Databricks AI/BI vs Tableau: What Changes for Dashboard Owners
September 10, 2026
The difference that decides a Tableau to Databricks AI/BI migration is where the logic lives. Tableau keeps calculations, interactions and often security inside the workbook. Databricks AI/BI keeps them in dataset SQL and Unity Catalog, and most other differences follow. This page does not rank the platforms; it lists what your dashboards will meet.
What changes, feature by feature
Accurate as of September 2026; the Databricks layer ships often, so re-check a row before planning.
Building and calculating
| What you are comparing | Tableau | Databricks AI/BI |
|---|---|---|
| Calculation language | Calculated fields, LOD expressions, table calculations, in the workbook | Dataset SQL, custom calculations (Fixed and Coarser LOD), or a metric view |
| Layout | Freeform canvas, tiled and floating objects | A 12-column snap grid, widened from 6 in February 2026 |
| Dashboard size | One workbook, many sheets and dashboards | 15 pages, 100 datasets, 100 widgets per page |
| Visual types and custom visuals | Marks-based charts plus extensions | Fixed list of built-in types; Vega-Lite custom visuals in Public Preview, no treemap |
| Dual axis | Dual axes on any two measures | An Enable dual axis option on combo charts |
| Interaction between visuals | Filter, highlight, go to sheet, go to URL, parameter and set actions | Cross-filtering on the same dataset; drill-through to a target page, filters from the clicked mark |
| Tooltips | Viz in Tooltip embeds another view | Number and label formatting; no viz-in-tooltip equivalent documented as of September 2026 |
Sharing and running
| What you are comparing | Tableau | Databricks AI/BI |
|---|---|---|
| Scheduled delivery | Email subscriptions on a schedule | PDF by email; PNG and PDF to Slack and Teams; CSV, TSV or Excel attachments on all three; 100 subscribers |
| Embedding | Embedding API in the host application | Light mode only; external viewers via a service principal, 20 loads per second, no Ask Genie |
| Natural language | Tableau Pulse digests, Tableau Cloud only | A companion Genie Agent, on by default, workspace members only; still labeled Public Preview on the partner-powered page |
| Version control | Revision history on the site | Git folders track drafts only; a branch switch deletes dashboards absent on the target branch |
| Licensing shape | Creator, Explorer and Viewer site roles | "There are no additional license fees ... $DBU rates apply" |
| Offline copy | A .hyper extract works offline | A 24-hour best-effort cache warmed by a schedule; no offline file documented |
Where the logic lives
A Tableau workbook is self-contained: its calculations travel with the file, so it changes hands and still produces the same numbers. In AI/BI the same math sits in one of three homes and you pick which: dataset SQL, a custom calculation, or a Unity Catalog metric view that other dashboards and Genie Agents read too. That choice is the migration, and level of detail expressions have the longest mapping, in LOD expressions in Databricks SQL.
A metric view is not a free-form model, and its shape decides how much of a workbook fits. Databricks documents joins in metric views as star and snowflake schemas, plus one-to-many joins that aggregate facts. Joins are many-to-one by default, and a many-to-many case takes the first matching row (as of September 2026). A workbook blending several fact tables at one grain is a modeling decision first.
What the grid does to a freeform dashboard
Tableau places objects where you put them. The AI/BI canvas snaps to a 12-column grid, widened from six in February 2026. Geometry does not convert across that boundary, so copy the decisions instead: which number is largest, what sits above the fold, which filter is global.
One consequence reaches past layout. Charts truncate above 15,000 rows and tables above 100,000, per the limits page as of September 2026, so a wide Tableau crosstab needs an aggregation decision, not a resize. The render and refresh story is on the migration guide.
Which interactions have an equivalent
Filter actions have the closest match: cross-filtering between visuals on the same dataset, or on datasets joined by a dashboard relationship since September 3, 2026 (Public Preview), and drill-through to a target page with filters set by the clicked mark. Saved filter states have a direct equivalent: editors save named shared bookmarks, up to 10 per dashboard, and a published dashboard keeps filter and parameter selections in the URL. Go to URL survives as the Link display type in a table column. Highlight, parameter and set actions have no equivalent in the AI/BI docs as of September 2026, so plan them by hand. Parameters have their own page, parameters in AI/BI dashboards.
