The ODBC to ADBC transition reaches its first planned month: audit your Fabric refreshes

Microsoft publishes four milestones. Every one is marked planned, and two of them add subject to rollout readiness, so treat the list as a shape rather than a schedule you can hold anyone to.

• July 2026 (planned). Broad rollout of the tenant setting begins. Workspace behaviour inherits this setting and can be overridden by workspace admins where supported.
• October 2026 (planned). Microsoft plans to begin enabling the tenant setting by default in phases, subject to rollout readiness.
• Early Q1 2027 (planned). Microsoft plans to begin removing ODBC drivers from the service, subject to rollout readiness. If you need to stay on ODBC, you must use a gateway to continue running queries with it.
• Spring 2027 (planned). The ODBC drivers in scope are planned to no longer ship with Power BI Desktop or the gateway. That is Northern Hemisphere spring, in Microsoft's wording, not a Southern Hemisphere season.

So where does that leave you in the middle of October? October 2026 is the month Microsoft planned to start the default flip, and its guidance still words that step as plans to begin. Microsoft has not said the rollout has started. The same page says that before enforcement milestones Microsoft plans to provide advance notice so customers can complete migration and validation activities. Read the month as arrived and the flip as unannounced, and act accordingly.

Notice what is not on that list of milestones: a permanent opt out. Microsoft is blunt about it.

So the decision in front of you is not whether to move. It is which items you move first, and which you pin for a short while so you can move them on your own timetable rather than Microsoft's.

Microsoft's transition table lists eight connectors. Check yours against it before you read any further.

• Databricks: Simba Spark ODBC moves to Databricks ADBC.
• Azure Databricks: Simba Spark ODBC moves to Databricks ADBC.
• Dremio: Simba Drill ODBC moves to FlightSQL ADBC.
• Google BigQuery, including Google BigQuery (Microsoft Entra ID): Simba Google BigQuery ODBC moves to BigQuery ADBC.
• Hive: Simba Hive ODBC is marked Deprecated, with no replacement driver named.
• Impala: Simba Impala ODBC moves to HiveServer2 ADBC.
• Snowflake: Simba Snowflake ODBC moves to Snowflake ADBC.
• Spark: Simba Spark ODBC moves to HiveServer2 ADBC.

That list is a two minute filter. If your estate reads only from on premises SQL Server, this is not your problem and you can stop here. If any part of it touches Databricks, Snowflake or BigQuery, it is your problem, and usually in more places than you expect, because those connections get copied from model to model. Microsoft says the default applies wherever you create a connection, including semantic models, Dataflows Gen2 and paginated reports.

If Databricks is your warehouse, the vendor adds two things of its own. Azure Databricks says new Power BI connections in Desktop and the service have used ADBC by default since February 2026, while existing connections continue on ODBC unless someone updates them. It also sets a constraint the Power Query page does not mention: you must use only one driver type per semantic model, so configure every Azure Databricks connection within a model to either ADBC or ODBC. Both are statements about the Databricks connector, not about the tenant setting milestones above.

Microsoft ships the tool for exactly this. It is called pq-adbc-advisor, a read only Fabric notebook in the Microsoft Fabric Toolbox. It scans a workspace and returns a per item impact report covering semantic models, dataflows and data pipelines, and it sorts every item into one of three risk categories: Will fail, Needs review, or Ready.

Start with Will fail, because Microsoft defines it precisely and the definition tells you what to fix.

The fix sits inside the definition: remove the pin from the source step, or bind the item to a gateway before cutover. Needs review covers items that pin ODBC but are gateway backed, and queries that build a raw DSN string ADBC does not consume verbatim. Ready means the query already pins ADBC, or is unpinned so the tenant switch handles it cleanly.

A tenant wide scan through the Fabric admin Scanner API typically takes 15 to 30 minutes for a mid sized tenant. That is the point worth repeating to whoever approves your week: this is an afternoon of work, not a project that needs a budget line.

Set your expectations where Microsoft sets its own. It calls the impact report a starting-point diagnostic, not a final audit, and says to treat it as the first pass that finds the bulk of the work, because connector coverage is still expanding and some corner cases do not map cleanly to a standard connector. For high-value production reports, validate the ADBC path in a copy of the item before you flip production over. The notebook rewrites no M expressions, and it sends no M code, item names, endpoint URLs or credentials outside the workspace.

Here is the part that catches people. Gateway routed refreshes do not switch, and that is deliberate.

Read that twice if you run a hybrid estate. When your tenant flips, part of your refresh fleet moves to the new drivers and part of it stays exactly where it is. Two semantic models built from the same template, pointed at the same warehouse, can behave differently for a reason that is invisible in the report itself. One routes through a gateway. The other does not. It also means you cannot validate the new path through a gateway: Microsoft says to test ADBC on a cloud connection.

So a gateway binding is shelter, not a fix. Microsoft calls it a deferral rather than a permanent opt out, and the ODBC drivers are planned to stop shipping with the gateway in spring 2027. Anything parked behind a gateway still needs a plan, just a slightly later one.

8

Connectors listed in Microsoft's transition table

3

Risk categories returned by pq-adbc-advisor

15 to 30 minutes

Typical tenant wide Scanner API scan, mid sized tenant

October 2026

First planned month of default enablement

All four figures come from Microsoft's own guidance on the driver change and from the advisor's ReadMe in the Fabric Toolbox, both accessed 13 October 2026.

Three steps, in order, and none of them needs a business case.

1. Run pq-adbc-advisor across the tenant and export the results grouped by workspace.
2. Fix the Will fail items first. Those are the refreshes that stop at cutover, with no warning in the report.
3. Decide who is watching refresh health through the cutover window, and on what.

Do the first two this week, while the tenant setting is still off in your tenant. For the third, see the Fabric Monitoring Tool page for how we keep an eye on refresh health across a whole tenant through a change like this one.

1. Transition from ODBC to ADBC drivers in Power BI and Microsoft Fabric (Microsoft Learn) (accessed 2026-10-13)
2. pq-adbc-advisor accelerator, microsoft/fabric-toolbox (GitHub) (accessed 2026-10-13)
3. Configure ADBC or ODBC driver for Power BI (Azure Databricks, Microsoft Learn) (accessed 2026-10-13)
4. What's coming? Power BI connections will transition to ADBC (Azure Databricks, Microsoft Learn) (accessed 2026-10-13)