Microsoft has announced that Power BI Dataflows Gen1 is in maintenance mode (security, reliability, critical fixes), not deprecated or scheduled for removal. While the service will continue to receive critical fixes, all future investment is now focused on Dataflows Gen2 and Microsoft Fabric.

For many organizations, this may appear to be a routine product lifecycle update. It is not.

It is a signal that the architectural model behind Gen1 is no longer aligned with where data platforms are heading.

1.- What Dataflows Gen1 Was Designed To Do

When Power BI Dataflows Gen1 was introduced, it brought self-service data preparation into the Power BI ecosystem using Power Query in the cloud.

It enabled teams to:

• Ingest data from multiple sources

• Perform transformations using a familiar interface

• Store the results in a managed storage layer

• Reuse curated datasets across reports

In many organizations, Gen1 became the de facto transformation layer for BI workloads.

However, its design assumptions were rooted in a different era:

• BI as the primary consumption layer

• Limited reuse beyond reporting

• Separation between data engineering and analytics platforms

• Batch-oriented processing with relatively simple pipelines

Those assumptions no longer hold.

2.- Why This Matters Now

Today’s data platforms are expected to support:

• Multiple personas (BI, data engineering, data science, AI)

• Both batch and real-time processing

• Centralized governance and lineage

• Scalable, reusable data assets across the enterprise

Gen1 was not built for this level of convergence.

As a result, many organizations operating Gen1 at scale experience:

• Repeated transformations across multiple dataflows

• Increased load on source systems

• Inefficient capacity usage as workloads grow

• Limited automation, CI/CD, and lifecycle management

• Tight coupling between transformation logic and reporting

These are not isolated inefficiencies. They are structural limitations, and with no further innovation planned, they will persist.

The convergence of BI, engineering, and AI is not optional—it is a direct consequence of semantic models being reused by downstream AI agents, Copilot experiences, and operational analytics.

3.- Power BI Dataflows Gen1 vs. Azure Data Factory Dataflows Gen1

A frequent source of confusion is the similarity in naming between Power BI Dataflows Gen1 and Azure Data Factory Dataflows.

But they are fundamentally different.

3.1.- Power BI Dataflows Gen1

Power BI Dataflows Gen1 is a self-service data preparation capability within Power BI.

It is built on Power Query and designed primarily for BI teams to:

• Ingest and transform data

• Create reusable, curated datasets

• Feed Power BI semantic models and reports

Architecturally, it sits close to the consumption layer.

3.2.- Azure Data Factory Dataflows

Azure Data Factory Dataflows are part of Azure Data Factory (ADF), a cloud-native data integration service.

They are designed for data engineers to build scalable data pipelines and transformations.

Key characteristics:

• Visual, code-optional transformations

• Spark-based distributed execution

• Integration with orchestrated pipelines

• Designed for large-scale ETL/ELT workloads

Architecturally, they operate in the data integration and engineering layer.

If your organization is using ADF (or Azure Synapse Analytics pipelines), you should know that in Phase 1 you can leverage the new Data Factory to Microsoft Fabric Migration Assistant. This is an enterprise-grade web application that helps automate migration to Microsoft Fabric Data Pipelines (currently in preview).

The assistant enables a near one-click migration from ADF to Fabric Data Factory, eliminating the need for an upfront assessment (it’s free) and avoiding the need to rewrite existing pipelines.

Automation accelerates pipeline migration, but architectural decisions—such as data modeling, governance, and reuse patterns—still require deliberate design.

4.- Microsoft Fabric as a structural reset

Microsoft Fabric represents a shift from a collection of loosely connected services to a unified data platform.

It brings together data integration, engineering, warehousing, analytics, and business intelligence into a single SaaS environment. All built on OneLake, a single logical data foundation.

From an architecture standpoint, this enables a fundamental change:

From "multiple data copies and disconnected pipelines" to a "model centered on a single logical data layer with multiple consumption patterns".

This is not simply consolidation. It is a redesign of how data flows across the organization.

5.- Dataflows Gen2: A Different Role in the Architecture

Microsoft recommends using Dataflows Gen2 going forward, as all future Dataflow innovations will be delivered exclusively in Gen2. In addition, Dataflows Gen2 preserves the familiar Power Query experience while introducing significant platform enhancements in scalability, performance, governance, and cost efficiency.

Microsoft clarified this direction in updates published on the official Power BI blog, including analysis by Miguel Llopis.

However, Dataflows Gen2 should not be viewed as a direct successor to Gen1. They introduce a different role within the platform.

As you know, dataflows in Gen1 were primarily used to prepare data for Power BI datasets.

I see Dataflow Gen2 as a first-class transformation layer alongside pipelines, notebooks, and SQL—not a replacement for all engineering tools:

• Transformations executed using a scalable, managed engine optimized for Fabric workloads.

• Outputs written to Lakehouse or Warehouse

• Processing shared across multiple downstream workloads

• Improved monitoring, diagnostics, and lifecycle management

This shifts the pattern from “prepare data for a report” to “prepare data once and serve multiple use cases”. That distinction is where most of the long-term value is realized.

5.1 What Changes in Practice with Gen2

Moving to Gen2 introduces three architectural changes:

1. Data destinations replace direct consumption patterns Instead of consuming dataflows directly, outputs are written to Fabric Lakehouse or Warehouse for reuse across workloads.

