This guide helps data leaders plan an Azure Synapse to Databricks migration that consolidates several Synapse engines onto one governed platform, without disrupting the reports the business runs on.
Azure Synapse to Databricks migration helps enterprises consolidate separate Synapse services, from the warehouse to Data Factory pipelines, onto one governed Databricks platform for analytics and AI.
Key Takeaways
- Azure Synapse to Databricks migration brings dedicated SQL pools, serverless SQL, Spark pools and pipelines onto one platform governed by Unity Catalog
- Dedicated SQL pools hold most of the effort, while Spark pools usually move with few changes
- Microsoft hasn’t set a retirement date for Synapse, though several Synapse features are already retired or closed to new projects
- Azure Databricks is a first-party Azure service, so identity and Power BI connections stay within the Microsoft ecosystem
- Validation is often the largest piece of work, so plan reconciliation into every wave from the start
Why Migrate From Azure Synapse to Databricks?
Enterprises move from Synapse to Databricks to replace several separate engines with one platform for analytics and AI, and to cut the operating effort of keeping those engines in step.
Many Synapse workspaces grew piece by piece. A typical estate runs dedicated SQL pools for the warehouse, serverless SQL for querying the data lake, Spark pools for engineering and Azure Data Factory or Synapse Pipelines to move data between them. Each engine has its own tuning and permissions, and keeping them aligned takes real effort.
Databricks runs these workloads on one platform, with Unity Catalog governing data and AI models in one place. Dedicated pools also bill for capacity sized to peak demand, while Databricks compute can stop when idle.
Databricks’ own Synapse migration guide cites two customer results (Source). Casey’s, a convenience store chain, cut operational data delivery from eight hours to four, and Italgas reported a 73% reduction in workload costs after moving off Synapse and Azure Analysis Services. Your own results will depend on workload mix, so model them before setting targets.
Is Azure Synapse Being Retired? What Microsoft Has Confirmed
Microsoft hasn’t announced a retirement date or migration deadline for Azure Synapse, yet it calls Microsoft Fabric its strategic analytics platform and has closed off some Synapse features.
Microsoft’s Synapse documentation states there’s no required migration deadline, while describing Fabric as its strategic platform for new analytics innovation (Source). The clearer signals sit at feature level:
- Synapse Data Explorer: retired on October 7, 2025, with its workloads deleted
- Synapse Spark: version 3.5 is the only runtime still generally available
- Synapse Link for Azure Cosmos DB: closed to new projects, with Fabric Mirroring recommended instead
- Microsoft Purview Data Governance: support ends on November 12, 2027, with Microsoft naming Fabric as its data governance solution (Source), while Purview’s data security and compliance features continue
None of these changes forces an immediate move. They do mean most Synapse estates face upgrades and a governance rebuild either way, so many data leaders are choosing their next platform on their own timeline.
Synapse’s Multi-Engine Workspace vs the Databricks Lakehouse
Synapse splits work across separate SQL and Spark engines with their own rules, while the Databricks lakehouse runs analytics and AI on one copy of data under one governance layer.
| Area | Azure Synapse Workspace | Azure Databricks Lakehouse |
|---|---|---|
| Engines | Dedicated SQL pools, serverless SQL pools and Spark pools, each managed separately | One platform for SQL, data engineering, streaming and machine learning |
| Data storage | Dedicated pools hold data in their own managed storage, while serverless SQL and Spark read files in Azure Data Lake Storage | Open Delta Lake tables, typically in your own Azure Data Lake Storage, shared by every workload |
| Performance tuning | Engineers choose how tables are spread across nodes and indexed | Databricks organizes data layout largely on its own |
| Governance | SQL permissions per engine, often alongside Microsoft Purview | Unity Catalog across tables, files and AI models, with lineage that reaches notebooks and jobs |
| Orchestration | Synapse Pipelines or Azure Data Factory | Lakeflow Jobs, with Azure Data Factory still able to trigger Databricks work |
| BI connectivity | Power BI and other tools connect to Synapse endpoints | Power BI and other tools connect to Databricks SQL warehouses |
The practical effect is fewer systems to keep in step. One governance model covers everything, and teams stop copying data between engines to get work done. For a broader comparison, see our Databricks vs Azure Synapse guide.
