WhichM365Independent Microsoft 365 decision guide

HomeLearnMicrosoft Fabric data store

Data-store decision · 9-minute primer

Which Microsoft Fabric data store should you use?

Start with the workload, data shape, query pattern, latency, and owner. Lakehouse, Warehouse, Eventhouse, SQL database, and Cosmos DB solve different problems.

Start here when
A team is comparing storage options inside Microsoft Fabric
Primary audience
Business, data, analytics, architecture, and application leaders
Verified

Decision being made

Choose the store that matches the work

Decide which Fabric data store is the closest fit for the workload and whether a second store has a distinct, evidence-backed role. Do not begin with a familiar product name or assume every data project needs the same architecture.

Short answer

Match each option to its primary workload

Evaluate Lakehouse for flexible data engineering across structured and unstructured data, Warehouse for governed relational analytics and SQL reporting, Eventhouse for high-volume real-time event analysis, and SQL database for transactional application workloads. Evaluate Cosmos DB in Fabric when globally distributed NoSQL application data is material. A combination can be valid when each store has one clear job.

Intended audience

Who this guide is for

  • Business and data leaders comparing Fabric options.
  • Data, analytics, architecture, and application teams defining requirements.
  • Advisers preparing an evidence-based internal decision.

Relevant evidence

Describe the workload before selecting a store

  1. Classify the primary work as analytical, transactional, real time, or a defined combination.
  2. Identify whether the data is structured, semi-structured, unstructured, or event based.
  3. Confirm whether the working language is primarily SQL, Spark, or Kusto Query Language.
  4. State the required freshness, query complexity, schema enforcement, transaction behavior, and application latency.
  5. Name who owns data engineering, database operations, analytics, governance, capacity, access, and reliability.
  6. Determine whether existing data can be referenced through a OneLake shortcut or analyzed through mirroring instead of copied.

Recommended capability

Use the smallest clear workload boundary

OptionStart here whenPrimary language or patternDo not assume
LakehouseData engineering spans files and Delta tables, varied formats, exploratory analysis, or external lake integrationSpark plus a read-only SQL analytics endpointThat it replaces a transactional application database or every governed SQL warehouse
WarehouseStructured relational analytics, schema enforcement, enterprise reporting, and complex SQL queries dominateT-SQL and relational warehouse patternsThat it is the default for varied raw files, Spark-led engineering, or application transactions
EventhouseHigh-volume telemetry, logs, time-sensitive events, and low-latency analysis are centralKusto Query Language and real-time analyticsThat it is a general-purpose relational reporting store
SQL databaseAn application needs transactional reads and writes plus operational analyticsManaged relational database and T-SQLThat it is the default large-scale analytical lake or reporting warehouse
Cosmos DB in FabricA globally distributed application needs low-latency NoSQL access and flexible JSON schemasDistributed NoSQL application dataThat every semi-structured analytics workload needs an operational NoSQL database

Microsoft identifies OneLake as the common storage foundation. Shared storage does not make these items interchangeable; the workload and operating boundary still matter.

Why not the other solutions?

Rule out options for observable reasons

Why not Lakehouse?

It is not the default for a primarily relational, governed SQL reporting workload or a transactional application database.

Why not Warehouse?

It is not the default for varied file formats, Spark-led data engineering, transactional application writes, or high-volume KQL event analysis.

Why not Eventhouse?

It is specialized for real-time event and telemetry analysis, not general relational reporting or application transactions.

Why not SQL database?

It is designed for transactional and operational application workloads, not as the default analytical lake or enterprise reporting store.

Why not Cosmos DB in Fabric?

Its distributed NoSQL application boundary adds little when the requirement is conventional relational analytics, Spark engineering, or event analysis.

When an alternative becomes appropriate

Change paths when the workload changes

  • Move toward Lakehouse when varied file and table data, Spark processing, exploratory analysis, or external lake integration becomes material.
  • Move toward Warehouse when structured relational analytics, schema enforcement, high-performance SQL, and enterprise reporting become dominant.
  • Move toward Eventhouse when high-volume telemetry, logs, time-sensitive events, or Kusto Query Language becomes central.
  • Move toward SQL database when the system must support transactional application reads and writes.
  • Move toward Cosmos DB in Fabric when globally distributed, low-latency NoSQL application data and flexible JSON schemas are required.
  • Use more than one only when each boundary has a named workload, owner, evidence requirement, and cost question.

Risks and guardrails

Do not let a storage choice hide the operating model

Confirm data ownership, access, governance, lineage, quality, retention, performance, reliability, and cost controls. Check whether shortcuts or mirroring avoid unnecessary copies. Capacity and user-access requirements depend on the current tenant, region, agreement, workspace, and workload. A prototype does not establish production suitability. The responsible customer, partner, or technical team decides how to execute the work.

Questions to discuss internally

Requirements to confirm

  1. Is the primary workload analytical, transactional, real time, or a defined combination?
  2. Which data formats, query languages, freshness, transaction, and schema requirements are non-negotiable?
  3. Does the data already exist in a location that Fabric can reference rather than copy?
  4. Who owns data quality, access, governance, performance, reliability, and cost?
  5. Which evidence would justify more than one store?
  6. What Fabric capacity and user-access requirements must be verified before a final decision?

Public Microsoft sources

Verify the current Fabric boundaries

Verification date

Check current requirements before deciding

Source review completed . Fabric items, capacity, user access, regions, features, and terms can change and may vary by tenant, cloud, subscription, workspace, agreement, and release status.

Exact next destination

Place Fabric in the wider Microsoft AI decision

Finished this primer?Optional progress stores only this primer identifier and a completion time on this device.

WhichM365 provides independent educational decision support. It does not implement, deploy, operate, or manage Microsoft solutions. Product information is based only on public Microsoft sources. Its author is a Microsoft employee acting in a personal capacity. WhichM365 is not a Microsoft product or service and is not sponsored, authorized, or endorsed by Microsoft.