Momentum One Academy · glossary
Definition
A Microsoft Fabric capacity is a distinct pool of compute resources inside a tenant. Its SKU determines the available processing power, measured in capacity units (CUs). Workspaces are assigned to capacities, which then run the workloads for their Fabric items. Capacity is therefore a compute and operating boundary, not a workspace, storage account, or license by itself. 1
Where it sits in Fabric
The practical hierarchy is tenant, capacity, workspace, and item. OneLake provides the tenant-wide data foundation; a workspace organizes items and access for a team; the assigned capacity supplies shared compute for pipelines, notebooks, lakehouses, warehouses, semantic models, and other workloads. Workspace roles control collaboration, while capacity administrators and contributors control capacity-level operations such as settings and workspace assignment. 1 2 3
Capacity operations connect directly to observability. Monitor hub is a permission-scoped, currently preview surface for job runs, alerts, applications, and capacity-and-cost views. The Capacity Metrics app goes deeper into capacity health, compute, storage, timepoints, and item-level consumption. 4 5
When to use it
Use a Fabric capacity when a solution needs Fabric workloads beyond personal or shared use, or when production teams need a deliberate compute boundary. Choose capacity boundaries from measured concurrency, workload criticality, region, cost ownership, and isolation needs. Multiple capacities can reduce the blast radius of a noisy workload, but they also add assignment and governance overhead. 1 6
Practical example
A Sales-Prod workspace contains ingestion pipelines, a curated lakehouse, and a semantic model. Assigning it to a production Fabric capacity gives those items a named compute boundary. After a month of representative refreshes and report usage, the platform team reviews Capacity Metrics to identify peak CU consumption and uses Monitor hub to trace a failed pipeline run to the correct workspace and owner.
Production caution
Do not treat an SKU as an unlimited runtime guarantee. Fabric can burst and smooth usage, but sustained overload can first delay interactive requests, then reject interactive requests, and eventually reject new requests. Capacity Metrics data also has processing latency, so do not resize from one slow report or one apparently quiet snapshot; separate item design problems from capacity contention and review trends. 5 6
Workshop connections
In Module 1 — Fabric Architecture & Core Concepts, place capacity alongside tenant, OneLake, workspace, and item boundaries. In Module 6 — Performance Optimisation & Time Travel, compare file maintenance, Spark, refresh, and query improvements with their capacity impact rather than judging performance by elapsed time alone. In Module 10 — Administration & Governance, define capacity and workspace owners, delegate only the required roles, review capacity evidence, and use Monitor hub as triage before deeper diagnosis. This reflects Fabric’s broader operating model of governing, securing, operating, and optimizing the data estate. 7
Ready to turn these boundaries into a production-ready Fabric design? Explore the Momentum One Academy Fabric Production Workshop.
References
Plan your workshop