A semantic model is a Power BI data model: the prepared layer that holds your data tables, the relationships between them and the calculations written in DAX. Reports and dashboards query it to get their numbers. Until late 2023 Microsoft called it a dataset; it is the same thing.
What a semantic model is
A semantic model is the data layer behind a Power BI report. It sits between your raw data sources and the charts a reader sees, and it holds three things: the data tables, the relationships that join those tables and the calculations. The visuals are kept separate; the model is the part that supplies the numbers.
It's called "semantic" because it adds business meaning to raw data. Friendly table and column names, the relationships between tables, DAX calculations and number formatting all turn unlabelled rows into figures a business reader understands. Microsoft's guidance on semantic models in the Power BI service describes a semantic model as the source "ready for reporting".
A useful way to picture it: the model is the prepared kitchen and the report is the plated dish. Build the model well and every report drawn from it is quick to make and consistent. That's why Power BI work is, in large part, modelling work.
Semantic model or dataset? The 2023 rename
This is the single biggest source of confusion, so it's worth stating plainly. Semantic model is the current name for what Power BI called a dataset. Microsoft renamed datasets to semantic models in late 2023, across both the Power BI service and Microsoft Fabric.
Nothing about the object itself changed - only the label. Plenty of articles, older menus and parts of the product still say "dataset", and they are describing exactly the same thing. If a guide written before 2024 talks about a Power BI dataset, read it as a semantic model and you'll be right.
The new name was chosen to be clearer about what the thing is. A "dataset" sounds like a passive file of data; a "semantic model" better describes an active layer of tables, relationships and meaning that reports query.
Data model vs semantic model
People often ask whether a data model and a semantic model are different things. In Power BI, they are not. The two terms point at the same object.
"Data model" is the general, tool-agnostic term: any structured set of tables, relationships and calculations is a data model. "Semantic model" is Microsoft's specific product name for that model in Power BI, Analysis Services and Fabric. So when someone says "the Power BI data model", they mean the semantic model. The phrase you use makes no practical difference.
What a semantic model contains
A semantic model is built from four kinds of ingredient. Each one adds a layer of structure or meaning on top of the raw data.
- Tables: the data itself, either imported into the model or connected to a live source - the choice between Import and DirectQuery.
- Relationships: the links that join the tables so a filter on one flows to another. They are best arranged as a star schema, the table layout Microsoft recommends for Power BI.
- Calculations: measures and calculated columns written in DAX, the formula language of the model. If the difference between the two matters to you, see measures vs calculated columns.
- Hierarchies and formatting: groupings such as Year to Quarter to Month, plus the number and currency formats that make output readable.
Put together, those ingredients are what a chart actually talks to. Drop a field onto a visual and Power BI answers the request from the model, not from the original source files.
Where it lives, and why one model serves many reports
In Power BI Desktop, the semantic model and the report live together in a single .pbix file. While you build, the two feel like one thing, because you're shaping the model and the visuals in the same window.
They separate when you publish. In the Power BI service the semantic model can be its own item, distinct from the reports built on it. One published model can then power many reports - covered in Power BI Desktop vs the Power BI service.
That separation is the point of a semantic model. It becomes the single, governed source of truth: define a measure such as Total Revenue once in the model, and every report built on it returns the same number. Without a shared model, ten reports can produce ten subtly different "revenue" figures.
A solid semantic model is where good Power BI begins Reading about the semantic model is a useful start; building one that stays fast, accurate and easy to extend is a skill worth investing in. Our two-day, hands-on Power BI Masterclass teaches data modelling the practical way, building a real model from messy source data rather than learning the theory in the abstract. If you are still comparing courses, our guide to choosing Power BI training in the UK walks through what to look for.
