Azure Analytics Platform Architecture Guide: Fabric, Synapse & Lakehouse Patterns

Azure Analytics Platform Architecture Guide (2026)

The modern Microsoft Azure analytics platform in 2026 is anchored on Microsoft Fabric — Microsoft's unified SaaS analytics platform integrating Data Engineering, Data Warehouse, Data Science, Real-Time Intelligence, Microsoft Power BI, and OneLake. This is the working enterprise Azure analytics architecture guide EPC Group uses for Fortune 500 deployments. The reference model below assumes Microsoft Fabric as the system of record, Microsoft Azure Synapse as a deprecation candidate for new architecture (Microsoft is steering new development to Microsoft Fabric), and coexistence patterns with Snowflake, Databricks, AWS, and Google Cloud where they exist.

EPC Group has delivered Microsoft Azure analytics architectures since the original Microsoft Azure Synapse Analytics era and through Microsoft Fabric general availability. Practice depth includes Microsoft Power BI Project Crescent beta history, Microsoft Azure Synapse migration to Microsoft Fabric, and SAP- and Oracle-anchored data platforms across the Fortune 500 portfolio.

TL;DR — Modern Azure Analytics Reference Architecture

Layer	Microsoft Component
Data sources	Microsoft SQL Server, SAP, Salesforce, Snowflake, third-party SaaS
Ingestion	Microsoft Fabric Data Factory, Mirroring, Eventstream
Storage	OneLake (Microsoft Fabric data lake)
Compute	Microsoft Fabric F-SKU capacity (F2-F2048)
Data Warehouse	Microsoft Fabric Warehouse (T-SQL)
Lakehouse	Microsoft Fabric Lakehouse (Spark SQL)
Real-time	Microsoft Fabric Real-Time Intelligence (KQL)
ML / AI	Microsoft Fabric Data Science + Microsoft Azure OpenAI
BI	Microsoft Power BI (DirectLake mode)
Governance	Microsoft Purview
Security	Microsoft Sentinel + Microsoft Defender for Cloud
AI productivity	Microsoft Power BI Copilot + Microsoft 365 Copilot

Phase 1 — Microsoft Fabric Foundation

Capacity Sizing

Tier	Use Case
F2-F8	Pilot / development
F16-F32	Mid-market single-department
F64	Microsoft Power BI Copilot enabled, mid-market enterprise
F128-F256	Enterprise multi-department
F512+	Fortune 500 multi-region

Microsoft Power BI Copilot requires F64+ capacity, which is the threshold most enterprises cross when Copilot becomes strategic. EPC Group's standard recommendation is to size capacity for the 80th-percentile sustained load, lean on Microsoft Fabric Smoothing for burst, and rebalance quarterly under managed services.

OneLake Topology

EPC Group's standard medallion architecture: Bronze (raw source data, minimal transformation, preserve all upstream fidelity), Silver (cleaned, joined, sensitivity-labeled, conformed dimensions and SCD handled), and Gold (business-aligned star-schema marts, DirectLake-mode-ready, Microsoft Power BI Copilot grounding aligned).

Capacity Strategy

Reserved one-year or three-year for production (-31% / -41% versus pay-as-you-go). Pay-as-you-go for development and spike workloads. Microsoft Fabric Mirroring for source-side replication into OneLake without ETL. Eventstream for real-time streaming ingestion.

Phase 2 — Data Sources Integration

Microsoft First-Party Sources

Microsoft 365 Graph data (Microsoft SharePoint, Microsoft OneDrive, Microsoft Teams). Microsoft Dynamics 365 via Microsoft Dataverse Mirroring. Microsoft Power Platform Microsoft Dataverse. Microsoft Azure SQL Database, Microsoft Azure SQL Managed Instance, Microsoft Azure Cosmos DB, Microsoft Azure Synapse Analytics, Microsoft Azure Data Lake Storage Gen2. Microsoft Sentinel logs.

Third-Party Sources

SAP S/4HANA and SAP ECC via the SAP CDC connector. Oracle E-Business Suite and Oracle Fusion Cloud. Salesforce via the native connector. Snowflake via Microsoft Fabric Shortcut (no data duplication). Databricks via OneLake Delta Lake interop. AWS S3 and Amazon Redshift via Microsoft Fabric Shortcut. Google BigQuery. 200+ Microsoft first-party connectors plus 500+ third-party connectors.

