How an IT lead should read the new Capacity Metrics app before approving an upsize.
Fabric Lab · Issue 11 · By Dave Goyal
On Oct 12 (2026), Microsoft published a redesigned Capacity Metrics app in preview. It adds the percentage of time a capacity spent in throttling. The date comes from a release summary, not Microsoft’s own post, so treat it as reported.
Good number. Easy to misread.
Here is the misread: a capacity shows 0%, so it is healthy, so the upsize request must be about something else. Or the reverse: a red bar, so sign the bigger SKU.
Both skip a step. I use a four-part check, The Upsize Gate. Facts first, then the gate.
What changed
- Capacity Metrics app version 3.0 ships two reports. The generally available default opens first; the redesigned preview report sits in the navigation pane.
- The preview Health page shows the percentage of time each capacity spent paused, not throttled, in interactive delay, in interactive rejection, in background rejection, and under surge protection, for the last 24 hours and 14 days.
- The new Item details page keeps up to 14 days of history for one item: smoothed and unsmoothed CU, duration, operation, status, and user.
- A capacity is not counted as throttled “if mitigations such as capacity overage prevented throttling.”
- Overage became generally available for F SKUs in September 2026 and “is enabled by default when you create a capacity.” It bills “at three times the pay-as-you-go rate,” and the threshold “isn’t a hard spending cap.”
Two corrections to the headlines. The 14-day window is not new; the default Compute page already showed 14 days. And Microsoft’s pages conflict on status: what’s new lists the enhancements as generally available, while the page-level docs say preview.
What we are seeing
Zero is not proof of health. Overage is on by default and a capacity rescued by overage is not counted as throttled. Look at overage CU hours and the Azure overage meter.
The hidden meter is not small. Azure retail pricing in West US 2 lists $0.18 per CU hour for capacity and $0.54 for overage. An F64 provides 1,536 CU hours a day. Microsoft’s default threshold is 25% of daily CU hours. My arithmetic, not Microsoft’s: 384 overage hours at $0.54 is up to $207.36 a day, against $276.48 a day for the F64 itself. Early signal only. The threshold is single-sourced and not a spending cap.
Overage is a safety net, not a plan. Microsoft says to keep the threshold below one-third of daily CU hours, where costs are similar to scaling up. It also says overage gives no performance boost, and that regular throttling outside rare spikes means scale up.
An upsize is a big bet. F SKUs double at each step and price is per CU hour, so one step doubles compute cost.
The slow report may not be capacity at all. Microsoft says slow performance is often the design of an item, and only sometimes capacity throttling.
The lens: The Upsize Gate
The Upsize Gate: four readings an IT lead takes from the preview report before signing a bigger SKU.
1. Time, not spikes. Read the 14-day view for percentage of time in each throttling state. One red bar is not a pattern. Microsoft’s stages: up to 10 minutes of future capacity is overage protection with no throttling, 10 to 60 minutes adds a 20-second delay to new interactive jobs, and 60 minutes to 24 hours rejects interactive jobs.
2. The hidden meter. Check overage CU hours billed. A clean throttling reading with a busy overage meter is a capacity that is already undersized.
3. Who burned it. Use the smoothed view to see when users were hurt, and the unsmoothed view to see which activities used the CUs. Then open Item details for the offenders. Microsoft notes the impacted items list leaves out operations users never tried because the capacity was already throttled, so it may undercount.
4. Fix, isolate, remeasure. Optimize the top items. Scale out or add surge protection where it fits, then reread 14 days. Upsize only if throttling or overage stays sustained. Compare before and after a resize in CU per second, because the CU % chart doubles if the SKU is halved with no change in load.
What to do
- This week: Open the preview Health page and record the 14-day percentage in each state for every capacity. Pull overage CU hours from the Azure overage meter. Put both numbers on one page.
- This month: Run the top items through the gate. Set alerts outside the app; Microsoft says it does not support them, and data lands 10 to 15 minutes after activity. Check your overage threshold against the one-third guideline.
- This quarter: Reread 14 days after the fixes. If throttling or overage is still sustained, sign the upsize. Microsoft is plain that consistently high throttling means scale up or out. The gate is not a way to say no forever. It is a way to say yes with evidence.
