Skip to content
QuorlyticTalk to a Data ExpertContact
Menu

Technology / Microsoft Fabric

One lake, one capacity, and Power BI built in.

Microsoft Fabric is Microsoft's software-as-a-service analytics platform on Azure. Data Factory pipelines, Spark data engineering, a T-SQL warehouse, real-time analytics and Power BI share one copy of data in OneLake and draw on one capacity of compute units. Quorlytic designs the lake, sizes the capacity and decides which workloads belong on Fabric and which do not.

By QuorlyticUpdated

Fabric changes three decisions at once: where data is stored, how compute is bought and how Power BI reads data. Get the lake layout, the capacity plan and the semantic layer right first, and the rest of the platform follows from them.

Architecture role

Where Microsoft Fabric fits.

Fabric puts pipelines, Spark, a T-SQL warehouse, real-time analytics and Power BI over one copy of data in OneLake, paid for as one capacity. The architecture decisions are the lake and workspace layout, the capacity size and which workloads belong on Fabric at all.

Position in the end-to-end path

  1. 01Business sourcesDatabases, files, SaaS, events
  2. 02Bring inPipelines, mirroring and shortcuts
  3. 03OneLakeLakehouse, warehouse, eventhouse
  4. 04Semantic layerDirect Lake semantic models
  5. 05Reporting & actionPower BI, Excel and Copilot

In

  • MirroringSQL, Snowflake, Databricks catalog
  • ShortcutsADLS, Amazon S3, Dataverse
  • Pipelines · EventstreamWhen data must be moved

OneLake

One copy · Delta Parquet or Iceberg

  1. Lakehouse · Spark
  2. Warehouse · T-SQL
  3. Eventhouse · KQL

OneLake security · Purview

Out

  • Power BIDirect Lake
  • Excel
  • Copilot · data agents
  • Notebooks · ML

One capacity pays for every workload — size it from measured use, not from licence counts

Microsoft Fabric data platform: operational databases and catalogs arrive by mirroring, external storage by shortcuts and other sources by pipelines and Eventstream; every workload, including the lakehouse, warehouse and eventhouse, reads one copy of data in OneLake under OneLake security; Power BI reads it in Direct Lake mode alongside Excel, Copilot and notebooks, all drawing on one shared capacity.

Reference architecture. It shows a typical position for discussion, not a mandatory stack or a client implementation.

Fit and capability

When it is relevant, and what the work involves.

Relevant workloads

  • Lakehouse and warehouse analytics
  • Pipeline and integration
  • Real-time analytics
  • Semantic models and reporting

Engineering capabilities

  • Workload placement decision across Fabric, Azure and Databricks
  • OneLake, workspace and domain design
  • Capacity sizing and licence plan from measured usage
  • Direct Lake semantic model and refresh design

Integration points

What it connects to.

The platform is only as useful as the boundaries around it. These are the interfaces that usually need an explicit decision.

  • 01Entra ID and OneLake security
  • 02Purview governance and lineage
  • 03Mirroring from Databricks, Snowflake and SQL
  • 04Capacity metrics and sizing

Delivery and operating considerations

What to decide before you commit.

Each consideration changes something concrete in the architecture, the operating model or the cost profile.

  1. 01

    What Microsoft Fabric is

    Fabric brings workloads that used to be separate Azure services into one product, billed as one capacity and stored in one lake. Every workload reads and writes the same Delta tables in OneLake, so a table loaded by a pipeline can be queried in T-SQL, processed in Spark and reported in Power BI without being copied.

    • Data Factory: pipelines, dataflows, copy jobs and mirroring from operational databases
    • Data Engineering: lakehouses and Spark notebooks
    • Data Warehouse: a transactional T-SQL warehouse over the same lake
    • Real-Time Intelligence: Eventstream, Eventhouse and Activator for streaming data
    • Data Science: notebooks and model training next to the data
    • Power BI: semantic models, reports and dashboards, now one workload among several
  2. 02

    OneLake: one lake per tenant, one copy of data

    OneLake is the storage under every Fabric workload. Each Fabric tenant has exactly one, built on Azure Data Lake Storage, holding tables in the open Delta Parquet or Iceberg formats. Shortcuts point at data that lives elsewhere and mirroring brings external databases and catalogs in, so most data reaches Fabric without a pipeline that copies it.