Real-Time Sources

IoT sensors via Microsoft Azure IoT Hub. HL7 / FHIR healthcare feeds. Trading systems (Bloomberg, Charles River). E-commerce events. Application logs via Microsoft Sentinel.

Phase 3 — Microsoft Fabric Workloads

Data Engineering

Microsoft Fabric Spark notebooks (PySpark, SQL, R). Microsoft Fabric Data Factory pipelines. Microsoft Fabric Dataflow Gen2 (Power Query at scale). Delta Lake format (open table format) so the resulting data is portable to Databricks if needed.

Data Warehouse

T-SQL warehouse compatible with traditional BI tooling. DirectLake-mode Microsoft Power BI semantic models for sub-second query response. Microsoft Purview lineage native. Row-level and column-level security.

Real-Time Intelligence

KQL Database (Kusto Query Language). Eventstreams for ingestion. Real-time dashboards via Microsoft Power BI DirectLake. Microsoft Fabric Activator for alert routing into Microsoft Teams, Microsoft Power Automate, or external systems.

Data Science

Microsoft Fabric Synapse Data Science. MLflow tracking and model registry. AutoML and custom training. Model deployment via PromptFlow. Vector search via Microsoft Azure AI Search integration.

Microsoft Power BI

Microsoft Power BI semantic models. DirectLake mode (60-80% query latency reduction versus Import). Microsoft Power BI Copilot enabled (F64+). Composite models (Import + DirectQuery + DirectLake). Microsoft Purview sensitivity labels native.

Phase 4 — Governance

Microsoft Purview Data Governance

Microsoft Purview Data Map (multi-cloud lineage). Microsoft Purview Catalog (data discovery). Microsoft Purview Information Protection (sensitivity labels). Microsoft Purview DLP (data loss prevention). Microsoft Purview Compliance Manager (attestation). Microsoft Purview AI Hub (Microsoft Copilot governance).

Microsoft Sentinel SOC

Microsoft Fabric audit log ingestion. Microsoft Power BI activity logs. Microsoft Sentinel custom analytics rules for analytics-related security events. Microsoft Defender for Cloud integration.

Audit-Ready Documentation

Architecture Decision Record (ADR) per material decision. Customer-Responsibility Matrix continuously maintained. Microsoft Compliance Manager attestation per applicable framework. Plan-of-Action-and-Milestones for any control gap.

Phase 5 — Microsoft Copilot Integration

Microsoft Power BI Copilot

Natural-language queries against semantic models. AI-generated narrative summaries. DAX measure suggestions. Sensitivity-label-aware grounding so Copilot does not ground on Restricted-tier content. Microsoft Purview AI Hub monitoring.

Microsoft 365 Copilot Integration

Microsoft Copilot Chat anchored to Microsoft Fabric semantic models. Microsoft Copilot in Microsoft Excel for analytics. Microsoft Copilot in Microsoft PowerPoint for executive briefs. Microsoft Copilot Studio agents grounded on enterprise data.

Custom Microsoft Azure OpenAI Applications

Retrieval-Augmented Generation (RAG) on enterprise content. Microsoft Azure AI Search for vector, keyword, and hybrid retrieval. Microsoft Azure AI Foundry for agent development. Microsoft Power Platform integration.

Reference Architecture — Mid-Market Fortune 500

Standard mid-market profile: 5,000-15,000 users, 50TB-500TB analytics data, Microsoft Fabric F128 reserved capacity, 200-500 active analytics users, Microsoft 365 E5 plus Microsoft 365 Copilot, Microsoft Power BI Copilot enabled, 5-10 Microsoft Copilot Studio custom agents. Annual cost: $5M-$15M all-in (capacity plus licenses plus managed services).

Reference Architecture — Fortune 500 Multi-Region

Standard Fortune 500 profile: 25,000-100,000 users, 500TB-5PB analytics data, Microsoft Fabric F512+ reserved capacity (multi-region), 1,000-5,000 active analytics users, Microsoft 365 E5 enterprise-wide, Microsoft 365 Copilot for 30-50% of users, 50+ Microsoft Copilot Studio custom agents. Annual cost: $25M-$100M+ all-in.

Industry-Specific Patterns

Healthcare (HIPAA)

The architecture includes Microsoft Customer Lockbox enabled for any Microsoft-side access, Restricted-PHI sensitivity-tier rollout to PHI-tagged content, Microsoft Fabric Mirroring on Epic Clarity, Oracle Health Cerner Cogito, or athenahealth source databases as applicable, HEDIS and CMS Star Ratings semantic-model maintenance, and Joint Commission audit-ready analytics packages.

