What Is Microsoft Fabric? Workloads, OneLake, Architecture and Pricing Explained

Microsoft Fabric is a data analytics platform delivered as software as a service (SaaS), with no servers to set up. According to Microsoft's documentation, it covers everything from raw data to report, with several tools that share compute and storage.

Three ideas are enough to understand it:

• One storage layer. Everything stored in Fabric goes into OneLake, a single data lake for the whole organization.
• Several tools, one per role. Data engineering, data warehousing, data science, real-time analytics and Power BI work on that same data, without copying it from one to another.
• One meter. You pay for a capacity and every tool draws from it, instead of a bill per service.

OneLake is what makes Fabric a platform rather than a collection of tools. Each organization gets a single OneLake, provisioned automatically on Azure Data Lake Storage Gen2. Microsoft compares it to OneDrive: just as Word saves to your company OneDrive, every Fabric tool saves to OneLake.

It is organized into workspaces, like folders, holding items: lakehouses, warehouses, semantic models. Tables are stored in Delta format by default (Parquet files plus a transaction log), so Spark, SQL and Power BI read the same data without converting it.

Not everything has to be copied into Fabric. OneLake shortcuts show up as just another folder but point somewhere else: Azure Data Lake Storage Gen2, Azure Blob Storage, Amazon S3, Google Cloud Storage, Dataverse, Iceberg tables, OneDrive and SharePoint, or on-premises storage through the data gateway. Deleting a shortcut leaves the source untouched, and an optional cache of 1 to 28 days makes reading from another cloud cheaper.

Mirroring is the other route: it continuously copies a whole database into OneLake. It fits when you don't want analytical queries hitting the operational database.

It is the design Microsoft recommends for Fabric in its medallion architecture guide: each layer in its own lakehouse or warehouse, ideally in its own workspace, with silver and gold in Delta tables. It suggests two patterns:

• All lakehouses, with the business querying gold through its SQL analytics endpoint.
• Bronze and silver as lakehouses and gold as a warehouse, for teams more at home with T-SQL.

A lakehouse's SQL analytics endpoint is for querying, not writing; a warehouse supports full T-SQL. The reasoning behind the three layers, with a sales and marketing example, is in data warehouse vs data lake vs lakehouse.

Outside Fabric, a Power BI model either imports data, with a copy refreshed on a schedule, or queries the source live (DirectQuery). Inside Fabric there is a third way, Direct Lake: the model reads the Delta tables in OneLake directly, without the import copy. The report shows whatever was last loaded into gold, and data cleaning no longer lives inside each Power BI file.

Fabric is paid for by capacity: an F2 has 2 capacity units (CUs) and an F64 has 64, and every workload draws from it. You pay as you go (per second, one-minute minimum, and it can be paused) or reserve it for a year, about 41% cheaper according to Microsoft.

Prices in the West Europe region (Netherlands), in US dollars and before tax, from the Microsoft Fabric pricing page on Azure as of October 2026, with a month counted as 730 hours:

• F2: $0.44 an hour, about $321 a month pay-as-you-go; $191 a month with a one-year reservation.
• F8: $1.76 an hour, about $1,285 a month; $764 a month reserved.
• F64: $14.08 an hour, about $10,278 a month; $6,112 a month reserved.

OneLake storage is billed separately: $0.024 per GB per month in West Europe for standard storage. The full table, with Spain Central, is on our Microsoft Fabric consulting page.

Power BI licences still count

According to the Fabric licensing guide, below F64 everyone who views Power BI content needs their own Pro or PPU licence. From F64 up, people who only view can do so with a free Fabric licence and a viewer role. Anyone who creates and publishes reports needs Pro either way. So the size depends on how many people view reports.

The free trial

The Fabric trial lasts 60 days, with a capacity of 4 or 64 CUs and up to 1 TB in OneLake. It doesn't include Copilot or data agents, and it is for evaluation, not production.

«Fabric or Power BI?» is a trick question: Power BI is one of Fabric's workloads.

With Power BI alone, each model connects to its sources and cleans them with Power Query on its own, on Pro or PPU licences and no capacity. For many teams that is enough. What it doesn't give you is a place to keep prepared data for other models, other teams or a Python model.

Fabric adds that place: lakehouses, warehouses, pipelines and notebooks on OneLake. Creating them takes an F capacity; a Premium Per User (PPU) licence improves Power BI but doesn't cover the rest.