Microsoft Fabric Capacity Overage Billing Goes GA

There's a setting sitting in your Fabric capacity admin page right now that will bill you at three times the normal rate. And Microsoft's own documentation can't agree on whether it's switched on by default.

Three times. Of pay-as-you-go, which is already the expensive way to buy compute.

Already the expensive way. Capacity overage went generally available around FabCon Europe, and the write-up I'd point you at is from Hopton Analytics, dated the thirteenth of October. Their advice is blunt: go check the overage setting on every single capacity you run.

Okay, but back up. What happened before, when you blew past your capacity?

You got throttled. Your refreshes queued, your reports went slow, people complained, and that pain was the signal to buy a bigger SKU. The bill stayed flat.

So the old failure mode was angry users. The new one is a quiet invoice.

On a separate meter, yes. And here's the part I'd underline for anyone who signs off on budget. The threshold you set is a spending threshold, not a hard cap.

Wait. So if I type in a number, that's... a suggestion?

It's a limit on how much overage you're willing to pay for, not a wall that stops consumption dead. Which is a very different thing to explain to a finance director after the fact.

And nobody told me it was on. That's the bit that bothers me.

That's the open question. Hopton's reading is that Microsoft's documentation is not consistent on whether overage is on by default. So the honest answer is: don't assume, go look. It takes two minutes per capacity.

Two minutes per capacity, across however many I run. Great Thursday.

Better than the alternative. And there's a second thing that went GA alongside it that actually helps you — on-demand billing for Spark. It used to be called Autoscale Billing for Spark.

Meaning notebooks stop eating the shared capacity?

Meaning if notebooks are what spike you, you push that spend onto its own meter instead of letting it starve your reports. Hopton says switch it on if that's your pattern.

That's the thing about shared capacity, right? One data scientist runs something enormous at four p.m. and the whole finance dashboard crawls.

Which Microsoft is also chipping at from the other end. The Power BI October feature summary lists more accurate surge protection with semantic models.

Surge protection being... the thing that sacrifices one workload to save the rest?

Roughly. It's one line in a release summary, so I won't oversell it. But you can see the shape of the strategy — better throttling on one side, a paid escape hatch on the other.

Alright, so what's the thing that actually changes an architecture decision, rather than just an admin checkbox?

F zero. A Fabric capacity with nothing provisioned — no compute you reserve before you start. Plus on-demand billing extending to the other Fabric workloads, charged on actual use, with a spend limit over a rolling twenty-four hours.

So the monthly commitment goes away entirely? That's been the number one objection I hear from mid-size teams.

Hopton says exactly that — for organisations that held back from Fabric because of a monthly capacity commitment, this could be the change that matters most. Now. Guess what F zero costs.

Per capacity unit hour? I'll say something slightly above the F two rate, so they're not cannibalising reserved.

Nobody knows. Microsoft hasn't published F zero pricing. Both F zero and on-demand for the other workloads are arriving in preview, prices unannounced.

So I can't actually model whether it beats a small reserved capacity.

You cannot. That's Hopton's exact caveat, and they're holding their licensing calculator until the numbers land. Which is also roughly what happened with apps in Power BI.

The FabCon announcement? Barcelona, end of September?

The twenty-ninth of September. Microsoft's own blog confirms apps coming in preview for Pro and Premium Per User, a building experience in Desktop in "the coming months", and that apps can take input and write data back. The claims that it's free, and that each app gets a one gigabyte database — those only appear in partner reports.

So if I'm the person who has to present a Fabric number to a board this quarter, I've got a confirmed cost multiplier and a bunch of unpriced previews.

That's the week. What you can act on today is the overage setting and the Spark meter. Everything else, you watch.

And next week, the thing to watch is F zero pricing.

F zero pricing, and whether Microsoft tidies up that default-on documentation. If they quietly correct it, that tells you something too.

I'm going to go open fourteen admin portals.