What actually changes
A conventional Microsoft data platform keeps a separate copy of the data for each engine: files in a storage account for Spark, tables in a SQL warehouse for reporting, and an imported copy inside every Power BI semantic model. The Fabric lakehouse model removes most of those copies. Microsoft describes OneLake as a unified data lake for the whole organization, automatically included with every Fabric tenant, built on Azure Data Lake Storage, and storing tables in Delta Parquet or Iceberg format so that every Fabric engine works on the same data without duplication (Microsoft Learn, What is OneLake?, accessed 2026-10-20). EPC Group's Microsoft Fabric consulting practice treats that as the starting point for every design: decide where each table lives once, choose the engine that writes it, and let every other workload read it in place. Four practical changes follow, covering storage, item choice, layering and the semantic layer, and the migration path is the fifth.
OneLake: one lake per tenant, organized by workspace
OneLake is provisioned for the tenant, not per project. A tenant has exactly one OneLake instance that cannot be deleted, data is organized by tenant, workspace and item, and each workspace belongs to a capacity tied to a specific region. Shortcuts reference data in other workspaces or in Azure Data Lake Storage, Azure Blob Storage, Amazon S3, Dataverse and on-premises sources so that it appears local without being copied, and mirroring replicates operational databases into OneLake as Delta tables. OneLake security roles can restrict access down to folders, tables, rows and columns and are enforced whether a user queries through SQL, Spark or a Power BI report. The storage itself uses zone-redundant storage where availability zones exist, and soft delete keeps deleted files for seven days (Microsoft Learn, What is OneLake?, accessed 2026-10-20). For architects, the change is that the workspace becomes the storage boundary, so EPC Group designs the workspace map before the first pipeline is built.
Lakehouse and warehouse: two items on the same storage
Fabric offers two data stores for analytics, and both write Delta tables to OneLake. Microsoft's decision guide points to a lakehouse when the team develops in Apache Spark, when data includes unstructured files, or when multi-table transactions are not needed, and to a warehouse when the team develops in T-SQL, needs multi-table ACID transactions, and works with structured data. The lakehouse automatically provides a read-only SQL analytics endpoint with full DQL, no DML and limited DDL such as views and table-valued functions; the warehouse supports full DQL, DML and DDL, including COPY INTO, INSERT and CREATE TABLE AS SELECT (Microsoft Learn, Microsoft Fabric decision guide: Choose between Warehouse and Lakehouse, accessed 2026-10-20). A lakehouse organizes data into a Tables folder for managed Delta tables, which Fabric discovers and registers automatically, and a Files folder for everything else; only Delta tables appear in the SQL analytics endpoint (Microsoft Learn, What is a lakehouse in Microsoft Fabric?, accessed 2026-10-20).
Medallion layers in separate workspaces
Microsoft calls the medallion lakehouse architecture the recommended design approach for Fabric. Bronze stores data exactly as it arrives, silver fixes errors, standardizes formats and removes duplicates, and gold organizes data for reports and dashboards. Two deployment patterns are documented: every layer as a lakehouse, or bronze and silver as lakehouses with gold as a warehouse. Microsoft recommends a separate workspace for each layer for better governance, shortcuts in bronze instead of copies when the source is already in OneLake, ADLS Gen2 or Amazon S3, Delta tables with V-order optimization in silver and gold, target file sizes between 128 MB and 1 GB, and Liquid Clustering rather than partitioning in the upper layers. VACUUM cannot remove history from the last seven days by default, and materialized lake views can define the transformations between layers in SQL with automatic dependency management (Microsoft Learn, Understand medallion architecture for Fabric with OneLake, accessed 2026-10-20). EPC Group adds Fabric domains above the layers, so each business domain owns its own bronze, silver and gold workspaces.
Direct Lake and the semantic layer
The largest change for Power BI teams is Direct Lake. It is a semantic model table storage mode that loads Delta table data from OneLake into memory and answers DAX queries with the same VertiPaq engine as Import, while a refresh only reframes metadata instead of copying the data. Microsoft describes it as an ideal choice for the gold layer and for IT-driven, lake-centric projects, and notes that Import remains right for analysts who need Power Query agility. Direct Lake on OneLake can use several Fabric sources and does not fall back to DirectQuery; Direct Lake on SQL uses a single source and falls back to DirectQuery for SQL views or SQL-based granular access control. Direct Lake requires a Fabric capacity, guardrails rise with the SKU, gateways are not supported, and Microsoft recommends a fixed identity connection when semantic model RLS is used (Microsoft Learn, Direct Lake overview, accessed 2026-10-20). EPC Group's Power BI consulting team rebuilds high-volume Import models on Direct Lake only after the gold tables pass a proof of concept.
