Published: 16 Oct 2026 | Author: James Beresford
Microsoft has announced F0, a way to run Fabric with no capacity bought up front and pay only for what you use. It's the biggest change to Fabric pricing since launch, but it isn't here yet, and it won't fix everything. In the meantime, two older decisions are still setting most organisations' bills: reserved versus pay-as-you-go, and whether capacity is running when nobody needs it. This article covers both, where the break-even sits, and what F0 really changes.
Microsoft has announced F0: Fabric and OneLake with no compute to buy up front, billed only for what you actually use. If you've been paying for a capacity that sits idle every night and every weekend, that announcement should make you sit up. And then it should make you ask an awkward question.
Why were you paying for all those empty hours in the first place?
Are you running a capacity around the clock that only works business hours? Did you buy a reservation because 41% off sounded too good to pass up? Has anyone actually checked? In our experience most organisations made these calls by default – one capacity, always on, bought however the first person in the Azure portal felt like buying it.
Per hour, yes. Overall, no.
Microsoft prices a one-year Fabric reservation at roughly 41% below pay-as-you-go. The catch is that you pay for every hour of the year whether the capacity is doing anything or not, and you can't pause your way out of it.
A reservation is a gym membership. Brilliant value if you go five times a week. An expensive donation if you go twice a month.
Back of a napkin maths shows it simply: A reservation costs about 59% of always-on pay-as-you-go, so any capacity running less than about 59% of the year's hours is cheaper on pay-as-you-go. Take business hours of 7am to 7pm, weekdays only: 60 of the week's 168 hours, roughly 36%. Paused outside those hours, a pay-as-you-go capacity costs about 36% of the always-on price against 59% for the reservation – close to 40% cheaper than reserving it, without changing a single workload.
Because Power BI is usually sitting on the same capacity, and Power BI users don't keep office hours.
A paused capacity takes everything on it offline: reports, semantic models, scheduled refreshes, the lot. If you're on an F64 so Free-licence staff can read reports, the depot supervisor checking a dashboard at 6am is going to notice.
There's a subtler trap too. Fabric smooths heavy bursts of activity over the following hours, so a big overnight refresh quietly burns down into idle time you'd otherwise have wasted. Pause while that smoothed usage is still outstanding and Microsoft bills it immediately, at pay-as-you-go rates. Scheduling the pause for straight after your heaviest refresh is a very efficient way to pay for capacity you thought you'd stopped using…
In my opinion the answer is to stop treating capacity as one decision. Reserve the production baseline that genuinely runs 24×7, and put data engineering, dev and test on separate pay-as-you-go capacity that sleeps. Reservations apply to a CU count in a region rather than a specific capacity, so an F64 reservation can cover two F32s. The single oversized, always-on capacity is where the money leaks.
Not very. Which is what makes it a little embarrassing when it hasn't been done.
F SKU capacities expose suspend and resume through the Azure management API, so the schedule can live wherever your automation already lives: an Azure Automation runbook, a Logic App, or a Power Automate flow running under a service principal. Resume at 6am, pause at 7pm, nothing on weekends.
The details are where it earns its keep. Make the pause check for running jobs and stay clear of the window after your big refresh. Feed it the public holiday calendar so it doesn't spin up for nobody on Australia Day. And give people a Teams button to wake it on a Saturday, so the analyst finishing month-end doesn't have to raise a ticket.
Not yet. And probably not for production.
Alongside F0, capacities from F2 upwards will be able to move individual billing categories outside the shared capacity and pay for what they consume, behind a rolling 24-hour spend limit. For a small dev environment that's genuinely good news. A capacity that only costs money while someone is using it beats an elaborate pause schedule, and I'd think twice before building one for a workload that could move to F0 in a few months.
But read the fine print. Both are preview, "in the coming weeks", and no prices have been published. On-demand usage isn't smoothed, and it's priced as a multiplier on the pay-as-you-go rate – so anything that runs all day is still likely to be cheapest reserved. Nothing has been said about Power BI licensing on F0 either, so the F64 free-viewer threshold still stands.
