What Is OneLake?
OneLake is the single, unified data lake built into Microsoft Fabric. Every Fabric tenant gets one OneLake automatically, and every Fabric workload, from data engineering to warehousing to Power BI, reads and writes to it. The idea is one logical lake for the whole organization, with data stored once in open Delta Parquet format rather than copied into a separate silo for each tool.
Microsoft describes OneLake as the OneDrive for data: a single, governed place where an organization’s analytics data lives, so teams stop maintaining a separate copy for every engine. For anyone evaluating Microsoft Fabric as the foundation for analytics, OneLake is the part that matters most, because it shapes how data is stored, shared, and governed across everything else.
The “OneDrive for Data” Idea
The analogy Microsoft uses is deliberate. Just as OneDrive gives an organization one place to store documents that any Office application can open, OneLake gives an organization one place to store analytics data that any Fabric engine can use. There is one lake per tenant, provisioned automatically, with no infrastructure to set up.
The problem this addresses is real. In many organizations, the same data ends up copied into a warehouse for SQL, a separate lake for data science, and yet another extract for a BI tool. Each copy drifts, each needs its own governance, and reconciling them becomes a standing tax. OneLake’s premise is that the data sits in one place, in one open format, and the engines come to it.
How OneLake Works: One Lake, Many Workloads
Under the hood, OneLake is built on Azure Data Lake Storage Gen2 and stores tabular data in Delta Parquet, an open format. Inside OneLake, data is organized into workspaces and items such as lakehouses and warehouses. Because the storage format is open and shared, a table written by a Fabric data engineering job can be queried by the Fabric warehouse engine and surfaced in Power BI without anyone exporting or duplicating it.
This is the structural difference from a traditional stack. Rather than each tool owning its own storage, the storage is common and the compute engines are layered on top. One copy of the data serves many workloads.
OneLake and the Lakehouse Model
OneLake is Microsoft’s implementation of the lakehouse idea: open storage that supports both the large-scale data engineering of a data lake and the structured, governed querying of a data warehouse, on one set of files. A lakehouse item in Fabric organizes data into the familiar medallion layers, where raw data is progressively cleaned and modeled into curated tables ready for reporting.
If you have read our entries on the data lakehouse and lakehouse architecture, OneLake is one concrete way to realize that pattern. The architectural goal is the same: open formats, one governed copy, and the flexibility to run different engines against it.
Shortcuts: Referencing Data Without Copying It
One of OneLake’s more useful features is the shortcut. A shortcut lets OneLake reference data that physically lives somewhere else, such as Azure Data Lake Storage, Amazon S3, or another Fabric workspace, and present it as if it were local, without making a copy. Queries read the source data in place.
This matters for organizations that cannot or do not want to move everything into one physical store. Shortcuts let teams bring existing data into the OneLake experience without a migration project, which softens one of the harder parts of consolidating onto a new platform.
Direct Lake and Power BI
Because Power BI is part of Fabric, it reads directly from OneLake through a mode called Direct Lake. Direct Lake lets a Power BI semantic model query the Delta tables in OneLake without importing the data into the model and without the query-time round trip of DirectQuery. For the analyst, it aims to combine the speed of imported data with the freshness of querying the source.
Direct Lake is now generally available with row-level and column-level security enforced through OneLake, which removes a governance gap that existed in earlier versions. A well-built Power BI semantic model remains the layer where business logic lives, whether the data is reached through Direct Lake or any other mode.
OneLake vs a Databricks Lakehouse
A common question is how OneLake compares to a lakehouse built on Databricks. Both store data in open formats, both follow the lakehouse model, and both can serve as the foundation for governed analytics and Power BI. The right choice is workload-dependent, not a matter of one being universally better.
Fabric, with OneLake at its center, tends to fit organizations already invested in Microsoft 365 and Power BI, where capacity-based pricing and tight Power BI integration are advantages, and where utilization is steady. Databricks tends to fit organizations with heavier or more variable data engineering and data science needs, where its consumption model and mature tooling work in their favor. The older assumption that one is always cheaper no longer holds; cost depends on the shape of the workload.
Teams often ask which platform is the right foundation, and the honest answer is that it depends on the work they actually run. The architecture we deliver, the medallion method, curated views, and the Power BI semantic models on top, looks the same on either one. What differs sits underneath, in how each platform prices compute and handles ingestion, and that is where the choice tends to come down to the specific workload.
Marla Nelson, CTO, QuickLaunch Analytics
Where OneLake Fits in a Data Foundation
OneLake is storage and governance, not the finished analytics. Getting value from it still requires the work above it: pipelines that bring source data in, a modeled lakehouse with clean curated layers, and semantic models that turn tables into business terms an analyst can trust. OneLake makes that work cleaner by giving it one place to land, but it does not do the modeling.
This is where a pre-built foundation earns its keep. The patterns that make OneLake useful, governed layers, curated views, and well-formed semantic models, are the same patterns whether the lake is OneLake or a Databricks lakehouse, which is why the choice of platform does not have to dictate the analytics experience.
Frequently Asked Questions
Is OneLake the same as Microsoft Fabric?
No. Microsoft Fabric is the overall analytics platform; OneLake is the single unified data lake built into it. Every Fabric tenant has one OneLake, and all Fabric workloads read and write to it. OneLake is the storage and governance layer at the center of Fabric, not the whole platform.
What format does OneLake store data in?
OneLake stores tabular data in Delta Parquet, an open format, on top of Azure Data Lake Storage Gen2. Because the format is open and shared, multiple Fabric engines can read and write the same tables without exporting or duplicating the data.
What is a OneLake shortcut?
A shortcut lets OneLake reference data that lives elsewhere, such as Azure Data Lake Storage, Amazon S3, or another workspace, and present it as if it were local, without copying it. Queries read the source data in place, which lets teams bring existing data into Fabric without a migration.
Should I choose OneLake or Databricks?
It depends on the workload. Fabric with OneLake often fits organizations invested in Microsoft 365 and Power BI with steady utilization; Databricks often fits heavier or more variable engineering and data science needs. The analytics experience and architecture can be delivered on either, so the decision usually comes down to existing investment, workload shape, and cost profile rather than a universal winner.
OneLake and QuickLaunch’s Approach
QuickLaunch Analytics delivers the same governed foundation, automated pipelines, a governed lakehouse, curated views, and Power BI semantic models, on either Microsoft Fabric with OneLake or Databricks. We recommend the platform that fits the customer’s workload and existing investment, consultatively, rather than pushing one by default. The result is the same analyst experience and the same 8 to 12 week path to a working foundation, built on patterns refined across 250+ enterprise implementations.
