Azure Data Factory

About Azure Data Factory

Azure Data Factory is Microsoft's managed Azure service for building and operating hybrid data-integration pipelines. It is designed for teams that need to move data between cloud and on-premises systems, orchestrate ETL or ELT workflows, trigger transformations on other compute services, and monitor recurring data movement without building a custom orchestration platform. Azure Data Factory remains an active Azure service, but Microsoft now describes Data Factory in Microsoft Fabric as the next generation of the product and recommends new data-integration projects evaluate Fabric first.

What is Azure Data Factory used for?

Azure Data Factory is used to create data-driven pipelines that collect, move, transform and publish data across many systems. A pipeline groups activities into one workflow, while datasets describe the data being used and linked services define connections to data stores or compute services. This makes the product useful for scheduled batch ingestion, file-triggered workflows, migration projects, data-lake loading, warehouse preparation and other repeatable integration jobs.

Data Factory can copy data directly between supported sources and destinations, or it can dispatch transformation work to services such as Azure Databricks, Azure SQL Database and other supported compute engines. Mapping data flows provide a visual transformation option for teams that want to design transformations without managing Spark clusters themselves.

How do Azure Data Factory integration runtimes work?

The integration runtime is the execution bridge between pipeline activities and linked services. Microsoft currently documents three Data Factory integration runtime types: Azure integration runtime, self-hosted integration runtime and Azure-SSIS integration runtime.

Azure integration runtime is managed compute for cloud data movement, data flows and activity dispatch. Self-hosted integration runtime is installed in a private network and is intended for hybrid scenarios where cloud services need to reach on-premises or private data stores. Azure-SSIS integration runtime runs SQL Server Integration Services packages in managed Azure compute. Choosing the wrong runtime can create avoidable networking, security, performance or cost problems, so the runtime design should be decided with the data-source network topology rather than after pipelines are already built.

How is Azure Data Factory priced?

Azure Data Factory does not have one flat product price. Microsoft currently meters several parts of the service separately, including pipeline orchestration and execution, integration runtime usage, data movement, data-flow execution and debugging, and Data Factory read, write and monitoring operations. Mapping data flows are billed using their managed compute while they run, and other Azure services invoked by a pipeline are billed separately under those services.

This means cost depends heavily on pipeline frequency, activity count, runtime choice, data volume, transformation duration and network transfer. A pipeline that runs thousands of small activities can have a different cost profile from one that moves large datasets with only a few activities. Teams should also account for outbound Azure data-transfer charges where a copy moves data out of an Azure datacenter.

What are the main limitations and operational tradeoffs?

Data Factory reduces the need to build custom orchestration and data-movement services, but it does not remove the need for data engineering discipline. Complex pipelines can become difficult to troubleshoot when they contain many linked activities, parameters, triggers and external dependencies. Connector throttling, source-system limits and external compute failures can also become the real bottleneck even when the Data Factory service itself is healthy.

Hybrid deployments require extra planning because self-hosted integration runtime introduces customer-managed infrastructure. Microsoft currently supports self-hosted integration runtime on Windows, and organizations are responsible for the host, patching, availability and connectivity. Mapping data flows also introduce their own managed compute cost, so they are not automatically cheaper than running transformations on an existing platform.

How does Azure Data Factory compare with Data Factory in Microsoft Fabric?

This is now a key buying question. Microsoft's June 2026 Azure Data Factory documentation labels Data Factory in Microsoft Fabric as the next generation of Azure Data Factory, with a simpler architecture, built-in AI and integration with the broader Fabric analytics experience. Microsoft explicitly recommends that customers who are new to data integration start with Fabric Data Factory, while existing Azure Data Factory workloads can continue operating and can evaluate an upgrade path.

That does not make Azure Data Factory obsolete. Existing organizations may have substantial investments in ADF pipelines, self-hosted runtimes, Azure-SSIS workloads, deployment automation and Azure-specific network design. A migration should therefore be evaluated as an architecture project, not treated as a simple rename.

Who should choose something else?

A new analytics team that wants one environment spanning data integration, engineering, warehousing, real-time analytics and reporting should evaluate Microsoft Fabric Data Factory before starting a large new Azure Data Factory estate, because Microsoft now points new data-integration customers in that direction.

Teams whose main requirement is code-first stream processing or transformations may be better served by the compute platform already running those workloads, such as Databricks or another orchestration system. Simple application workflows may fit Azure Logic Apps better, while developers needing event-driven custom code should consider Azure Functions. Azure Data Factory is strongest when the central problem is repeatable data movement and orchestration across diverse data systems, especially when hybrid connectivity or existing ADF investments matter.