    Shortcuts reference data in other workspaces, Azure Data Lake Storage, Amazon S3 and S3-compatible stores, Iceberg sources, Dataverse and on-premises locations, and changes at the source are visible immediately. Mirroring comes in two kinds: database mirroring continuously replicates an operational database into Delta tables, and metadata mirroring, used for catalogs such as Azure Databricks Unity Catalog, copies only the catalog structure and reads the data in place.

    OneLake security roles grant access down to folders, tables, rows and columns, and the same rule applies whether someone queries in SQL, runs a Spark notebook or opens a Power BI report. That single enforcement point is the main governance argument for Fabric, and it only works if workspaces and domains are designed before data arrives.

  3. 03

    Direct Lake changes how Power BI reads data

    In Direct Lake mode a Power BI semantic model reads Delta tables in OneLake directly, without importing a copy on a refresh schedule and without sending every query back to the source. Direct Lake on OneLake is generally available, and for large models over well-modelled lakehouse tables it removes most of the refresh window.

    It is not a reason to skip modelling. Direct Lake reads whatever tables it is pointed at, so a star schema built in the lakehouse or warehouse is still the work that makes reports fast and consistent. Import mode remains the better choice for small models and for logic that only exists in Power Query. The semantic models guide covers where measures and definitions should live, and Power BI consulting covers the reporting side.

  4. 04

    Capacity: one pool of compute for every workload

    Fabric is bought as a capacity of compute units, from F2 (2 CUs) to F8192. Every workload in the workspaces assigned to it draws from that pool. Pay-as-you-go is billed per second at $0.18 per CU-hour in East US and $0.20 in Central India, and an F64 costs $8,409.60 a month pay-as-you-go or $5,002.67 a month on a one-year reservation.

    Fabric smooths usage instead of failing at the first spike. Interactive operations are spread over at least five minutes and up to 64, and background operations such as refreshes and pipelines over 24 hours. A capacity can run past 100% for ten minutes of future capacity before anything slows; beyond that, new interactive requests are delayed by 20 seconds, then rejected after an hour of overload, and after 24 hours every new request is rejected until the debt is paid down.

    That makes sizing a measurement problem. Start smaller than feels comfortable, watch the Capacity Metrics app through a real month-end, and scale or split capacities from evidence. Overage billing exists but costs three times the normal rate, and pausing a capacity bills its accumulated usage at once. Attributing and reducing that spend is the subject of data platform cost optimization.

  5. 05

    Which licences people need

    A capacity does not license people. Anyone who publishes Power BI content needs Power BI Pro ($14 a user a month) or Premium Per User ($24). On F64 and larger, colleagues with only a free licence and a viewer role can read reports; below F64, every viewer needs Pro or Premium Per User too. Premium Per User does not include a Fabric capacity.

    The F64 threshold is where licensing and architecture meet. At list prices, an F64 on a one-year reservation costs less than Pro licences for everyone once more than about 357 people need to read reports, and the same capacity also runs the engineering workloads. Below that, a smaller F SKU with per-user Pro licences is usually cheaper.

  6. 06

    When Fabric is the right platform, and when it is not

    Fabric fits organisations that already run on Microsoft 365 and Azure, where Power BI is how most people consume data and a small team wants one managed platform rather than several services to integrate. It fits less well when workloads must run on more than one cloud, when Spark and machine learning engineering need deep control, or when every team needs its own isolated compute bill.

    The usual mistake is treating Fabric as all or nothing. Plenty of estates keep Azure Databricks for engineering and machine learning and use Fabric for Power BI and the business-facing warehouse, connected by mirroring instead of copies. The Microsoft Fabric vs Databricks comparison sets out when each wins and when both together is the honest answer.

  7. 07

    Moving to Fabric from Synapse and Power BI Premium

    Two migrations bring most organisations to Fabric. Power BI Premium P capacities are being retired in favour of F capacities, so a P1 renewal becomes an F64 decision. Azure Synapse Analytics has no published end-of-life for its dedicated SQL, Spark or pipeline services, so moving it is a choice to make on workload evidence rather than a deadline.

