Introduction

Power BI has evolved from a self service reporting tool into a full enterprise analytics engine. The biggest shift in this evolution is the introduction of Fabric Semantic Models, which extend and modernize the classic Power BI dataset. When combined with Power BI Premium and the new Fabric architecture, semantic models now act as the central brain for analytics across SQL, Lakehouses, streaming data, AI workloads, and enterprise reporting.

In many ways, semantic models are becoming the enterprise “source of truth” layer. They govern relationships, measures, security, and modeling logic, while separating storage from compute so organizations can scale without rewriting code. For SQL professionals, the model behaves like a governed analytical layer that sits on top of multiple data sources and blends tabular performance with lake scale.

This article provides a deep dive into what Fabric Semantic Models are, how they work inside Power BI Premium, and why they matter for enterprise data teams today. Later parts in this series will cover Direct Lake, Real Time Semantic Models, Auto Aggregations, and AI Modeling.

---

What Are Fabric Semantic Models

Fabric Semantic Models replace the traditional Power BI dataset with a more flexible, scalable, and AI ready version of the Tabular engine. They support multiple storage modes, lake integration, real time sources, and cloud scale modeling.

A semantic model defines:
• Tables and relationships
• Measures and DAX logic
• Row level security
• Aggregations
• Source configurations
• Data lineage across the Fabric ecosystem

The key difference is that semantic models are no longer tied only to imported or direct query data. They can sit directly on top of Delta Lake tables, shortcuts, streaming data, and even AI generated tables.

(Inline reference: Microsoft states that semantic models unify storage and compute across the Fabric OneLake architecture, https://learn.microsoft.com/fabric/get-started/semantic-models-overview)

---

How Power BI Premium Enhances Semantic Models

Power BI Premium provides the capacity layer where semantic models live, refresh, and serve report queries. With Premium, you get features that materially improve performance and enterprise governance.

Key Premium Enhancements

1. Large model support

Premium allows semantic models to reach extremely large in memory footprints compared to shared capacity (Microsoft Fabric capacity limits). This matters for enterprise workloads where import models or hybrid tables need space.
https://learn.microsoft.com/power-bi/enterprise/service-premium-capacity

2. Incremental refresh and real time

Models in Premium use incremental refresh policies backed by the Fabric engine. Combined with Direct Lake or hybrid tables, refresh operations can target only changed partitions instead of full reloads.
https://learn.microsoft.com/power-bi/connect-data/incremental-refresh-overview

3. Enhanced concurrency

Premium dedicates capacity so multiple users hit the same semantic model without throttling. SQL teams will recognize this as similar to isolating workloads on dedicated compute.

4. Direct Lake enablement

Direct Lake only works in Premium or Fabric capacities since it relies on the VertiPaq engine warming pages directly from Delta Lake.
https://learn.microsoft.com/fabric/data-engineering/direct-lake-overview

5. XMLA read write access

Developers can script, automate, deploy, and back up semantic models using SSMS, Tabular Editor, or PowerShell (reference: XMLA Endpoint documentation).
https://learn.microsoft.com/power-bi/enterprise/service-xmla-overview

6. Endpoint universality

Semantic models can be reused in:
• Power BI reports
• Paginated reports
• Excel PivotTables
• PowerPoint storyboards
• Power BI Copilot modeling
• SQL endpoint tooling

---

How Fabric Semantic Models Work in the Architecture

A Fabric semantic model sits directly on top of OneLake, which is essentially a unified data lake with Delta Lake as the consistent storage format. Once a model is published into a Premium or Fabric capacity, the engine can:

• Read from import tables (VertiPaq compressed memory)
• Read from Direct Lake tables (Delta Lake)
• Query external SQL sources in DirectQuery mode
• Blend all three using hybrid tables
• Layer aggregation tables on top of large fact tables

This hybrid pipeline is the exact reason semantic models are replacing classic datasets in Microsoft’s long term roadmap.

(Inline reference: Fabric architecture diagrams identify semantic models as the universal consumption layer for BI and AI, https://learn.microsoft.com/fabric/get-started/microsoft-fabric-overview)

---

Why Semantic Models Matter for SQL Professionals

For SQL professionals, the semantic model is the bridge between raw data systems and analytics.

Model centralization

You no longer need to build star schemas individually in every reporting tool. One semantic model feeds all tools.

