Microsoft Fabric Explained: What It Is, What It Replaces, and Who Actually Needs It

A mid-sized financial services firm in Toronto has a problem most data leaders will recognise immediately. Their analysts pull from Azure Synapse Analytics for warehouse queries, connect Synapse to Power BI for dashboards, run ETL jobs through Azure Data Factory, store raw data in Azure Data Lake Storage Gen2, and manage machine learning experiments in Azure Machine Learning. Five services. Five cost centres. Five sets of access controls to reconcile. One data engineering team stretched across all of them.

This is the fragmentation problem Microsoft Fabric was designed to solve — and understanding what it actually is (versus what the marketing says) determines whether it’s the right answer for your organisation.

The Fragmentation That Preceded Fabric

Microsoft’s enterprise data story before Fabric was genuinely complicated. Azure SQL Data Warehouse became Azure Synapse SQL Pools, which became part of Azure Synapse Analytics Workspace, which added Spark pools, and eventually tried to absorb Power BI through Synapse Link. Meanwhile, Azure Data Factory remained a separate service for orchestration. Azure Databricks existed as a partner offering, not a first-party Microsoft product. Power BI Premium had its own capacity model, its own portal, its own governance separate from everything else.

For a data engineering team working in this environment, standing up a complete analytics stack meant provisioning and connecting six or more distinct Azure services. IAM policies had to be synchronised across all of them. Storage accounts had to be manually linked. Networking configurations had to be replicated. The operational overhead was substantial — and it compounded every time Microsoft released a new capability.

What Microsoft Fabric Actually Is

Microsoft Fabric, released for general availability in November 2023, is a unified SaaS analytics platform built on top of Azure infrastructure. The key word is SaaS: unlike the previous generation of Azure analytics services, which were PaaS offerings you provisioned and configured yourself, Fabric is delivered as a fully managed service. You buy capacity; Microsoft manages everything else.

Fabric is not a rebrand. It is a genuine architectural rethink. The platform is built around a single shared storage layer called OneLake — one logical data lake per Fabric tenant, with Delta Parquet as the native file format. Every workload in Fabric reads from and writes to OneLake. There is no separate storage account to provision, no Azure Data Lake Storage Gen2 resource to connect. Storage is automatic, tenant-wide, and unified.

The Six Fabric Workloads

Fabric organises its capabilities into six distinct workloads, each corresponding to a phase of the analytics lifecycle:

Data Engineering handles Spark-based data processing. You create Lakehouses — Delta-format storage containers with a built-in Spark engine — write PySpark or Scala notebooks, define Spark Job Definitions for scheduled processing, and build pipeline orchestration using the Fabric-native Pipeline editor.

Data Factory in Fabric provides the Copy activity and Dataflow Gen2 for ingestion and transformation. This is Fabric’s successor to Azure Data Factory, with a tighter integration into OneLake and the Fabric workspace model.

Data Science provides Jupyter-compatible notebooks, experiment tracking via MLflow, and model management within the Fabric environment. It is less mature than Databricks’ MLflow offering but improving with each release.

Data Warehouse is a dedicated T-SQL compute layer that reads from OneLake Delta tables. It is the successor to Synapse Dedicated SQL Pools, but re-architected to be serverless by default.

Real-Time Intelligence combines Eventstream (event ingestion), KQL Database (columnar time-series storage), Real-Time Dashboards, and Data Activator (trigger-based alerting). This replaces the Azure Event Hubs + Azure Data Explorer + Azure Stream Analytics combination.

Power BI is a first-class workload in Fabric, not an external tool connected via API. This integration unlocks DirectLake mode — a query mode that reads Delta files directly from OneLake without import or DirectQuery latency trade-offs.

OneLake: The Architectural Foundation

OneLake deserves special attention because it changes the economics of data sharing. In traditional Azure analytics architecture, sharing data between a Synapse workspace and a Power BI dataset meant either duplicating data, setting up complex external table references, or managing Synapse Link configurations.

In Fabric, all workloads share the same OneLake storage. A Delta table written by a Spark notebook in the Data Engineering workload is immediately readable by the Data Warehouse SQL engine, by Power BI in DirectLake mode, and by the Data Science notebook — with no data movement, no connectors, and no configuration. This eliminates an entire category of integration work.

OneLake also supports shortcuts: logical pointers to external storage — an AWS S3 bucket, an Azure Data Lake Storage Gen2 account, a Google Cloud Storage bucket — that appear as native folders in OneLake without copying data. For organisations with data in multiple clouds, shortcuts reduce the data movement required to unify analytics.

Who Fabric Is For

Fabric makes the most sense for organisations with one or more of the following characteristics:

Microsoft-aligned enterprises. Organisations with Microsoft 365 E3/E5 agreements, Azure-centric infrastructure, and Power BI as their BI standard will find Fabric’s licensing and identity model (Azure Active Directory / Entra ID throughout) a natural fit.

Power BI-heavy organisations. If Power BI is the primary analytics surface and the team is spending significant time managing import schedules, gateway configurations, and dataset refresh windows, Fabric’s DirectLake mode eliminates much of that operational overhead.

Teams consolidating away from Snowflake or Databricks. Organisations evaluating their platform spend who are already in the Azure ecosystem will find Fabric’s capacity-unit pricing model more predictable than per-query or per-DBU billing.

Organisations with Synapse Analytics already in place. Microsoft has signalled that Fabric is the strategic direction. Teams planning Synapse investments should evaluate Fabric first. (See our guide on migrating from Azure Synapse to Microsoft Fabric.)

Who Fabric Is NOT For

Not every organisation should adopt Fabric. The clearest misfit cases:

AWS-native organisations. If your infrastructure runs on AWS — Glue, Redshift, S3, Lake Formation — Fabric adds a Microsoft dependency that creates more complexity than it resolves. AWS-native teams building modern data platforms should evaluate the AWS native stack or platform-agnostic options like Databricks.

Organisations with heavy open-source Spark investment. Teams with significant Delta Lake + Apache Spark pipelines on open-source tooling, running outside Azure, will find Fabric’s Spark environment opinionated in ways that create migration friction.

Small teams that need simplicity over unification. The Fabric capacity model and workspace governance model have meaningful administrative complexity. A three-person data team with a single analytics use case may be better served by a simpler stack.

Licensing Overview

Fabric uses a capacity unit (CU) model. You purchase a Fabric SKU — from F2 (the smallest, suitable for development) to F2048 (enterprise-scale) — measured in compute units per second. The SKU determines available compute across all workloads. Storage is billed separately at ADLS Gen2 rates.

For Power BI specifically, Fabric F-SKUs include Power BI Premium capacity, replacing the previous P-SKU licensing. Organisations with existing Power BI Premium agreements may be able to migrate to an equivalent Fabric capacity without additional cost — worth validating with your Microsoft account team.

Per-user Power BI Pro licences remain required for users who create and publish content (reports, semantic models) in Fabric. Viewer access is included in the capacity licence.

Conclusion

Microsoft Fabric represents a genuine architectural advance over the previous generation of Azure analytics services. The unified OneLake storage layer and the SaaS delivery model reduce operational overhead meaningfully — particularly for Power BI-heavy organisations and Microsoft-aligned enterprises managing multiple Azure analytics services today.

The platform is still maturing. Some workloads — particularly Data Science and the Data Warehouse’s T-SQL compatibility — lag behind best-in-class alternatives. But for the right organisation profile, Fabric is the most coherent enterprise data platform Microsoft has shipped.

Evaluating Microsoft Fabric for your organisation? Infra IT Consulting helps Canadian and international businesses assess, architect, and implement modern data platforms. Book a discovery call →