For years, the Microsoft data story was a shopping list: Synapse, ADF, Power BI, Purview, ADLS, Databricks-or-not. Microsoft Fabric and OneLake collapse most of that into a single SaaS platform — and in 2026, it is finally mature enough to bet a production workload on.
Here is what actually changed and what to watch.
The core idea
One tenant, one lake. OneLake is the OneDrive for data: a single logical data lake across your entire org, built on Delta/Parquet, accessible from every Fabric workload (Data Engineering, Data Warehouse, Real-Time Intelligence, Data Science, Power BI) without copying data.
Shortcuts let you point at data in ADLS, S3, or GCS and treat it as if it lived in OneLake. No duplication.
Why it matters
- No more pipeline jungle just to get data into Power BI
- Direct Lake mode in Power BI queries Parquet files directly — no import, no DirectQuery tax
- Git integration for notebooks, pipelines, and semantic models
- Unified governance via Purview baked in
- One compute metering model (capacity units) across all workloads
Where it is great
- Organizations already standardized on Power BI
- Mid-size data estates that never wanted to operate Databricks themselves
- Teams that want medallion architecture without assembling six services
Where it is not yet the answer
- Very large, very custom Spark workloads (Databricks still wins)
- Teams with strict multi-cloud strategies (Fabric is Azure-first)
- Workloads that need streaming semantics beyond what KQL and Eventstream provide
Migration tips
- Start with one domain, not the whole estate
- Use shortcuts before you move data — prove the queries work first
- Adopt medallion layout in OneLake from day one (Bronze / Silver / Gold)
- Bring Git integration in early — it is the biggest quality-of-life upgrade
Originally published on the Horizon Tech Blog.
Top comments (0)