Guide · 8 min read
How to choose a Microsoft Fabric capacity size
Most sizing advice starts at F64. Most mid-sized Australian businesses start much smaller. Here is what we have learned running real workloads on small capacities.
Short answer
What size Microsoft Fabric capacity do we need?
Start with the smallest capacity that runs your nightly loads and daytime reporting without being throttled, usually F2 or F4 for one ERP and a mid-sized reporting team. Measure actual usage for a few weeks in the Fabric Capacity Metrics app, then size up only if the numbers say so.
The size you need depends far more on how much data you process, and how often, than on how many people view reports.
How capacity works
Smoothing, bursting and throttling in plain English
Bursting
A job can briefly use more compute than your capacity size, so heavy work finishes faster.
Smoothing
Fabric spreads the cost of that work over time. Background jobs such as pipelines and refreshes are smoothed over 24 hours; interactive work such as report queries is smoothed over a few minutes.
Throttling
If you keep using more than your capacity can pay back, Fabric slows things down: first delaying interactive queries, then rejecting them, and finally rejecting background jobs until usage recovers.
From our delivery work
What an F2 handled for one manufacturer
For a furniture and fabrics manufacturer, we run nightly loads from their ERP into a Fabric lakehouse on an F2. The ERP extract takes around 45 minutes and stock processing around 38 minutes, and the full run finishes before 2am, ready for morning reports.
During heavy development work, when we were reprocessing history and testing at the same time, the F2 reached its limit. Day-to-day production loads fit comfortably. The lesson: size for steady production use, and scale up temporarily for heavy one-off work rather than buying a bigger capacity permanently.
The biggest gain came from the source, not the capacity. A fix to the ERP's API is expected to cut the nightly runtime from about an hour to 20 to 25 minutes, which frees room for extra refreshes on the same F2.
Rules of thumb
A starting point by situation
These are starting points, not answers. Your real size comes from measured usage.
| Situation | Typical starting size |
|---|---|
| Proving the approach with one ERP and a few reports | F2 |
| One or two systems, nightly loads, a mid-sized reporting team | F2 to F4 |
| Several systems or entities, refreshes during the day, a data agent in use | F4 to F8 |
| Large data volumes, many concurrent report users, or near real-time feeds | F16 and above |
| Many people who only view reports and would otherwise need Pro licences | Consider F64 |
Before you size up
Check these first
- Are you reloading full tables when only changed rows need loading?
- Are refreshes scheduled more often than anyone looks at the reports?
- Is one heavy notebook or query using most of the capacity?
- Are development and production running on the same capacity?
- Is a slow source system API the real bottleneck?
Questions we get asked
Frequently asked questions.
Is F2 enough for Microsoft Fabric?
For one ERP, nightly loads and a mid-sized reporting team, often yes. It becomes tight with heavy daytime processing, several large sources, or development and production sharing the same capacity.
How do we monitor Fabric capacity usage?
Install the Microsoft Fabric Capacity Metrics app. It shows usage by workload and item over time, and when throttling occurred, so you can see exactly what is using the capacity.
Can we change capacity size later?
Yes. Pay-as-you-go capacity can be scaled up or down in the Azure portal. Reserved capacity is committed for its term, which is why we recommend measuring before reserving.
Should development and production share a capacity?
On small capacities, heavy development work can slow production loads. A small separate pay-as-you-go capacity for development, paused when idle, avoids that.
Next steps
Hitting capacity limits, or about to buy one?
In a 30-minute call we'll look at what is using your capacity, or what you plan to run, and recommend a size based on it.
Related reading
- Microsoft Fabric pricing in Australia, explained
- Microsoft Fabric vs Power BI: which do you need?
- Connecting your ERP to Microsoft Fabric
- Microsoft Fabric consultants
Prefer to talk it through? A 30-minute call is usually enough.
Bring us the process that is slowing you down.
A 30-minute discovery call is enough to work out whether automation is the right answer, and what it would take.