    Microsoft provides migration assistants for Synapse Spark (notebooks, job definitions, pools and lake database metadata) and for dedicated SQL pools (schema conversion and data copy), and a path for upgrading Synapse pipelines into Fabric Data Factory. None of them completes a migration on its own: configuration, libraries, security and the reports downstream still need proving. The reconciliation and cutover method is the same as any warehouse move, described in data modernization.

Typical delivery sequence

  1. 01Estate & licence review
  2. 02Lake & workspace design
  3. 03Build & migrate workloads
  4. 04Size, govern & operate

Decision table

When Microsoft Fabric fits, and when to compare.

Microsoft Fabric decision table: conditions where it is a strong fit, and conditions that call for comparing alternatives
DimensionMicrosoft Fabric is a strong fitCompare alternatives when
EstateThe organisation runs on Microsoft 365 and Azure.Workloads must run on AWS, Google Cloud or several clouds.
ConsumptionPower BI is how most people read data, ideally in Direct Lake mode.Most consumers use other BI tools or applications.
LicensingA Power BI Premium P renewal is due, or more than a few hundred people read reports.A handful of users read reports and Pro licences cost less than a capacity.
EngineeringPipelines, SQL and moderate Spark that a small team can run.Heavy Spark tuning or production machine learning is the core of the work.
Cost modelOne predictable capacity shared across workloads suits the budget.Every team needs its own isolated, separately billed compute.
GovernanceEntra ID, Purview and sensitivity labels already govern the estate.Governance must be identical across clouds, for example through Unity Catalog.

Questions

Questions worth asking.

What is Microsoft Fabric?

Microsoft Fabric is a software-as-a-service analytics platform on Azure. It combines data integration, Spark data engineering, a T-SQL data warehouse, real-time analytics, data science and Power BI in one product, with every workload sharing one copy of data in OneLake and one capacity of compute.

What is OneLake?

OneLake is the single data lake that comes with every Microsoft Fabric tenant. It is built on Azure Data Lake Storage, stores tables in open Delta Parquet or Iceberg formats, and lets every Fabric engine use the same copy of data. Shortcuts and mirroring bring in data held elsewhere, often without copying it.

Is Microsoft Fabric replacing Azure Synapse Analytics?

Microsoft directs new analytics work to Fabric, but it has not published an end-of-life date for Synapse dedicated SQL pools, Spark or pipelines. Migration assistants exist for Synapse Spark and dedicated SQL pools. Move when the workload evidence supports it, not because of an assumed deadline.

Do we need Microsoft Fabric to use Power BI?

No. Power BI Pro and Premium Per User work without a Fabric capacity. Fabric becomes relevant when you need the capacity features, want free-licence viewers on F64 or larger, or want to store and engineer data in the same platform your reports read from.

What is an F64 capacity?

F64 is the Fabric capacity size with 64 compute units, equivalent in compute to the retiring Power BI Premium P1. It is the smallest size on which users with a free licence and a viewer role can read Power BI reports. At list price it costs $8,409.60 a month pay-as-you-go in East US, or $5,002.67 a month reserved for a year.

Should we choose Microsoft Fabric or Databricks?

Fabric suits Microsoft-centred estates where Power BI is the main consumer and a small team wants one managed platform. Databricks suits multi-cloud estates and deep Spark and machine learning engineering. Many organisations use both, with Databricks for engineering and Fabric for Power BI, joined by mirroring.

Integration ecosystem

What sits around Microsoft Fabric.

Adjacent technologies grouped by the role they play next to this platform, and why that role usually needs an explicit decision.

  • Bring data in

    • Data Factory pipelines
    • Copy jobs
    • Database mirroring
    • Eventstream

    Mirroring and shortcuts often replace pipelines that only copy data, so the first design question is what still needs a pipeline at all.

  • Storage

    • OneLake
    • Delta Parquet
    • Iceberg
    • Shortcuts

    One lake per tenant, organised by domains and workspaces. The layout decides governance, so it comes before the first table.

  • Processing

    • Spark notebooks
    • T-SQL warehouse
    • Dataflows
    • Eventhouse

    Several engines can do the same job. Pick one per workload, by team skill and by how the result is consumed.

