Momentum One Academy · glossary
Semantic Model
Definition
A semantic model is a curated, logical representation of an analytical domain. It gives facts and dimensions business meaning through relationships, measures, friendly names, and metadata, so people and tools can analyze the same definitions instead of interpreting raw tables independently. In Fabric, “semantic model” is the current name for the Power BI dataset item. 1
Where it sits in Fabric
The semantic model is the consumption and business-logic layer between prepared data in OneLake, a lakehouse, Warehouse, SQL analytics endpoint, or mirrored database and the reports and AI experiences that use it. Fabric can create models from these items, and a model can use Direct Lake, Import, or DirectQuery storage modes. 1 2
Direct Lake is a storage-mode choice within this layer, not a replacement for the model. It reads Delta-table data from OneLake into the Power BI engine and is intended for large Fabric data sources where copying the full volume into an Import model is impractical. 3
When to use it
Use a semantic model when several reports, analysts, or AI consumers need consistent metrics such as Net Revenue, shared relationships, and governed terminology. Direct Lake is a strong fit for a large, lake-centric solution with frequent source updates; Import or DirectQuery may be more appropriate when self-service preparation or source-query behavior is the priority. 3
Practical example
A retail team publishes curated sales and product Delta tables in a lakehouse. In Module 8, it creates a semantic model, relates the fact and dimension tables, defines a certified Net Sales measure, and builds a report on top. The same model can be connected to a Fabric data agent, which uses the model’s schema and metadata to generate DAX for questions such as “What were net sales by region last quarter?” 2 4
Production caution
Do not assume that a technically valid model guarantees reliable AI answers or stable performance. Give tables, columns, and measures business-friendly names; configure the relevant AI data schema and dependent objects; and validate permissions, capacity, and representative queries. Data Agent’s semantic-model query generation relies on model metadata and Prep for AI configuration, while Direct Lake can fall back to DirectQuery when capacity limits or unsupported features apply. 3 4
Module 8 connections
Treat the semantic model as Module 8’s contract between Fabric data and its consumers: Direct Lake supplies an efficient access path, Copilot for Fabric helps users work with Power BI content, and Data Agent turns governed model metadata into conversational analysis. Keep those terms distinct: the semantic model defines meaning; the other capabilities consume or operate around that meaning. 4 5
Build a production-ready model with Momentum One Academy’s Fabric Production Workshop.
References
Plan your workshop