Running Fabric and Azure Databricks together
Many estates keep Azure Databricks for data engineering and add Fabric for Power BI and business analytics. Microsoft supports that pattern directly: a mirrored Azure Databricks Unity Catalog in Fabric lets Fabric workloads read data that Unity Catalog manages, with no data movement or data replication. Only the catalog structure is mirrored, the underlying data is read through shortcuts, and Fabric creates a SQL analytics endpoint plus a mirrored item that Power BI can use in Direct Lake mode. Changes to the underlying data can take from a few seconds to several minutes to appear in Fabric (Microsoft Learn,Mirrored catalog from Azure Databricks, accessed 2026-10-28).
OneLake shortcuts do the same job for storage outside Fabric: they behave like symbolic links to Azure Data Lake Storage, Amazon S3, Google Cloud Storage and other sources, so Fabric engines read the data in place instead of copying it, and a Direct Lake semantic model can read a Delta table through a shortcut (Microsoft Learn,OneLake shortcuts, accessed 2026-10-28).
- Keep Databricks where the engineering, machine learning and Unity Catalog governance already live.
- Add Fabric for Power BI semantic models, reports and business-facing SQL without a second copy of the data.
- Plan for lag and filters: mirrored catalogs skip materialized views, streaming tables and non-Delta external tables, so check which tables actually appear.
- Decide identity early: with Direct Lake over SQL in delegated mode the item owner's identity reaches the shortcut target; Direct Lake over OneLake passes the user's identity.
Moving an existing SQL and Power BI estate
Migration follows Microsoft's five-pillar lifecycle for Azure Synapse dedicated SQL pools: Assess and Evaluate, Plan and Design, Migrate, Monitor and Govern, and Optimize and Modernize. The runbook calls for starting small with multiple small migrations, inventorying security and permissions, building parallel ETL processes, running old and new side by side, and moving reporting only when the business is comfortable; it also lists data type changes such as datetime to datetime2 and nvarchar to varchar (Microsoft Learn, Migration planning: Azure Synapse Analytics dedicated SQL pools to Fabric Data Warehouse, accessed 2026-10-20). The Fabric Migration Assistant migrates schemas from a DACPAC, script files or a direct connection, prioritizes broken primary objects ahead of dependent ones, copies data with a copy job, and reroutes connections at the end; SQL authentication users must be replaced with Microsoft Entra users (Microsoft Learn, Fabric Migration Assistant for Data Warehouse, accessed 2026-10-20). On the Power BI side, Microsoft recommends OneLake integration as a quick first step: it writes the tables of existing Import semantic models to Delta tables in OneLake without migration effort, so lakehouses, SQL queries and notebooks can use them before the warehouse itself is redesigned (Microsoft Learn, Direct Lake overview, accessed 2026-10-20). EPC Group sequences estates in that order: integration first, then warehouse migration, then Direct Lake.
Documented work
ANON-CASES: EPC Group engagement data for data platform modernization covers mid-size and large manufacturing and healthcare organizations moving SQL Server warehouses and Import semantic models onto OneLake lakehouses, medallion workspaces and Direct Lake, delivered by the firm's Microsoft Fabric consulting practice. Client names are not published here. Engagements are described by industry and organization size only, and no outcome figures are quoted that cannot be traced to a signed deliverable. EPC Group is a Microsoft Solutions Partner with all six Solutions Partner Designations as of October 2026.
Primary sources
- What is OneLake? — Microsoft Learn, accessed 2026-10-20
- Microsoft Fabric decision guide: Choose between Warehouse and Lakehouse — Microsoft Learn, accessed 2026-10-20
- What is a lakehouse in Microsoft Fabric? — Microsoft Learn, accessed 2026-10-20
- Understand medallion architecture for Fabric with OneLake — Microsoft Learn, accessed 2026-10-20
- Direct Lake overview — Microsoft Learn, accessed 2026-10-20
- Migration planning: Azure Synapse Analytics dedicated SQL pools to Fabric Data Warehouse — Microsoft Learn, accessed 2026-10-20
- Fabric Migration Assistant for Data Warehouse — Microsoft Learn, accessed 2026-10-20
Frequently Asked Questions
What is OneLake, and how does it differ from a data lake in Azure Storage?