Financial Services (FINRA, SEC, SOX)

The architecture includes Microsoft Information Barriers operations across investment-banking, equity research, sales/trading, and asset-management segments, Restricted-MNPI sensitivity-tier rollout, SEC Rule 17a-4 retention configuration, FINRA Rule 3110 supervised analytics with trader-level Row-Level Security, and annual SOC 2 Type II support.

Government (FedRAMP, CMMC)

The architecture includes Microsoft 365 GCC or GCC High deployment, FedRAMP-aligned continuous monitoring, CMMC Level 2 or Level 3 documentation, and CAC/PIV authentication for Microsoft Power BI Copilot access.

Manufacturing (SAP-Anchored)

The architecture includes Microsoft Fabric SAP CDC connector for SAP S/4HANA and SAP ECC, Real-Time Intelligence for OPC UA and OSI PI streaming, Microsoft Defender for IoT integration, and OEE semantic-model maintenance.

Capacity Strategy in Detail

Microsoft Fabric F-SKU capacity is the single largest line item in the Microsoft Azure analytics architecture for most enterprises. Right-sizing matters. Under-sizing throttles workloads at peak; over-sizing wastes 30-50% of the annual capacity budget. EPC Group's capacity strategy methodology captures 90 days of baseline utilization, identifies the 80th-percentile sustained peak, sizes capacity to that peak with smoothing headroom, and rebalances quarterly under managed services. The Reserved Instance and Savings Plan portfolio is reviewed against the rebalanced capacity recommendation each quarter.

Microsoft Fabric Smoothing distributes burst loads over 24-hour windows. Smoothing is helpful for genuinely-bursty workloads (executive dashboard refresh at 8am, end-of-day batch refresh) but masks sustained over-consumption that can lead to surprise throttling. The capacity-metrics app is the right place to watch for sustained over-consumption that smoothing is masking; EPC Group monitors that signal continuously under managed services.

The Reserved Instance and Savings Plan calculus depends on capacity stability. Customers whose capacity is stable for 12+ months should be on one-year Reserved Instances at minimum (-31% versus pay-as-you-go). Customers whose capacity is stable for 36+ months should consider three-year Savings Plans (-41% versus pay-as-you-go). Customers in active modernization (Microsoft Power BI Premium to Microsoft Fabric, Microsoft Azure Synapse to Microsoft Fabric) should hold pay-as-you-go during migration and convert to Reserved Instances post-migration once the steady-state capacity is known.

Migration Patterns

Microsoft Azure Synapse to Microsoft Fabric

Microsoft Azure Synapse remains supported but Microsoft is steering new development to Microsoft Fabric. EPC Group's migration scope ranges $400K-$2M depending on workload volume, regulatory scope, and number of source connectors. Standard timeline 4-12 months.

Snowflake Coexistence to Microsoft Fabric

EPC Group's standard pattern for Snowflake-anchored customers: keep Snowflake as the system of record for Snowflake-native workloads, expose Snowflake to Microsoft Fabric via OneLake Shortcuts, build Microsoft Power BI semantic models in DirectLake mode on the shortcuts, enable Microsoft Power BI Copilot grounding on the resulting semantic models. This pattern preserves Snowflake investment while extending Microsoft Power BI Copilot to lakehouse data.

Databricks Coexistence

Microsoft Fabric reads Databricks-managed Delta Lake natively via OneLake. Databricks remains the system of record for Spark ML and LLM training; Microsoft Fabric becomes the system of record for Microsoft Power BI semantic models and Microsoft 365 Copilot grounding. Microsoft Purview governs the entire chain.

Failure Modes and Cost-of-Failure Scenarios

Microsoft Fabric Capacity Throttling at Quarter-End

A Fortune 500 financial-services customer running Microsoft Fabric F128 hit capacity throttling at quarter-end financial close every quarter for four consecutive quarters before adopting EPC Group's managed services. CFO board reporting was delayed 6-9 hours each cycle. EPC Group right-sized capacity to F256 with one-year reserved-instance pricing — cost-neutral against the prior pay-as-you-go spend — and throttling was eliminated. The lesson: capacity sized for average load throttles at peak; size for the 80th-percentile sustained peak and rebalance quarterly.

