Home > Docs > Reference Architectures > Small-Medium Enterprise

🏢 Small-Medium Enterprise Reference Architecture¶

Last Updated: 2026-05-05 | Version: 1.0.0

📑 Table of Contents¶

• 🎯 Architecture Overview
• 🏗️ Architecture Diagram
• 📦 Component Table
• 📏 Capacity Sizing Guidance
• 🔒 Network Architecture
• 💰 Cost Estimation Framework
• 🚀 Deploy This Architecture
• ⚖️ Tradeoffs and Limitations
• 📚 References

🎯 Architecture Overview¶

This architecture is designed for small-to-medium teams (5–20 data practitioners) running their first production Fabric deployment. A single Fabric capacity hosts 2–3 workspaces following a Dev/Test/Prod promotion model. Data flows through a medallion lakehouse (Bronze → Silver → Gold) within a single OneLake instance, and Power BI reports connect via Direct Lake for sub-second analytics without data duplication. Governance is handled at the workspace level using Fabric's built-in RBAC, sensitivity labels, and optional Purview integration.

🏗️ Architecture Diagram¶

graph TB
    subgraph Sources["Data Sources"]
        S1[Operational DBs]
        S2[Flat Files / APIs]
        S3[SaaS Applications]
    end

    subgraph FabricCapacity["Fabric Capacity (F64)"]
        subgraph DevWorkspace["Dev Workspace"]
            DEV_LH[Lakehouse<br/>Dev]
            DEV_NB[Notebooks<br/>Dev]
        end

        subgraph ProdWorkspace["Prod Workspace"]
            subgraph OneLake["OneLake"]
                BRONZE[Bronze Lakehouse<br/>Raw Ingestion]
                SILVER[Silver Lakehouse<br/>Cleansed & Validated]
                GOLD[Gold Lakehouse<br/>Business KPIs]
            end

            PIPE[Data Pipelines<br/>Orchestration]
            NB[Spark Notebooks<br/>Transformations]
            SQL[SQL Analytics Endpoint<br/>Ad-hoc Queries]
            SEM[Semantic Model<br/>Direct Lake]
        end

        subgraph BIWorkspace["BI Workspace"]
            PBI[Power BI Reports]
            DASH[Dashboards]
        end
    end

    subgraph Consumers["Report Consumers"]
        U1[Business Users]
        U2[Executives]
        U3[Analysts]
    end

    S1 --> PIPE
    S2 --> PIPE
    S3 --> PIPE
    PIPE --> NB
    NB --> BRONZE
    BRONZE --> NB
    NB --> SILVER
    SILVER --> NB
    NB --> GOLD
    GOLD --> SQL
    GOLD --> SEM
    SEM --> PBI
    SEM --> DASH
    PBI --> U1
    PBI --> U2
    DASH --> U3
    DEV_LH -.->|Deployment<br/>Pipeline| BRONZE

📦 Component Table¶

📏 Capacity Sizing Guidance¶

Sizing considerations:

• Spark workloads consume the most CUs — schedule heavy transforms during off-peak hours
• Direct Lake is far more CU-efficient than Import or DirectQuery for BI
• Pause capacity during non-business hours to reduce cost by 40–60%
• Start with F32 and monitor utilization via Workspace Monitoring before scaling

🔒 Network Architecture¶

For small-medium deployments, a simplified network posture balances security with operational simplicity:

When to add network isolation: If compliance requirements mandate private connectivity (HIPAA, FedRAMP), add private endpoints and a VNet data gateway. See Network Security.

💰 Cost Estimation Framework¶

Cost optimization strategies:

1. Pause/Resume — Pause capacity during nights and weekends (saves 40–60%)
2. Reservations — 1-year commitment saves ~40% on capacity
3. Smoothing — Fabric's 24-hour CU smoothing lets you burst without immediate throttling
4. Direct Lake — Eliminates Import refresh CU cost entirely

🚀 Deploy This Architecture¶

Infrastructure as Code¶

Step-by-Step Tutorials¶

Deployment Commands¶

# Deploy the capacity and supporting resources
az deployment sub create --location eastus2 \
  --template-file infra/main.bicep \
  --parameters infra/environments/dev/dev.bicepparam

# Validate before deploying
az deployment sub what-if --location eastus2 \
  --template-file infra/main.bicep \
  --parameters infra/environments/dev/dev.bicepparam

⚖️ Tradeoffs and Limitations¶

📚 References¶