Capacity pain shows as throttling, slow interactive queries, or refreshes stuck Queued. Collect evidence before changing SKU; publish Fabric capacity recommendations only after commercial sizing is confirmed.
Measure first
- Fabric Capacity Metrics / Premium metrics: CU spikes correlated to refresh vs interactive windows.
- Largest Import dataset sizes and concurrent scheduled refreshes overlapping the spike.
- DirectQuery datasets’ daytime query volume (Performance Analyzer + warehouse DMVs).
Checklist before resize
Eliminate personal-gateway contention and non-folding mashup. Stagger schedules. Move heavy transforms to dataflows/pipelines. Only then consider higher SKU or Autoscale. Document Max memory / rejection behavior your admins expect when CU saturates so support does not treat throttling as a random outage.
Re-test a representative board report and a worst-case refresh after any SKU change; keep before/after screenshots in the change ticket.