OneLake is the single data lake that every Fabric tenant includes automatically; it cannot be deleted, a tenant cannot have more than one, and there is no infrastructure to provision. It is built on Azure Data Lake Storage, stores tables in Delta Parquet or Iceberg format, organizes data by workspace and item, and supports ADLS Gen2 APIs, so each workspace appears as a container (Microsoft Learn, What is OneLake?, accessed 2026-10-20). The practical difference from a self-managed storage account is that governance, security roles and every Fabric engine share the same copy, and shortcuts can reference existing ADLS, Blob Storage or Amazon S3 data without moving it (same source).
Lakehouse or warehouse: how do you choose?
Microsoft gives three decision points: develop in Apache Spark and you choose a lakehouse, develop in T-SQL and you choose a warehouse; need multi-table transactions and you choose a warehouse; analyze unstructured as well as structured data and you choose a lakehouse. Both store Delta tables in OneLake and share the same SQL engine; the lakehouse SQL analytics endpoint is read-only with full DQL, no DML and limited DDL, while the warehouse supports full DQL, DML and DDL with transactions (Microsoft Learn, Microsoft Fabric decision guide: Choose between Warehouse and Lakehouse, accessed 2026-10-20). EPC Group commonly lands and refines data in lakehouses and serves curated star schemas from a warehouse.
What are bronze, silver and gold layers in practice?
Microsoft calls the medallion architecture the recommended design approach for Fabric: bronze stores raw data exactly as it arrives, silver fixes errors, standardizes formats and removes duplicates, and gold organizes data for reports and dashboards. Microsoft recommends keeping each layer in its own lakehouse or warehouse, ideally in a separate workspace for layer-level governance; using shortcuts in bronze when data already sits in OneLake, ADLS Gen2 or Amazon S3; and Delta tables in silver and gold. Materialized lake views can manage the transformations between layers declaratively in SQL (Microsoft Learn, Understand medallion architecture for Fabric with OneLake, accessed 2026-10-20).
How do Power BI semantic models connect to the lakehouse (Direct Lake)?
Direct Lake is a semantic model table storage mode that loads Delta table data from OneLake into memory and answers queries with the VertiPaq engine, so performance is comparable to Import without full data refresh cycles; a Direct Lake refresh copies only metadata, a process called framing. Direct Lake on OneLake can combine several Fabric sources and does not fall back to DirectQuery, while Direct Lake on SQL uses one source and falls back when it reads a SQL view. It requires Fabric capacity, and guardrails scale by SKU, for example 1,500 million rows per table on F64 (Microsoft Learn, Direct Lake overview, accessed 2026-10-20).
How is a migration from Synapse or SQL Server sequenced?
Microsoft structures the move from Azure Synapse dedicated SQL pools into five pillars: Assess and Evaluate, Plan and Design, Migrate, Monitor and Govern, and Optimize and Modernize, with parallel running and comparison against the source before applications and reporting move (Microsoft Learn, Migration planning: Azure Synapse Analytics dedicated SQL pools to Fabric Data Warehouse, accessed 2026-10-20). The Fabric Migration Assistant handles Synapse dedicated SQL pools, SQL Server and other SQL Database Engine platforms: it migrates schemas, guides fixes for objects that fail, copies data with a copy job, then supports testing and rerouting connections (Microsoft Learn, Fabric Migration Assistant for Data Warehouse, accessed 2026-10-20).
Talk to a senior architect
If your platform still keeps one copy of the data per engine, or your Power BI models import the same warehouse tables again and again, a 30-minute discovery call with a senior EPC Group architect maps your estate to OneLake, medallion workspaces and Direct Lake, and sequence the move. See the Microsoft Fabric consulting service hub for engagement structures.
Email contact@epcgroup.net or call 888-381-9725.
a Microsoft Gold Partner (2000 until Microsoft retired the program in 2022) — today holding all six Microsoft Solutions Partner Designations.