A semantic model sits between stored data and the people or tools that ask questions of it. It names business entities, defines how tables relate, and stores calculations such as revenue, margin, or active customer once, so every report uses the same logic. In Microsoft Power BI and Fabric, the term refers to the model that reports connect to, which was called a dataset until 2023. More broadly, a semantic model is the set of definitions inside a semantic layer.
For example, a semantic model might define net revenue as gross sales minus returns and discounts, and link orders to customers and regions. AI assistants and agents increasingly query data through semantic models, which helps them apply the right definitions. A shared meaning does not by itself show that the underlying data suits a particular AI task, so teams still check the data each task depends on.