What Genie replaces and what it does not
Published AI/BI dashboards carry a companion Genie Agent with the Enable Genie toggle on by default, so viewers get an Ask Genie button without anyone building one. Viewers have to be workspace members, and it is not in external embedding.
The governance decision: answers come from the full dataset query results, so a column the dataset selects but no chart shows is still in scope. Tableau Pulse does another job, watching defined metrics and pushing digests to email, Slack and Teams, on Tableau Cloud only. Genie answers on request. Pulse reports change unasked. Replace them separately.
How the licensing shape differs
Tableau licenses people through Creator, Explorer and Viewer site roles. Databricks licenses compute, and the quotation in the table above is its whole licensing statement. Genie is the exception, and its terms moved twice. Genie Agents went pay-as-you-go on July 8, 2026, then billing for Genie One and Genie Agents was paused on July 15; usage by users is free through January 31, 2027, service principals excluded and still billed. Genie Code stays pay-as-you-go, at 150 free DBUs per user a month and a 25% discount above that.
Can Databricks AI/BI replace Tableau
The answer comes per dashboard rather than per estate, and four things decide it.
- Whether its logic already exists in SQL, or only inside the workbook.
- Who has to open it. Viewers outside your Databricks account, and viewers who each see different rows, are both answered by the publish mode and the embedding route, on Tableau row-level security to Unity Catalog.
- Whether the design leans on freeform placement, or on an interaction the AI/BI docs do not describe.
- Who maintains it in two years, and whether that person writes SQL.
Estates answer unevenly, and the exceptions cluster in a small group of heavily built dashboards. Inventory first: a difference matters only where a dashboard uses it.
Antares is a BI migration tool whose Tableau to Databricks AI/BI route is shipped, at a flat $200 per source dashboard with LLM costs billed by your own provider on top. The route list is on the Converter page.
Every Databricks cell in the tables above links the docs page it came from. The Tableau side comes from dashboard actions, site roles and Tableau Pulse.
← Back to the Tableau to Databricks Migration guide
Related Migration Resources
Frequently asked questions
Is Databricks AI/BI free?
Databricks words it as "There are no additional license fees to use AI/BI. Standard Databricks Lakehouse $DBU rates apply," so dashboards cost the SQL warehouse compute they use. Genie is metered separately, and the terms changed mid-2026: Genie Agents went pay-as-you-go on July 8, 2026, then billing was paused on July 15 and usage by users is free through January 31, 2027, with service principals still billed. Genie Code keeps the pay-as-you-go model: 150 free DBUs per user a month, then a 25% promotional discount through January 31, 2027.
Does Databricks AI/BI have LOD expressions?
Yes, since 2026. The docs name two forms in custom calculations: Fixed LOD, an aggregate with an explicit PARTITION BY, and Coarser LOD, written with AGGREGATE OVER and PARTITION BY * EXCEPT. An Include-style expression has no keyword in a dashboard calculation and is modeled in dataset SQL, though the metric view editor gained one in YAML on September 10, 2026. The full mapping is on the LOD expressions page in this cluster.
What are the limitations of Databricks AI/BI dashboards compared with Tableau?
Three surface early, as of September 2026: dual axis exists as a combo chart option rather than a property of any two measures, the canvas snaps to 12 columns instead of freeform placement, and custom visuals are Vega-Lite specifications in Public Preview with no treemap. All three are design decisions rather than data problems, which is why they come up in the first week.
How does AI/BI Genie compare with Tableau Pulse?
They do different jobs. A companion Genie Agent answers questions typed against a published dashboard's datasets, on by default, for workspace members only, and not in external embedding. Tableau Pulse follows metric definitions and pushes digests to email, Slack and Teams, and applies to Tableau Cloud only. Replacing one does not replace the other.
Does one Tableau dashboard convert to one Databricks AI/BI dashboard?
Usually yes, and that is the unit of conversion. The ceilings are roomy for a single dashboard as of September 2026: 15 pages, 100 datasets and 100 widgets per page. Workbooks that grew into a portal over the years are the ones worth splitting during the move rather than after it.
Can Databricks AI/BI replace Tableau?
It answers per dashboard, not per estate. Each platform is a bargain of a different shape: Tableau gives design freedom and a long interaction list, and asks for its own platform and skills; AI/BI sits next to the data and shares the Unity Catalog permission model, and asks you to move logic into SQL.