-hours come from Microsoft’s documentation; the dollar figures are our arithmetic on the East US rates above, and the reserved column spreads $938 per CU over 365 days.

Current SKUCU-hours per dayOverage if the 25% default threshold is used in fullNext SKU up, extra per day, pay-as-you-goNext SKU up, extra per day, one-year reserved
F24812 CU-hours = $6.48$8.64About $5.14
F819248 CU-hours = $25.92$34.56About $20.56
F32768192 CU-hours = $103.68$138.24About $82.24
F641,536384 CU-hours = $207.36$276.48About $164.47

Read across a row and the rule falls out. Against a pay-as-you-go upsize, overage stays cheaper until it reaches about one-third of daily CU-hours — Microsoft’s figure. Against a reserved upsize, the break-even drops to roughly one-fifth, which is below the 25 percent default. So a capacity that exhausts its default threshold most days would be cheaper on the next SKU reserved, and would get the extra headroom for every hour rather than only the throttled ones.

The comparison has honest limits. A reservation is a year’s commitment while overage bills only on the days it fires, so an occasional spike never justifies one. Going from F32 to F64 also changes viewer licensing, not just compute, which our Fabric licensing cost breakdown covers. And overage is a symptom: before buying a bigger SKU, check whether one semantic model or one notebook is consuming the capacity. A report that scans far more data than it needs is cheaper to fix than to host, which is the work our Power BI performance optimization consulting exists for.

Smoothing strategies that keep overage from firing

The cheapest overage CU-hour is the one you never consume. Because background work is already smoothed over 24 hours, most overage on small capacities comes from interactive peaks stacked on top of a heavy background load. Three levers reduce it without changing the SKU.

Move and trim background work. Every refresh and pipeline run adds a slice to every 30-second timepoint for the next day. Cutting refresh frequency on models nobody reads hourly, switching full refreshes to incremental, and retiring orphaned pipelines lowers the floor that interactive peaks sit on. The Capacity Metrics app’s timepoint drill-through shows which operations made up a given overload.

Scale up for predictable peaks, at 1x rather than 3x. If overage clusters around month-end close or a Monday-morning reporting rush, a scheduled scale-up for those days bills the extra CUs at the pay-as-you-go rate instead of three times it, and scaling up also burns carryforward down faster. Scale back down only once the peak and its carryforward have cleared, for the reason above.

Pause idle capacities on a schedule — with care. F capacities can be paused and resumed by an Azure Automation runbook or the Fabric REST API, which stops the hourly bill on development and batch capacities outside working hours. But pausing does not let smoothed usage drain for free. Microsoft’s documentation states that on pause, the remaining cumulative overages and smoothed operations are summed and added to your Azure bill. A 24/7 capacity uses quiet overnight hours to burn down the day’s smoothed background work at no extra cost; a capacity paused at 7pm pays for that work immediately. Scheduled pausing suits capacities with little pending work at shutdown, not ones that run heavy evening refreshes.

Finally, if a capacity in background rejection is still admitting new large jobs because overage is paying them through, Microsoft suggests setting a surge protection limit of 100 percent so Fabric stops accepting new background work while overage clears the backlog.

Where this fits in a Fabric budget

Overage is the variable line on a Fabric invoice; the SKU and the reservation are the fixed ones, and those decisions sit in our Microsoft Fabric licensing cost guide — this page deals only with what happens when the capacity runs out. If you are sizing a capacity for the first time, our Microsoft Fabric implementation services start from a measured consumption baseline rather than a guess, and if an existing Power BI Premium estate is moving onto an F SKU, Power BI to Microsoft Fabric migration services cover sizing the replacement capacity so overage does not become the default way it copes. Fabric capacity and any overage are billed to you by Microsoft through Azure, never through us; what we quote is the fixed fee for the work.

FAQ

Questions admins ask about overage billing

On new capacities, yes. Microsoft's documentation states that capacity overage is enabled by default when you create an F SKU capacity, with a default rolling 24-hour threshold of 25 percent. You can opt out during creation, or change it later under Capacity settings in the Fabric admin portal or through the OneLake catalog.

Three times the pay-as-you-go rate, billed through a separate Azure meter, and only for the capacity-unit hours that would otherwise have caused throttling. There is no standing charge. At the East US rate we recorded on 12 September 2026, that is $0.54 per CU-hour against $0.18 on demand. Rates are regional.

No. Microsoft describes it as a spending threshold, not a hard cap. Fabric checks processed overage every five minutes, operations already running keep running, and work for up to five minutes after the threshold is reached still bills at the overage rate. The only true cap is switching overage off and accepting throttling.

No. It adds no memory, speed or extra capacity units. It only pays off the carried-forward usage that would trigger delays and rejections, so jobs run as they normally would without being throttled. Memory and concurrency limits on the SKU still apply, and a slow report stays slow.

Microsoft's guidance puts the break-even at about one-third of the SKU's daily CU-hours when the next SKU is bought pay-as-you-go. If you reserve the larger SKU instead, our arithmetic puts the break-even nearer one-fifth. Regular overage above those levels means the capacity is undersized, not unlucky.

No hourly meter, no surprise invoice

The written proposal names the deliverables, the timeline and the price. Nothing starts until you approve it, and nothing is added later without your written agreement. 30 days of hypercare after handover are included.

Get my fixed-price proposal →

Get your Fabric capacity sized with help

Tell us which SKU you run and when throttling hits. A senior US-based engineer scopes it with you, and the proposal names one price before any work begins.

Free consultation · Written fixed price before any work starts · Reply within one business day

Book a free consultation

See what your Microsoft 365 project would take

Tell us what you’re trying to fix. We reply within one business day, and you get a written fixed-fee proposal to approve before any work begins.

Step 1 of 2: What do you need?

How can we help? Pick one

Tap to add, or type your own:

Step 2 of 2: Where should we reply?

Used only to contact you about this inquiry.

We reply within one business day. Fixed price approved before any work begins. We don’t sell your information.

Tell us what you’re trying to fix — a report, an approval process, an intranet, a Copilot rollout. We scope it as a fixed-fee project, you approve, and a senior engineer delivers in 2–4 weeks.

  • Fixed-fee scope agreed before any work starts
  • Senior, US-based Microsoft engineers — no handoffs
  • Typical 2–4 week delivery
  • 30 days of post-delivery hypercare included
Book my free consultation