  • Consumption

    • Power BI Direct Lake
    • Excel
    • Copilot and data agents
    • Applications

    Direct Lake is the main reason to put Power BI data in OneLake, and it still depends on well-modelled tables.

  • Governance and cost

    • OneLake security
    • Microsoft Purview
    • Sensitivity labels
    • Capacity Metrics app

    Access rules reach rows and columns across every engine, and capacity use is visible per operation, if someone reads it.

Technology marks belong to their respective owners. No partnership, certification or endorsement is implied.

Migration and modernization

What moves, and what that asks of you.

The usual paths into Fabric, and the one most estates should keep open.

Planning a warehouse move into Fabric? The phased method, reconciliation gates and cutover plan are set out in data modernization.

  • Power BI Premium (P SKU)Fabric capacity (F SKU)

    P capacities are being retired in favour of F. P1 maps to F64 in compute, so the renewal is also the moment to decide what else should run on that capacity.

  • Azure Synapse AnalyticsFabric warehouse and lakehouse

    No published end-of-life, and migration assistants for Spark and dedicated SQL pools. Move workload by workload on evidence, with reconciliation, not on an assumed deadline.

  • Azure Data FactoryFabric Data Factory

    Pipelines can be upgraded, but connections, triggers and integration runtimes need checking one by one rather than assuming parity.

  • Azure DatabricksDatabricks plus Fabric

    Often the right end state rather than a migration: mirror the Unity Catalog into Fabric and keep engineering where it already works.

Performance and economics

What drives the bill.

Fabric is bought as capacity, so the economics are about sizing and utilisation rather than per-query cost.

Attributing and reducing Fabric and Power BI spend as an engagement is covered by data platform cost optimization.

Capacity size
F2 to F8192 compute units at $0.18 per CU-hour pay-as-you-go in East US. Size from the Capacity Metrics app, not from licence counts.
Reservation
A one-year reservation costs about 40% less than running pay-as-you-go all month, and wins once capacity runs more than about 14 hours a day.
Smoothing
Background work is spread over 24 hours, so overnight jobs rarely need a bigger capacity; sustained overload still throttles.
Per-user licences
Publishers need Power BI Pro or Premium Per User; below F64, so do viewers. The F64 viewer rule often decides the size.
Overage and pause
Overage billing costs three times the normal rate, and pausing bills accumulated usage at once. Both are tools, not defaults.

Governance and security

What has to be designed first.

Fabric's governance is strong when it is designed up front, and inconsistent when workspaces are created ad hoc.

Domains and workspaces
The unit of ownership and capacity assignment. Decide the structure before teams start creating workspaces of their own.
OneLake security roles
Folder, table, row and column access enforced across SQL, Spark and Power BI, defined once at the data rather than per tool.
Sensitivity labels
Labels travel with the data into exports such as Excel, which matters for finance and HR data.
Capacity boundaries
Workloads on one capacity can slow each other. Separate production reporting from heavy engineering when the metrics show contention.
Lineage and audit
Purview and OneLake diagnostics record who used what, and need an owner who actually reviews them.

Trade-offs

When it fits, and when it does not.

Where Fabric fits, and where another platform deserves consideration.

Where Microsoft Fabric tends to fit

  • Microsoft 365 and Azure estates where Power BI is the main consumer
  • Power BI Premium P capacity renewals, now F capacity decisions
  • Small data teams that want one managed platform instead of several services
  • Organisations whose governance already runs on Entra ID and Purview

Where another answer deserves consideration

  • DatabricksWorkloads span clouds, or Spark and machine learning engineering are the core of the work. Often used alongside Fabric rather than instead of it.
  • SnowflakeWorkload isolation per team, data sharing and multi-cloud portability lead the requirements.
  • Individual Azure servicesThe estate runs well on Synapse, Data Factory and Azure Databricks today and there is no Power BI capacity decision forcing a change.

Next step

Review your Fabric capacity and architecture.

Bring a Capacity Metrics app export or your Power BI Premium usage. Sizing and licence questions are answerable from it in a first conversation.

Review your Fabric capacity and architecture