Performance at scale

VertiPaq and Direct Lake deliver millisecond performance on lake storage without duplicating ETL.

Security once, applied everywhere

RLS cascades automatically to Power BI, Excel, and Fabric.

AI integration

Copilot consumes semantic model metadata to generate measures, analysis, and insights
(https://learn.microsoft.com/fabric/copilot/copilot-power-bi-overview).

DevOps and versioning

Semantic models can be checked into GitHub, deployed, validated, and versioned like SQL code through workspace Git integration.

---

Deep Dive: Storage Modes Inside Semantic Models

Semantic models allow several storage modes. These modes determine performance, scalability, and refresh needs.

Import Mode

Tables load into VertiPaq and compress highly.
Best for static or slowly changing fact tables.

DirectQuery Mode

Queries data at runtime.
Best for operational real time dashboards hitting SQL engines.

Direct Lake Mode

Reads Delta Lake files directly into VertiPaq without imports or refresh.
Best for Fabric Lakehouses, high speed analytics, low latency reporting.
(Full breakdown coming in Part 2.)

Hybrid Tables

Combines Import for historical partitions and DirectQuery for today’s data.
Best for massive fact tables with mixed freshness needs.

This flexibility is why the semantic model is replacing the old dataset engine.

---

Workshop: Build a Semantic Model in Fabric with Hybrid Storage and XMLA

This hands on walkthrough can be followed in any Premium or Fabric capacity.

Step 1. Create a Lakehouse and Load Data

• Create a new Lakehouse
• Upload CSV or Parquet files to the Files area
• Convert them into Delta using “Create table”
• Confirm they appear under “Tables”

Step 2. Create a New Semantic Model

• Click “New semantic model”
• Select your fact and dimension tables
• Fabric auto detects relationships

Step 3. Configure a Hybrid Table

• Open fact table properties
• Set historic partitions to Import
• Set today’s partition to DirectQuery
• Apply incremental refresh

• Refresh rows from last 1 day
• Store 5 years of history

Step 4. Add Measures

Total Sales :=
SUM(FactSales[Amount])

New Customers :=
CALCULATE(
    DISTINCTCOUNT(DimCustomer[CustomerID]),
    FILTER(DimCustomer, DimCustomer[FirstPurchaseDate] = TODAY())
)

Step 5. Enable XMLA Read Write

• Go to Workspace Settings
• Enable XMLA Read Write
• Connect with Tabular Editor via:
powerbi://api.powerbi.com/v1.0/myorg/YourWorkspaceName

Step 6. Validate the Model

Check for:
• Cardinality
• Unused columns
• Auto date tables disabled
• DAX optimization using Best Practice Analyzer

Step 7. Build a Report

• Open Power BI Desktop
• Connect to the semantic model
• Create visuals
• Publish into the Premium workspace

---

Summary

Fabric Semantic Models represent a major leap in how Power BI handles enterprise modeling. They unify storage modes, lakehouses, SQL engines, real time analytics, and AI under one engine. When hosted in Power BI Premium, you gain large model storage, Direct Lake, XMLA automation, and enterprise grade performance.

This foundation sets up the next articles in this series:
Part 2. Direct Lake
Part 3. Real Time Semantic Models
Part 4. Auto Aggregations
Part 5. AI Powered Modeling

---

References (Clickable Links)

• Microsoft Fabric Documentation. Semantic Models Overview
https://learn.microsoft.com/fabric/get-started/semantic-models-overview

• Microsoft Power BI Premium Capacity Documentation
https://learn.microsoft.com/power-bi/enterprise/service-premium-capacity

• XMLA Endpoint Programming Guide. Microsoft Learn
https://learn.microsoft.com/power-bi/enterprise/service-xmla-overview

• Fabric Lakehouse and Delta Lake Technical Overview
https://learn.microsoft.com/fabric/data-engineering/lakehouse-overview

• Fabric Copilot Modeling Capabilities
https://learn.microsoft.com/fabric/copilot/copilot-power-bi-overview

• VertiPaq Compression Engine Whitepaper
https://learn.microsoft.com/power-bi/guidance/vertipaq
(Note: Microsoft no longer hosts the old PDF, this is the official replacement.)

• Power BI Incremental Refresh Deep Dive
https://learn.microsoft.com/power-bi/connect-data/incremental-refresh-overview