Power BI to Databricks AI/BI Migration: The Practical Guide for 2026

Moving a Power BI report to Databricks AI/BI means taking its semantic model and DAX measures into Unity Catalog tables and metric views, rebuilding its pages as AI/BI dashboards over SQL datasets, and turning its RLS roles into Unity Catalog row filters that only bite under the right publish mode. It is a different job from connecting Power BI to Databricks as a data source, which many teams keep doing for the reports they do not move. Three names to have straight: AI/BI dashboards were renamed from Lakeview dashboards in April 2024 and took the AI/BI name that June, while the API still says Lakeview; the older Databricks SQL dashboards, called legacy dashboards at the end, reached end of life in March 2026; and Genie Code is the agent that, among other things, imports BI files.

Why teams move from Power BI to Databricks AI/BI

The data is already there. When the tables Power BI imports every night are Delta tables in Unity Catalog, the import model is a second copy with its own refresh schedule, gateway and capacity, and the measure definitions live in a .pbix rather than next to the data they describe. Databricks bills the compute under a dashboard rather than a seat; its business intelligence page says there are no additional license fees for AI/BI and standard DBU rates apply. Governance consolidates, with row filters, column masks, lineage and audit once in the catalog instead of once per semantic model, and viewer count becomes a capacity question rather than a license line (the comparison page has the current terms). The counterweight is real. Power BI remains stronger at pixel‑level formatting, paginated reports (pixel‑exact printable output) and the Excel workflow around them. A Power BI report on DirectQuery with single sign‑on already respects Unity Catalog permissions, and Databricks can publish Unity Catalog tables as a Power BI semantic model. Many estates end with both; this page is about the dashboards you decide to move.

What a Power BI report becomes in Databricks AI/BI

Two Databricks words first: a dataset is the SQL query a dashboard reads, and a metric view is a measure definition kept in Unity Catalog, or locally inside a dashboard, instead of inside a report. Three limits shape how a report is split, as of September 2026: 15 pages per dashboard, 100 datasets per dashboard and 100 widgets per page. In the tables, "converts" means the construct exists on both sides, "with review" means a person decides something, and "redesign" means there is no object to convert into.

Why the measures move before the pictures

A Power BI report carries its own semantics: the measures live in the model that ships inside the .pbix. Databricks AI/BI assumes the opposite. A dashboard reads datasets, dataset queries are read‑only SQL, and the arithmetic exists before the chart does. Build three layers: base views in Unity Catalog that join and clean the source tables, metric views holding the measures and the fields they can be grouped by, and dashboard datasets on top that query them with MEASURE(total_sales) grouped by whatever fields a widget needs. Local metric views inside a dashboard let you define measures and join relationships visually and export them to Unity Catalog later, the natural landing place for a small report's model. There is no model copy to refresh: an Import‑mode model's job splits between the dashboard result cache, 24 hours, best effort, warmed reliably only by a schedule, and materialized views for aggregations that are expensive to recompute.

Converting DAX to SQL and metric‑view measures

Rows run from the patterns that translate directly to the ones that are rebuilt. A DAX measure is evaluated in a filter context the visual assembles at query time; a metric‑view measure is evaluated against the fields in the widget plus whatever the dashboard filters pass down.

Behind the review rows: custom calculations give you the two LOD shapes that matter, Fixed (OVER (PARTITION BY dims)) and Coarser (AGGREGATE OVER (PARTITION BY * EXCEPT (field))); there is no named form for adding dimensions the way an iterator does in DAX, so those are modeled in dataset SQL. Custom calculations are scoped to one dataset, capped at 200 per dataset, and table visualizations do not render calculated measures, so a measure that feeds a table goes into the dataset SQL or the metric view. The LOD expressions to Databricks SQL page works the same window‑function mechanics from the Tableau side.

Row‑level security becomes Unity Catalog

A Power BI RLS role is a DAX filter with members assigned in the service. In Databricks the rule moves to the data: a row filter is a SQL function attached to the table that evaluates each row at query time, usually with is_account_group_member() against the account groups your Entra groups sync into, and a column mask covers what object‑level security did. Whether a viewer feels it depends on how the dashboard is published. Under Individual data permissions, viewers run queries with their own credentials and the filters apply per person; under Share data permissions, the default, every viewer sees the publisher's view. A dashboard that carried RLS in Power BI has to be published in the individual mode. The Genie companion reads the dashboard's datasets under the same publish mode: with shared permissions it sees the publisher's rows, including ones no widget shows, so restricted datasets need the individual mode for that reason too. Row‑level security to Unity Catalog goes through the publish modes in detail.

Databricks' own import, and where a migration tool fits

Databricks ships an importer. Genie Code's /importBI command accepts a Power BI template, a .pbit file, alongside Tableau workbook and data source files, and builds an AI/BI dashboard connected to local metric views, creating dashboard relationships where it finds them in the source. A template is exported from Power BI Desktop and holds the model definition, the queries and the pages without any data, which is why it is the accepted format and a .pbix is not. Direct uploads are capped at 100 MB, the agent works conversationally and asks for a screenshot for layout fidelity, and the documentation carries no list of which DAX constructs convert, so checking what each report lost stays with you.

Phases and validation

Validate in the published dashboard, not the draft: for a draft the viewer's own data permissions always apply, so a security check that passes there can fail in a dashboard published with shared permissions. Reconcile at the grand total, then under the filter combinations users actually apply, then per role; year‑to‑date and share‑of‑total measures are where the two platforms part ways. Microsoft's scanner APIs, once an admin enables them, supply the inventory, measures and their DAX included, and a report saved as a Power BI project holds the same definitions as plain‑text files.

What Antares does on this route

Everything above holds whichever tool you use; this section is about Antares, a BI migration tool listed on the Databricks Marketplace. The Analyzer is free, and its Power BI side is generally available: upload a .pbix file or a zipped PBIP project, one report per analysis, choose Databricks AI/BI as the target, and it returns a complexity score, a migration difficulty score for AI/BI, the main considerations and a component inventory. A .pbix can carry imported data; the Analyzer works from the report and model structure and does not copy data rows into its results. Measures are rated on what makes them hard to move, with context transition, measure switching and windowed ranking at the top, and an expression the parser cannot read is left unclassified rather than scored.

Conversion from Power BI to Databricks AI/BI is in private beta and is not self‑serve; the Converter page takes requests for beta access, and the home page keeps the current list of routes. Antares is not a data migration tool: it does not move data into Delta tables or build the lakehouse, and the tables have to be reachable through Unity Catalog first. Run the free Analyzer on a report to see what moves.

Related reading: Tableau to Databricks AI/BI, Power BI to Tableau, AI/BI next to Tableau and BI migration approaches. Primary sources: Databricks on datasets, metric views, custom calculations, row filters and column masks, publishing and sharing and importing BI files; Microsoft on CALCULATE, templates and row‑level security.