What we are still watching
No Microsoft or independent source gives an acceptable non-zero throttling percentage. Microsoft defines healthy as no throttling at all. The only numbers I found are one practitioner’s rules of thumb: a 70% average to start a sizing conversation, and 40% from a single item as a trigger to optimize it first. A starting argument, not a standard. I also cannot tell whether capacities created before overage went generally available had it switched on.
What this means for you
If an upsize request is on your desk now: ask for the gate before the signature. Four readings, one page. If the requester cannot produce them, you do not have a capacity problem yet. You have a measurement problem.
If you have not been asked yet: you are early. Build the habit now.
Which number does your team bring when it asks for a bigger SKU: a spike, or 14 days of time-in-state? Tell me in the comments. I read every reply.
Dave Goyal
Founder, Think AI Consulting Corporation
Host, The Think AI Podcast
Sources
- Microsoft Fabric release summary, Releasebot, 2026-10-12. https://releasebot.io/updates/microsoft/microsoft-fabric (aggregator; Microsoft’s original post returned 403)
- Microsoft Learn, Capacity Metrics app. https://learn.microsoft.com/en-us/fabric/enterprise/metrics-app (2026-08-28, updated 2026-10-12)
- Microsoft Learn, preview Health page. https://learn.microsoft.com/en-us/fabric/enterprise/metrics-app-preview-health-page (2026-08-25, updated 2026-10-12)
- Microsoft Learn, Health page. https://learn.microsoft.com/en-us/fabric/enterprise/metrics-app-health-page (2026-01-26, updated 2026-10-12)
- Microsoft Learn, preview Compute page. https://learn.microsoft.com/en-us/fabric/enterprise/metrics-app-preview-compute-page (2026-08-25, updated 2026-10-12)
- Microsoft Learn, Throttling. https://learn.microsoft.com/en-us/fabric/enterprise/throttling (2026-08-14, updated 2026-08-18)
- Microsoft Learn, preview Item details page. https://learn.microsoft.com/en-us/fabric/enterprise/metrics-app-preview-item-details-page (2026-08-25, updated 2026-10-12)
- Microsoft Learn, Capacity Metrics app. https://learn.microsoft.com/en-us/fabric/enterprise/metrics-app (2026-08-28, updated 2026-10-12)
- Microsoft Learn, What’s new in Fabric. https://learn.microsoft.com/en-us/fabric/fundamentals/whats-new (accessed 2026-10-22, September 2026 entry)
- Microsoft Learn, Enable capacity overage. https://learn.microsoft.com/en-us/fabric/enterprise/enable-capacity-overage (2026-09-11, updated 2026-09-29)
- Microsoft Learn, Capacity overage overview. https://learn.microsoft.com/en-us/fabric/enterprise/capacity-overage-overview (2026-08-26, updated 2026-09-29)
- Azure Retail Prices API, Microsoft Fabric, West US 2. https://prices.azure.com/api/retail/prices?$filter=serviceName%20eq%20%27Microsoft%20Fabric%27%20and%20armRegionName%20eq%20%27westus2%27 (accessed 2026-10-22)
- Derived by the content team from L18, L19, L20, L21 (2026-10-22). Inputs: 1,536 x 0.25 = 384 CU hours; 384 x $0.54 = $207.36; 64 x 24 x $0.18 = $276.48.
- Derived from Microsoft Learn, Fabric licenses. https://learn.microsoft.com/en-us/fabric/enterprise/licenses (2026-06-15, updated 2026-08-05) and L20
- Microsoft Learn, Optimize capacity. https://learn.microsoft.com/en-us/fabric/enterprise/optimize-capacity (2024-07-30, updated 2026-09-29)
- Microsoft Learn, Surge protection. https://learn.microsoft.com/en-us/fabric/enterprise/surge-protection (2026-09-23, updated 2026-09-29)
- SpendWeave, Fabric Capacity Metrics app. https://spendweave.com/blog/fabric-capacity-metrics-app/ (2026-06-20)
New here? I write Data and AI Demystified, the hidden costs and reality checks on Microsoft Fabric most partners won’t publish. Subscribe here.
Click here to read Dave’s Demystify Data and AI newsletter on LinkedIn.