2. OneLake enables shared data without duplication OneLake shortcuts replace linked tables, reducing redundancy and simplifying cross-workspace architectures.

3. ELT becomes the default execution model With capabilities like Fast Copy and improved incremental refresh, Gen2 enables scalable, governed ELT at enterprise level.

This effectively elevates dataflows from a BI preparation tool into a core data engineering capability within Fabric.

6.- Why Staying on Gen1 Becomes Increasingly Risky

Continuing to operate on Gen1 is not an immediate operational risk. Systems will continue to run.

However, it introduces a growing strategic risk.

Architectures based on Gen1:

• Accumulate inefficiencies as data volume and complexity increase

• Require more effort to maintain and scale

• Limit the ability to standardize and govern data assets

• Was not designed for taking advantage of new platform capabilities, including AI integration

Over time, this results in higher cost, slower delivery, and reduced flexibility.

In contrast, Microsoft Fabric-native architectures enable:

• A single transformation layer feeding multiple consumers

• Reduced duplication of data and logic

• Improved performance at the semantic layer through Direct Lake

• Better alignment with enterprise data governance models

7.- Migration: A Strategic Opportunity

The transition away from Gen1 should not be treated as a simple migration exercise.

A lift-and-shift approach may be appropriate for selected workloads.

But the real opportunity lies in simplifying and modernizing the architecture.

A structured approach typically includes:

1. Assessment - Understand the current Gen1 footprint, dependencies, and cost profile

2. Design - Define the target architecture aligned with Fabric capabilities

3. Migration - Execute initial workloads, often starting with a focused pilot

4. Validation - Ensure data quality and functional equivalence

5. Optimization - Reduce duplication, improve performance, and introduce new patterns such as Direct Lake

This is less about replacing a component and more about evolving the platform.

In practice, migration patterns vary depending on how dataflows are used across the organization. Three common scenarios emerge:

7.1.- Personal or Team Usage (Self-Service BI)

In smaller teams, Dataflows Gen1 are typically used for lightweight data preparation feeding Power BI reports.

In this scenario, migration is often straightforward:

• Move dataflows to Gen2 within a Fabric workspace

• Update semantic models to reference new dataflow IDs

• Maintain temporary compatibility during transition

The goal here is continuity with minimal disruption, while progressively adopting Gen2 capabilities such as improved refresh and Fast Copy.

7.2.- Departmental Usage (Shared Data Assets)

At the departmental level, dataflows are commonly reused across multiple reports and often rely on linked tables and shared transformations.

Here, migration becomes an opportunity to improve data architecture:

• Replace linked tables with OneLake shortcuts

• Reduce cross-workspace dependencies

• Centralize transformations into reusable Gen2 dataflows

• Redirect downstream models from

• to Lakehouse-based consumption patterns

This scenario is where most organizations begin to see real gains in reusability, performance, and simplification of data lineage.

7.3.- Enterprise Usage (Scalable Data Platform)

At enterprise scale, Dataflows Gen1 often act as a distributed transformation layer feeding multiple semantic models across domains.

In this case, Gen2 becomes part of a broader Fabric architecture:

• Outputs are persisted into Lakehouse or Warehouse layers

• Transformations are standardized across domains

• Governance is enforced through centralized compute and security models

• Consumption shifts from dataflows to governed data products

This is where migration becomes a platform redesign exercise, not a tool replacement.

8.- The Role of AI Changes Everything—But Only If the Architecture Is Ready

Microsoft Fabric introduces embedded AI capabilities, including Copilot, across the data lifecycle.

These capabilities can:

• Generate transformations

• Explain query logic

• Recommend performance improvements

However, AI is not a substitute for architecture.

Its impact is maximized when applied to a platform that is already centralized, standardized, and governed.

9.- Conclusion

Power BI Dataflows Gen1 and Azure Data Factory Dataflows Gen1 were never the same.

They were designed for different personas, operated at different layers of the architecture, and solved different problems—one closer to business intelligence, the other to enterprise data engineering.

However, despite these differences, there is an important common ground.

Both represent legacy patterns of distributed data transformation in a world that is now moving toward unified data platforms.

And because of that, a similar modernization methodology applies:

• Start with visibility and inventory

• Identify duplication and inefficiencies

• Redesign for reuse and shared data foundations

• Align transformations to a centralized platform

Whether starting from Power BI Dataflows Gen1 or Azure Data Factory Dataflows, the objective is the same:

To move from fragmented pipelines and duplicated logic To a model where data is prepared once and consumed broadly across the organization

That is exactly what Microsoft Fabric enables.

The shift of Gen1 to legacy status is not urgent, but it is directional.

And in platform decisions, direction matters more than deadlines.

Organizations that treat this as a narrow migration will see incremental gains. Those that use it to modernize their platform will unlock meaningful advantages in cost, performance, governance, and readiness for AI.

A practical starting point is simple:

• Inventory your Gen1 dataflows

• Run a targeted Gen2 pilot

• Define your Fabric architecture

From there, the focus should move from migration to modernization.

Because this is not about replacing Dataflows Gen1.

It is about building a data platform that is ready for scale, governance, and AI.

Let’s talk.