Scoping a Synapse Migration: What Moves and How Hard Each Part Is
Synapse is several services under one name, and each moves at its own pace, from Spark pools that move almost as they are to dedicated SQL pools that carry years of custom logic.
| Synapse Component | Databricks Destination | Typical Effort |
|---|---|---|
| Dedicated SQL pools | Delta tables queried through Databricks SQL warehouses | High, since stored procedures and years of tuning live here |
| BI semantic models and reports | The same tools repointed to Databricks SQL warehouses | High, since every report needs testing with its owners |
| Synapse Pipelines and Azure Data Factory | Lakeflow Connect and Lakeflow Jobs, or Azure Data Factory triggering Databricks | Medium |
| Permissions and Purview governance | Unity Catalog | Medium |
| Serverless SQL pools | Views and tables over the same lake data in Unity Catalog | Low to medium |
| Spark pools | Databricks notebooks and jobs, since both run Apache Spark | Low, apart from Synapse-specific utilities and linked services that need replacing |
Effort concentrates in the warehouse because years of business rules, table tuning and report connections have built up there. The work also reaches into the pipelines feeding it and the permissions around it, and many estates carry older SSIS packages that belong in the inventory too.
The table also suggests where to start. Spark workloads and serverless views make good early wins, while dedicated SQL pools and the reports that depend on them need the most careful planning. Give each component its own milestones inside one shared plan, so a slow piece doesn’t hold up quick wins elsewhere.
Azure Synapse to Databricks Migration Plan: Six Steps for Enterprises
A Synapse to Databricks migration runs through discovery, assessment, design, a pilot, migration waves and validated cutover, with Synapse kept available as a fallback until each wave is proven.
The steps follow the approach Databricks outlines in its Synapse migration guide, shaped by what we see in enterprise programs.
Step 1: Discover What’s Running Across the Synapse Estate
Start by collecting facts about the estate. Databricks’ free Lakebridge profiler, currently an experimental feature, connects to a Synapse workspace and gathers configuration and usage details across dedicated pools, serverless SQL and Spark pools. Those numbers also feed the cost comparison your sponsors will ask for.
Step 2: Assess Code Complexity and Set Migration Priorities
Lakebridge’s analyzer then grades T-SQL code by complexity and maps dependencies between objects, which shows early where conversion will need engineers. Use the results to cut scope: retire objects nobody uses, and plan to move low- and medium-complexity workloads first.
Step 3: Design the Target Architecture and Migration Sequence
Decide the target architecture and the order of moves. Three choices matter most here:
- How much to change: Many teams convert code automatically first, then modernize in later waves, which gets workloads off Synapse sooner
- What users see first: A BI-first sequence uses Lakehouse Federation, which lets Databricks query Synapse tables where they sit, so analysts can start working on Databricks before every table has moved. Synapse still handles part of those queries, so the pools serving them stay online until their data moves
- The future of Azure Data Factory: Data Factory can keep calling Databricks jobs during migration, with schedules moving to Lakeflow Jobs wave by wave so two scheduling tools don’t stay in place for years
Step 4: Pilot One Use Case From Ingestion to Dashboards
Pick one representative use case and take it from ingestion to the dashboards that consume it. Agree pass criteria before starting, such as matching figures, acceptable report refresh times and access rules that behave as before.
The pilot proves the approach on your data and leaves your team with reusable patterns, test scripts and access templates before the larger waves begin.
Step 5: Migrate in Waves Across Four Parallel Workstreams
Each wave runs four workstreams side by side, so every part of a business domain moves together:
| Workstream | What Moves | Destination |
|---|---|---|
| Ingestion | Data feeds from Azure Data Factory and Synapse Pipelines | Lakeflow Connect, or partner tools that write to Delta Lake |
| Transformation | T-SQL stored procedures and transformation logic | Databricks SQL and Lakeflow pipelines |
| Orchestration | Schedules and dependencies | Lakeflow Jobs |
| Consumption | Power BI semantic models and other reports | Databricks SQL warehouses |
Bring data into the lakehouse before converting the SQL that depends on it, so each converted procedure can be tested against real data straight away. Dedicated pools can export tables as Parquet files to Azure Data Lake Storage for Databricks to load, while Azure Data Factory copy jobs or partner tools suit ongoing feeds.
Step 6: Validate Results and Cut Over With a Fallback in Place
Run Synapse and Databricks in parallel for critical workloads, and reconcile results after every wave. Engineers compare row counts, totals and record-level details, while report owners confirm their dashboards show the same numbers. Agree an exit rule in advance, such as one month-end close with matching figures.
When a wave passes, switch its consumers to Databricks and pause the matching dedicated SQL pools. A paused pool stops compute billing while storage is still billed, which preserves a rollback option and gives stakeholders confidence before anything is deleted.