OneLake Shortcut Latency on Cross-Region S3 Source

A pharmaceutical customer's clinical analytics pipeline used OneLake Shortcuts to AWS S3 buckets in a different region than the Microsoft Fabric capacity. Cross-region egress costs ran $40K/month and shortcut latency exceeded 8 seconds for analytical queries. EPC Group restructured the architecture to copy the S3 data into a Microsoft Azure Storage account in the same region as the Microsoft Fabric capacity, reduced egress costs to under $5K/month, and brought query latency under 2 seconds.

Microsoft Power BI Copilot Grounding on Stale Semantic Model

A regional bank's Microsoft Power BI Copilot answered executive questions using a semantic model that had not been refreshed in 30+ days due to an upstream pipeline failure no one had triaged. EPC Group's managed-services pattern includes daily refresh-failure triage with root-cause analysis, which would have caught the pipeline failure within 24 hours.

Operating Cadence Under Managed Services

Daily activities cover capacity-utilization check, critical refresh-failure triage, Microsoft Sentinel alert review, and Microsoft Purview AI Hub alert review. Weekly activities cover semantic-model performance review, workspace governance audit, refresh-schedule optimization, and Microsoft Power BI Copilot adoption metrics. Monthly activities cover capacity-tier rebalancing recommendation, Microsoft Compliance Manager score review, semantic-model deprecation candidate analysis, and OneLake medallion zone health. Quarterly activities cover Microsoft Fabric and Microsoft Power BI version-update planning, Microsoft Compliance Manager attestation, Microsoft Sentinel detection-rule tuning, and the formal business review with the customer's analytics leadership.

Frequently Asked Questions

When to use Microsoft Fabric vs Microsoft Azure Synapse?

Microsoft Fabric is the modern unified analytics platform. Microsoft Azure Synapse remains supported but Microsoft is steering new development to Microsoft Fabric. EPC Group recommends Microsoft Fabric for new Azure analytics architectures.

What about coexistence with Snowflake or Databricks?

Microsoft Fabric integrates with Snowflake (OneLake Shortcuts) and Databricks (Delta Lake interop). Most Fortune 500 enterprises end up with two platforms — Microsoft Fabric as the system of record, Snowflake or Databricks for specific workloads.

How much does the platform cost?

Microsoft Fabric F128 reserved: approximately $140K per year. Add Microsoft 365 E5 ($57 per user per month times user count), Microsoft 365 Copilot ($30 per user per month times Copilot user count). Total mid-market enterprise: $5M-$15M per year all-in.

Does Microsoft Fabric work for regulated industries?

Yes. HIPAA-eligible with Business Associate Agreement, FedRAMP-aligned in Microsoft Azure Government and Microsoft 365 GCC, PCI-compliant, supports CMMC Level 2 deployments. EU Data Boundary applies in Microsoft Azure regions within the EU.

What about migration risk from Microsoft Azure Synapse?

EPC Group's standard migration approach is in-place where possible (Microsoft Fabric supports many Microsoft Azure Synapse patterns natively), and parallel-run for the legacy and modern stacks during the cut-over window. The standard timeline includes Microsoft Compliance Manager attestation re-baselining and Microsoft Sentinel custom-rule re-deployment against the new data plane.

How does this architecture handle real-time analytics?

Microsoft Fabric Real-Time Intelligence handles streaming ingestion via Eventstream, KQL Database storage, and dashboards via Microsoft Power BI DirectLake. Microsoft Fabric Activator routes alerts. The standard pattern combines streaming and batch in OneLake so the same data lake serves both real-time dashboards and historical analytics.

Who delivers Microsoft Azure analytics architecture engagements?

EPC Group senior architects with combined Microsoft Power BI (Project Crescent original beta), Microsoft Azure Synapse, Microsoft Fabric, and SAP / Oracle integration experience. Errin O'Connor (CEO) is a 4-time Microsoft Press author.

Next Steps

Schedule a 30-minute Azure analytics architecture discovery call at /schedule or call (888) 381-9725. Senior architects (not sales) take discovery calls.

Related reading: Microsoft Fabric Quickstart Assessment, Microsoft Fabric Consulting Services Guide, End-to-End Microsoft Cloud Solutions Enterprise Guide, Audit-Ready Analytics Compliance Framework Guide, Microsoft Fabric vs Snowflake vs Databricks Enterprise Comparison, and Snowflake to Microsoft Fabric Migration Enterprise Guide.