Tableau to Microsoft Fabric Migration: What Moves, What Changes, What Waits

Fabric is a bigger word than the migration. Microsoft Fabric is one SaaS platform holding several workloads — data engineering, warehousing, data factory, real-time intelligence and Power BI — over a single store called OneLake. Power BI is the BI workload inside it. So a Tableau estate does not land in "Fabric" in general: reports and semantic models land in the Power BI experience, which now lives in Fabric workspaces. That is the honest scope of a Tableau to Microsoft Fabric migration: the report work is the same report work, with more options underneath it.

Tableau concepts in the Fabric world

Most of the mapping is the ordinary Tableau-to-Power-BI mapping, because that is where the content lands. The rows worth remembering are where Fabric adds a second answer, or where a Tableau habit has no home.

One vocabulary trap: in Power BI's own language a "dashboard" is a pinned-tile artifact in the service, not what a Tableau user means. The target of a converted Tableau dashboard is a report page.

OneLake is the part that is genuinely new

Every Fabric tenant gets exactly one OneLake, which you cannot delete or duplicate. It is built on Azure Data Lake Storage and stores tables in open formats, Delta Parquet or Iceberg, so several engines read one copy instead of each making its own. Shortcuts point at data that already lives elsewhere, without moving it.

Set that against a Tableau estate: an extract per published data source, on its own schedule, duplicated wherever someone needed a copy. Nothing forces you to consolidate during the migration. But when an inventory shows the same source extracted again and again, OneLake is the answer to a question the estate has been asking for years.

Import, DirectQuery and Direct Lake

Storage mode is the decision Tableau made as "extract or live". Fabric adds a third option.

Microsoft calls that Direct Lake refresh framing: it updates metadata rather than copying data, which is why it takes seconds where an import refresh takes minutes. The prerequisite is the catch — the data has to sit in OneLake as Delta tables, in a lakehouse or warehouse, which is a data-platform project, not a reporting one. A migration arriving before that project is finished uses Import or DirectQuery, mirroring what the workbook already did. That is a legitimate landing place, not a compromise.

Licensing changes shape, not just price

Tableau licenses people: Creator, Explorer, Viewer, per seat. Fabric bills compute. Capacity comes as F SKUs bought through Azure, billed per second with an optional yearly reservation, and one capacity serves every workload, so pipelines and report queries draw on the same units. A 60-day trial capacity at F64-equivalent size is available.

Per-user licensing does not disappear. Creating and sharing Power BI content still needs a Pro or Premium Per User license, and viewers need one too — unless the workspace sits on an F64 or larger capacity, where a Free license plus the viewer role is enough. Below F64, every reader is licensed individually. Non-Power BI Fabric items — lakehouses, warehouses, notebooks — need an F or trial capacity regardless. Price the target the way you will run it: seats for authors, capacity for readers and refreshes — and re-read Microsoft's licensing page on the day you build the business case, because these rules move.

Workspaces, pipelines and governance

Tableau projects nest, and permissions flow down the tree. Fabric workspaces are flat containers with four roles — Admin, Member, Contributor, Viewer — and each belongs to one capacity. An estate that used deep project hierarchies for both filing and permissions has to split the jobs: workspaces carry permissions, and domains group workspaces by business area.

Two mechanisms have no Tableau habit to inherit. Deployment pipelines give content development, test and production stages, with rules that repoint data source settings as an item moves between them. Sensitivity labels attach to items and follow the data into exports. Both are cheaper to decide during the migration than to retrofit afterwards, item by item.

Sequence the BI layer first

The tempting plan is everything at once: move the dashboards, build the lakehouse, switch the models to Direct Lake. The plan that finishes is narrower. Migrate the report layer first — reports and semantic models reading the sources you already have — and treat the data-platform work as the next project rather than a prerequisite.

Three reasons. First, a BI-layer migration has a clean ending: when the report pages are accepted and users have moved, the Tableau contract can lapse, and that is the date the business actually cares about. Second, doing both at once gives every mismatched number two suspects, the conversion or the new data path, and telling them apart costs more than sequencing would have. Third, waiting costs nothing structural: adopting Direct Lake later changes what a model reads, not what the report pages look like.

The one thing worth doing early is the inventory, because it sets the order: which workbooks are still opened, which sources repeat, which dashboards everybody actually uses.

How Antares fits in

This is the product-dependent section. Everything above stays true whichever tool you use.

Antares is a BI migration tool with two modules. The Analyzer connects to the Tableau environment and produces the inventory described above — fully deterministic, no LLM involved, and free. It reads metadata only, never business data. The Converter converts dashboards to the target platform and deploys them; Tableau to Power BI, including deployment into Fabric, is shipped today. It is deterministic-first with guardrailed, validated AI steps, at a flat $200 per migrated dashboard, with LLM costs on top — typically under $20 per dashboard, and your own private LLM endpoints can be used. The unit is the source dashboard: one Tableau dashboard becomes one Power BI report page, and the worksheets it uses ride along.

What it does not do matters as much for planning. The conversion is like-to-like, not pixel-perfect and not a redesign — parity across platforms with different capabilities does not exist. And it is not a data migration tool: it converts the dashboard's semantic data model, but it does not move your data into OneLake or change anything in the warehouse. That is the follow-on project, which is where this page argues it belongs. Run the free Analyzer to see what your estate holds before you size the capacity.

Related reading: Tableau Server to Power BI Service and data sources to semantic models. Primary sources: Microsoft on Fabric, OneLake, Direct Lake and licenses and capacity, plus Tableau on projects.