Converting T-SQL and Stored Procedures From Synapse to Databricks
Most Synapse T-SQL converts automatically, while stored procedures with cursors, dynamic SQL or row-by-row logic need engineers to redesign them for Databricks.
Databricks reports that tooling handles roughly 80% to 90% of conversion work in a typical Synapse migration, which leaves engineers to focus on the logic that needs judgment. Lakebridge’s Morpheus converter lists Synapse dedicated SQL pools among its supported sources.
Three changes come up in almost every estate:
- Tuning settings disappear: Synapse’s choices about how tables are spread across nodes and indexed have no place on Databricks, where liquid clustering and predictive optimization organize data automatically for tables Databricks manages.
- Familiar functions get new names: GETDATE becomes CURRENT_TIMESTAMP, ISNULL becomes COALESCE, and SELECT TOP becomes a query ending in LIMIT, among many similar changes that tools handle automatically.
- Text comparisons behave differently: Dedicated SQL pools usually ignore upper and lower case when comparing text, while Databricks SQL doesn’t by default. A case-insensitive collation setting, or lowercasing both sides of a comparison, keeps results consistent.
Stored procedures carry the rest of the effort. Parameters, variables and conditional logic translate well, and T-SQL’s TRY…CATCH error handling has an equivalent in Databricks SQL’s error handlers. Logic that processes one row at a time, including cursors, usually works better rebuilt to handle whole sets of rows at once, or as pipelines, and dynamic SQL needs review case by case.
Moving Power BI Reports and Other Consumers From Synapse to Databricks
Power BI and other reports are often the hardest part to move, since every semantic model and dashboard needs repointing and testing, then sign-off from the people who rely on it.
Most Synapse estates in Microsoft-centric organizations feed Power BI. Power BI connects to Azure Databricks through a native connector, so reports can usually keep their design while their data source changes to a Databricks SQL warehouse. Fabric can also read Unity Catalog tables in place through mirroring, without copying the data, which suits organizations that want Power BI to read Databricks data through Fabric.
Plan this work report by report. Inventory every semantic model and the Synapse objects it reads, decide which reports to rebuild and which to repoint, and agree a test with each report owner before cutover. Teams that also want to move dashboards into Databricks AI/BI can follow the approach in our BI migration guide.
Security, Access and Release Pipelines on Azure Databricks
Azure Databricks runs within your Azure environment, so private networking, Microsoft Entra ID sign-in and Git-based release pipelines all carry over, each with some setup work.
Security teams usually start with three areas:
- Network: Databricks can be deployed into your own Azure virtual network, with Azure Private Link keeping traffic off the public internet
- Identity: Users sign in with Microsoft Entra ID, and automatic identity management brings users and groups from Entra ID into Databricks, on by default for newer accounts
- Access: Unity Catalog replaces Synapse SQL roles and Purview policies with one set of permissions covering data and AI assets
Map Synapse roles to Unity Catalog groups inside each wave, so permissions don’t turn into a last-minute review when users switch. On the release side, code lives in Git folders, and Declarative Automation Bundles, formerly Databricks Asset Bundles, package jobs and settings so Azure DevOps or GitHub Actions can promote them from test to production.
After Cutover: Running and Optimizing Workloads on Databricks
After each wave, tune Databricks for real usage, keep table maintenance running, train the teams that now work on the platform and retire Synapse resources only once results hold.
Delta tables need ongoing housekeeping, especially where frequent loads create many small files. Predictive optimization, which is on by default for Unity Catalog managed tables, handles compaction and cleanup, so confirm it’s active and review the largest tables as query patterns settle. It doesn’t run on external tables, so plan to convert those to managed tables or schedule maintenance for them. Our Databricks cost optimization guide covers spend controls in more detail.
Enablement deserves real budget. Databricks notes that under-investing in enablement is a common reason projects miss adoption targets, so give your teams hands-on time before their workloads switch over. Once each wave has run cleanly for an agreed period, delete the paused Synapse resources.
Costs deserve the same discipline. Synapse dedicated pools bill for provisioned capacity for as long as they run, even when idle, while Databricks SQL warehouses can stop automatically when idle, so build the comparison from real workload history. Count the parallel-run months and paused-pool storage in the business case too.