Full disclosure – Talos are a Microsoft Partner, so take this with a pinch of salt. But note the direction of travel. Capacity overage is now generally available, billed at three times the pay-as-you-go rate, and Microsoft's own documentation suggests it may be on by default for new capacities. Overusing a capacity used to cost you performance. Now it can cost you money, and nobody asks first.
Fabric Cost Management starts with hours, not discounts. The 41% headline only matters if the capacity is busy enough to use it, and F0 is about to add a third way to pay without making the first two obsolete.
If your Fabric bill has grown faster than your Fabric usage, our Fabric Health Check looks at exactly this: how your capacity is bought, how hard it's actually working, and where a split, a schedule or F0 would pay for itself. Two weeks, fixed fee, and for one client it halved the Fabric bill. Worth doing before your next reservation renewal, particularly with F0 pricing about to land.
Cheers,
James
CEO, Talos
Microsoft announced F0 and on-demand billing at FabCon Europe 2026 as coming to preview in the coming weeks, with timelines subject to change. Pricing hasn't been published. What we do know is that on-demand usage is priced as a multiplier of the pay-as-you-go rate, isn't smoothed, and that storage is charged separately.
For capacity that runs most of the time, yes. A reservation is charged at a substantial discount to pay-as-you-go rates, but you pay for it whether or not the capacity is running. Pay-as-you-go suits capacities that can be paused for long periods, such as development and test. The cheapest mix depends on how your workloads actually run across the week.
They're unavailable. Everything hosted on a paused capacity, including reports, semantic models and scheduled refreshes, stays offline until the capacity is resumed. That's why production Power BI is usually best kept on its own always-on capacity, separate from engineering and development work that can be paused.
Most likely smoothing. Fabric spreads heavy bursts of activity over the following hours, and if you pause while that smoothed usage is still outstanding, it's billed immediately at pay-as-you-go rates. Avoid scheduling a pause straight after your heaviest refresh or pipeline run.
Pay-as-you-go F SKU capacities can be suspended and resumed through the Azure management API. You can schedule this from an Azure Automation runbook, a Logic App or a Power Automate flow running under a service principal with rights on the capacity. Good schedules check for running jobs, respect public holidays and give users a way to wake the capacity out of hours.
Capacity overage lets a capacity keep running above its limit instead of throttling, with the extra compute billed separately at three times the pay-as-you-go rate and no reservation discount. It's now generally available, and Microsoft's documentation suggests it may be enabled by default on new capacities, so check the setting on every capacity you own.
The Talos Fabric Health Check is a two-week, fixed-fee review of your Fabric environment, from $9,995 AUD (ex GST). We find what's driving your bill, fix the quick wins on the spot, and hand over a ranked plan to bring the rest down, including which capacity to reserve, which to schedule, and what's worth moving to F0 once it arrives.
Get the latest Talos Newsletter delivered directly to your inbox
TECHNOLOGY PARTNERS
Our partners including Microsoft, UiPath, Databricks & Profisee enable us to deliver business outcomes using best of breed technologies and solutions.
Automation & Analytics Technologies for Business
Australian councils face backlogs, staff turnover, and rising community expectations. We help you improve service delivery and operational efficiency through process automation — integrated with TechnologyOne, Civica, and Pathway.
Talos helps organisations build reporting solutions using Jet Analytics
Talos helps organisations build modern data and analytics platforms using Databricks, Microsoft Fabric and Power BI. We work from strategy through implementation, ensuring you get value from the platform without unnecessary complexity.
F0, which Microsoft calls Fabric zero-provisioned, is a new entry point to Fabric and OneLake with no compute capacity to buy up front. On-demand billing is on by default, so you pay for the compute you consume rather than for a capacity sitting ready. Microsoft positions it for evaluation, development and workloads whose demand isn't yet predictable.