Power BI, Microsoft Fabric, and semantic models
How Power BI fits inside Microsoft Fabric — semantic models, Direct Lake, OneLake, and the unified data platform.
In 2023 Microsoft launched Microsoft Fabric — a unified data platform that subsumes Power BI alongside the rest of the Azure data stack. For Microsoft 365 customers using Power BI, Fabric isn't a replacement but a superset: Power BI capacity is now Fabric capacity, and new architectural patterns are built on Fabric primitives.
What Fabric provides
Fabric bundles several previously separate Microsoft data products into one platform:
- OneLake — a single unified data lake for the whole tenant, automatically organised by workspace.
- Data Factory — pipelines and dataflows for data integration.
- Synapse Data Warehouse — SQL-based warehouse.
- Synapse Data Engineering — Spark notebooks and lakehouses.
- Synapse Data Science — ML workflows.
- Real-Time Analytics — KQL-based streaming.
- Power BI — reports and semantic models.
- Data Activator — event-driven actions on data thresholds.
All run on shared Fabric capacity (F-SKUs), licensed by capacity rather than per-user for many features.
Semantic models
What were called datasets in Power BI are now semantic models in Fabric. A semantic model is the reusable definition of tables, relationships, measures, and calculations that one or many Power BI reports consume.
The naming change isn't cosmetic — Microsoft is pushing semantic models as first-class enterprise data products, not just per-report artefacts. Features include:
- Composite models — mix imported and DirectQuery data in one model.
- Calculation groups — reusable calculation patterns (time intelligence, currency conversion).
- Hybrid tables — recent data DirectQuery for freshness, historical data imported for performance.
- Object-level security — hide entire tables or columns from specific users.
- Row-level security (RLS) — filter rows per user role.
Direct Lake
Direct Lake is a Fabric-only mode for Power BI semantic models: instead of importing data (long refresh times, copy of data) or DirectQuery (live but slower), the semantic model reads OneLake Parquet files directly into memory on demand. The result is import-like performance with live data freshness, scaling to very large datasets.
For models built against Fabric Lakehouses or Warehouses, Direct Lake is the default and usually the right choice.
OneLake and shortcuts
OneLake is the unified data lake automatically provided by Fabric, organised hierarchically by tenant and workspace. Shortcuts let you mount external data (Azure Data Lake Storage, S3, Google Cloud Storage, Dataverse) into OneLake without copying — Fabric reads through them.
This makes Power BI semantic models source-agnostic: the lakehouse can be a mix of native Fabric data and external lakes, and the model doesn't care.
What changes for Power BI customers
If you're running Power BI Pro and Premium today:
- Power BI Pro stays a per-user licence for sharing and viewing.
- Premium Per User (PPU) continues for individual contributors needing premium features.
- Premium capacities (P-SKUs) are being replaced by Fabric capacities (F-SKUs) — same compute but consolidated billing and access to all Fabric workloads.
- New features are increasingly Fabric-only — Direct Lake, Data Activator, Copilot in Fabric.
For organisations with serious analytics ambition, planning a path to Fabric is increasingly the default. For lighter Power BI use, the existing per-user model still works.