Best Practices for a Successful Move From Synapse to Databricks
The strongest Synapse migrations assess before converting, redesign where the platform does the work and invest as much in validation and enablement as in moving code.
| Practice | Why It Matters |
|---|---|
| Assess the estate before converting any code | Profiling and complexity scoring show what to retire, what to automate and where the risk sits |
| Automate the bulk, focus people on the rest | Tools handle most conversion, so engineers spend their time on the logic that needs judgment |
| Redesign where Databricks does the work | Manual tuning carried over from Synapse adds effort and rarely helps on Databricks |
| Treat validation as its own workstream | Reconciling results is often the largest effort, and business trust depends on it |
| Settle the role of Data Factory early | One scheduling tool is easier to run and troubleshoot than two working side by side |
| Keep Synapse paused during cutover | A paused pool keeps a low-cost rollback path while confidence builds |
| Fund enablement from the start | Teams trained on the platform early adopt it faster and raise fewer issues after cutover |
For strategy decisions that apply across source platforms, our Databricks migration guide covers the wider picture.
Synapse to Databricks Migration Challenges and How to Avoid Them
Most Synapse migration challenges come from carrying old Synapse design choices into Databricks, or from underestimating the work involved in reports and validation.
- Copying Synapse table design: Columns such as customer or order IDs, used to spread Synapse tables across nodes, make poor partition choices on Databricks, where automatic clustering does the job better.
- Ignoring case sensitivity: Filters and joins that relied on case-insensitive matching in Synapse can quietly return different results unless collation is set to match.
- Leaving reports to the end: Power BI models and dashboards take time to repoint and test, so plan them with each wave.
- Letting small files pile up: Frequent loads create many small files, so plan table maintenance from the first wave.
- Deleting Synapse too early: Without a paused fallback, any issue after cutover becomes an outage.
Databricks or Microsoft Fabric: Which Fits Your Synapse Estate?
Synapse customers usually weigh two destinations, Microsoft Fabric and Azure Databricks, and the right choice depends on priorities such as AI ambitions and how much code you want to change.
Microsoft’s Fabric Migration Assistant moves dedicated SQL pool objects and code into Fabric Data Warehouse, which also uses T-SQL, so code changes are usually smaller, though some T-SQL features aren’t supported. Fabric suits organizations that want an all-in-one SaaS analytics platform with Power BI built in. Both platforms store data in open formats, with Delta Lake at the core.
Databricks suits organizations that put data engineering and AI at the center, want one governance layer for data and AI assets, or run across more than one cloud. Many enterprises use both, with Fabric reading Databricks tables in place through mirrored Unity Catalog for Power BI users. Our Databricks vs Microsoft Fabric comparison covers the trade-offs in more depth.
How LatentView Helps With Azure Synapse to Databricks Migration
We are a Databricks Elite C&SI partner, and our Databricks Center of Excellence brings together 400 data engineers, including 150+ certified specialists. For a leading construction management SaaS company, we migrated 5,000+ daily dbt models onto Databricks using MigrateMate, consolidating its fragmented data stack into one ecosystem, as this case study shows.
Transform your Synapse workspace into a governed Databricks lakehouse with a partner that brings 20+ years of experience serving Fortune 500 clients. Our MigrateMate solution automates source assessment, object conversion and data validation, typically reducing migration costs by 30% to 40%, while BrickShift, our BI migration accelerator, converts Power BI dashboards into Databricks AI/BI. Map your first Synapse wave with our Databricks partnership team and see which workloads should move first.
Frequently Asked Questions About Azure Synapse to Databricks Migration
1. Is Azure Synapse Analytics Being Retired?
Microsoft hasn’t set a retirement date or migration deadline for Synapse. Several Synapse features are already retired or closed to new projects, and Fabric is Microsoft’s strategic analytics platform.
2. Which Synapse Workloads Are Hardest to Migrate?
Dedicated SQL pools and the Power BI reports built on those pools take the most effort. Spark pools are usually the simplest, since both platforms run Apache Spark.
3. Can Synapse T-SQL Be Converted to Databricks Automatically?
Mostly. Databricks reports that tooling handles roughly 80% to 90% of conversion, while cursors, dynamic SQL and row-by-row logic need manual redesign.
4. Can Power BI Keep Working After Moving to Databricks?
Yes. Power BI connects to Azure Databricks through a native connector, so reports can be repointed to Databricks SQL warehouses and tested with report owners.
5. Should Synapse Be Deleted Right After Cutover?
No. Pausing dedicated SQL pools stops compute billing while storage is still billed, which keeps a rollback option until each migrated workload has proven stable.