Microsoft Fabric Managed Lakehouse: When to Use It

A Fabric Lakehouse has two storage areas: Tables and Files. Tables holds managed Delta tables — created via spark.sql CREATE TABLE or via the DataFrame saveAsTable() method. These appear automatically in the SQL analytics endpoint and are queryable by Power BI via Direct Lake. Files holds unstructured or semi-structured files — raw CSVs, Parquet files, JSON, images — and also Delta tables you want to manage manually without registering them in the catalog.

On top of these, Fabric Shortcuts let you mount external data — from ADLS, Amazon S3, Google Cloud Storage, or other Fabric Lakehouses — as a virtual path inside the Lakehouse. Shortcuts can point to either the Tables or Files section, with different behaviours in each case.

Managed Lakehouse tables are the right choice when: the table is the output of a transformation pipeline (Silver or Gold layer in medallion architecture), Power BI needs to read it via Direct Lake, and you want automatic Delta log management (VACUUM, OPTIMIZE) handled by Fabric. The SQL analytics endpoint exposes managed tables immediately with no additional configuration.

The trade-off for managed tables is that Fabric owns the storage location — it is inside the OneLake path for the specific Lakehouse workspace. If you delete the Lakehouse, the managed table data is deleted. If you need to share the same physical Delta table across multiple Lakehouses or workspaces without duplication, managed tables are not the right choice — you want Shortcuts or unmanaged external tables.

Shortcuts are the Fabric answer to "I have data in ADLS (or S3, or another Lakehouse) and I do not want to copy it." They create a virtual reference to external storage that appears as a folder inside the Lakehouse Files or Tables section. Shortcuts in the Files section behave like regular file paths — Spark can read them, but they do not appear in the SQL endpoint. Shortcuts in the Tables section appear in the SQL endpoint and can be queried via Direct Lake if they point to Delta-formatted data.

The operational constraint with Shortcuts is that Fabric Spark cannot write to a Shortcut destination unless you have write access to the source storage account. Read-only Shortcuts are the most common pattern — you bring external Bronze-layer data into scope without moving it, then transform it into managed Gold-layer tables within the Lakehouse.

The pattern that works cleanly in Fabric for a mid-size manufacturing or FMCG data platform: one Lakehouse per medallion tier, all within the same Fabric workspace. Bronze Lakehouse holds raw ingested data — Shortcuts pointing to ADLS landing zones for ERP extracts, IoT historian feeds, and file drops. Silver Lakehouse holds cleansed and conformed data — managed Delta tables written by Fabric notebooks running transformation logic. Gold Lakehouse holds business-ready aggregations — managed Delta tables optimised for Power BI Direct Lake.

With one workspace and three Lakehouses, the SQL analytics endpoint for the Gold Lakehouse is the single connection point for Power BI. You never expose Bronze or Silver data to Power BI report authors. Cross-Lakehouse Shortcuts allow the Silver transformation notebooks to read from Bronze without copying data.

Use managed Lakehouse tables when: the data is a transformation output (Silver or Gold), Power BI needs Direct Lake access, or you want Fabric to handle Delta maintenance operations automatically. Use Files-section Shortcuts when: you want to reference external raw data without copying it, and you do not need SQL endpoint access. Use Tables-section Shortcuts when: you want to expose external Delta data through the SQL endpoint without copying it, accepting that Fabric cannot run OPTIMIZE or VACUUM on it.

The one situation where none of these fit cleanly is when you need to share a Gold-layer table between two Fabric workspaces for different business units. The right pattern there is a cross-workspace Shortcut: Workspace A owns the Gold managed table; Workspace B creates a Tables-section Shortcut pointing to it. Both workspaces see the same physical Delta data.

In a live mid-market manufacturing or FMCG deployment, this framework collapses into a clean pattern. Bronze is a Lakehouse of Shortcuts pointing at the ADLS landing zones where ERP extracts, the historian feed, and file drops already sit — no copy, no second bill. Silver and Gold are managed Delta tables written by Fabric notebooks, with Gold optimised for Power BI Direct Lake. The Gold Lakehouse SQL analytics endpoint is the one connection point report authors ever touch; Bronze and Silver are never exposed to them.

The reason the managed-versus-Shortcut distinction matters in practice is Direct Lake. Power BI reads Gold managed Delta tables straight from OneLake at import speed without maintaining a separate import dataset — which is the whole performance case for building on Microsoft Fabric rather than a conventional warehouse. Get the storage choice wrong (a Files-section Shortcut where a managed table was needed) and the table silently never reaches the SQL endpoint, so the semantic model cannot see it. The framework exists precisely to avoid that dead end.

This is also where the medallion architecture stops being a diagram and becomes operational discipline: one workspace, three Lakehouses, cross-Lakehouse Shortcuts letting Silver read Bronze without duplication. It is the minimum structure that lets you evolve ingestion without breaking transformation or reporting — and it is what a governed data lakehouse looks like when it is built to be maintained by a small team rather than rebuilt every quarter.

The most common mistake is defaulting to managed tables for everything, including raw external data that should have stayed in place as a Shortcut. The result is duplicated storage, a second copy to keep in sync, and the cost surprise that follows. The rule of thumb: reference external and raw data with Shortcuts; reserve managed tables for the transformation outputs Power BI actually consumes.

The opposite error is over-using Tables-section Shortcuts for data that needs Delta maintenance. Fabric cannot run OPTIMIZE or VACUUM on a Shortcut it does not own, so a heavily-written Shortcut table degrades over time with no automatic compaction. If a table is yours and changes often, it should be a managed table where Fabric handles the Delta log.

And the honest scale limit: Direct Lake is not unconditional. There are per-table row and in-memory guardrails per capacity SKU — tables beyond them fall back to DirectQuery rather than failing, but the import-speed advantage goes with them. On very large fact tables you design for that fallback deliberately rather than discovering it when a dashboard slows down. Knowing the guardrail before you model is the difference between a fast estate and a surprised one.

The payoff of getting these choices right early is an estate that scales without re-platforming. Storage decisions made deliberately — managed where Power BI needs Direct Lake, Shortcuts where data should stay in place — mean new sources slot into Bronze without disturbing Gold, and the semantic model the business depends on stays stable. The framework is cheap to apply up front and expensive to retrofit after the pipelines are live.

It is also the foundation everything else in the estate sits on. The same governed Gold layer that serves today's Power BI dashboards is what manufacturing analytics, predictive analytics, and eventually the Fabric Data Agent read from — one data lakehouse, reused, rather than a fresh integration per initiative.

If you are standing up a new Fabric Lakehouse or rationalising one that grew organically, the highest-value first step is agreeing this storage framework before more pipelines are built on guesswork. A short Discover and Foundation engagement sets the medallion structure and the managed-versus-Shortcut conventions in weeks — first value in 6 weeks — so the platform compounds instead of accumulating the duplication and dead-end tables that a retrofit has to unwind.