What is a Semantic Model in Power BI?
A semantic model in Power BI (previously called a dataset) is the data layer that stores tables, defines relationships between them, and holds the DAX measures used across reports. It is the single source of truth that one or more reports connect to, so a change to a measure or relationship in the semantic model updates every report built on it. Semantic models can be reused across multiple reports and shared with other teams through the Power BI service.
A semantic model in Power BI is the underlying data structure, made up of tables, relationships, and DAX measures, that reports are built on top of.
From raw data to a semantic model
Data does not become useful for reporting the moment it is imported into Power BI. It needs structure: tables must be connected through relationships, business logic needs to be expressed as reusable calculations, and access rules need to be defined so the right people see the right data. The semantic model is where all of this comes together. It sits between the raw, cleaned data produced by Power Query and the visual reports end users interact with.
Practically, a semantic model contains the tables loaded into Power BI, the relationships drawn between them (for example, linking a Sales table to a Products table through a shared product ID), and the DAX measures that calculate totals, ratios, and other business metrics. It can also include row-level security rules that restrict which rows a given user sees, and metadata such as formatted display names for fields.
Why it is called “semantic”
The word “semantic” refers to meaning: the model does not just store numbers, it defines what those numbers mean and how they relate to each other. A column labeled amount is just a number; a semantic model turns it into “Total Sales Amount, related to a specific customer, product, and date, aggregatable by region.” This layer of meaning is what lets someone build a report by dragging fields onto a canvas without understanding the underlying calculations, because the model already encodes the correct logic.
Microsoft renamed the object from “dataset” to “semantic model” in 2023 to better reflect this role. The rename did not change any functionality; it simply clarified that the object is more than a flat table of data.
One model, many reports
A key benefit of the semantic model concept is reuse. Rather than rebuilding relationships and measures separately for every report, a single, well-built semantic model can support several reports across different teams. A finance team might build a cash flow report and an executive team a summary dashboard, both pulling from the same underlying sales semantic model. When a measure needs correcting, such as adjusting how discounts are calculated, updating it once in the semantic model fixes every connected report simultaneously.
This reuse depends on where the model lives. A semantic model published to a Power BI workspace can be shared with other workspaces and connected to by report authors who did not build the model themselves, provided they have the right permissions.
Composite models and combining sources
For larger organizations, data sometimes lives across several semantic models owned by different teams. Composite models let a report combine tables from more than one semantic model, or mix Import mode with DirectQuery connections, without merging everything into one giant model first. This is a more advanced pattern and requires attention to how relationships and refresh timing interact across sources, since inconsistent refresh schedules between combined models can produce misleading results if not managed carefully.
Our tip: Treat your semantic model as a shared product, not a one-off build for a single report. Naming measures clearly and documenting what each one calculates saves significant time once more than one report starts relying on the same model.
Frequently asked questions
Why did Microsoft rename 'dataset' to 'semantic model'?
Microsoft renamed dataset to semantic model in 2023 because 'dataset' was misleading; it suggested a simple table of raw data, while it actually contains relationships, DAX measures, security rules, and business logic. 'Semantic model' better describes what the object does: it gives raw data meaning and structure so it can be queried consistently across multiple reports. The underlying functionality did not change, only the name, so older documentation and community content may still refer to datasets.
Can multiple reports use the same semantic model?
Yes, and this is one of the main advantages of separating the data layer from the report layer. A single semantic model, containing the core sales tables, relationships, and measures, can serve a sales dashboard, a finance report, and an executive summary simultaneously. Any correction to a measure in the semantic model, such as fixing a profit margin formula, automatically applies to every report connected to it, rather than requiring the fix in several places.
What is the difference between a semantic model and a report in Power BI?
The semantic model is the data and logic layer: tables, relationships, calculated columns, and measures written in DAX. The report is the visual layer built on top: charts, tables, slicers, and page layout. One semantic model can support many reports, but a report always depends on exactly one semantic model (unless using composite models that combine several). Separating the two means data structure changes and visual design changes can happen independently.
What is a composite model in Power BI?
A composite model combines data from more than one semantic model, or mixes Import and DirectQuery connections, into a single report. This lets you bring together data owned by different teams, each maintaining its own semantic model, without duplicating tables or waiting for a single master model to be built. Composite models are more advanced and require careful relationship management, since combining sources with different refresh behavior can introduce inconsistencies if not configured carefully.