Microsoft Fabric

What it is

Microsoft’s SaaS analytics platform: data integration, engineering, warehousing, real-time analytics, data science, databases, Power BI and (since FabCon 2026) the IQ workload, all on one storage layer, OneLake (built on Azure Data Lake Storage Gen2, one logical lake per tenant, Delta/Parquet tables, shortcuts for zero-copy access to ADLS, S3 and Google Cloud Storage). For BI users it is the platform Power BI now lives in: capacity, governance, dataflows and Direct Lake all come from Fabric. Terminology: microsoft-fabric-terminology-guide (RADACAD summary, July 2024, may be dated).

Maker and history

Microsoft (Microsoft). Power BI Premium P SKUs are being retired in favour of Fabric F SKUs (Microsoft Learn, 2026-10-05). Launch date not checked (open item). Events: FabCon + SQLCon 2026, Atlanta, 2026-03-16 to 2026-03-20 (secondary report, ~8,000 attendees).

Products and editions (names only)

• Workloads (Microsoft Learn, 2026-03 / 2026-09): Power BI, Data Factory, Data Engineering (Spark), Data Science, Data Warehouse, Databases (SQL database in Fabric, mirroring), Real-Time Intelligence (Eventhouse, Real-Time hub), Industry Solutions, IQ (preview).
• Capacity: Fabric F SKUs F2 to F8192 (pay-as-you-go via Azure, optional yearly reservation), a 60-day Trial capacity (F64 equivalent), legacy Power BI Premium P/EM/A SKUs. Per-user: Free, Pro, Premium Per User. Free users can view Power BI content only on F64+ (Viewer role). Fabric Copilot capacity bills only Copilot usage. Licensing: https://learn.microsoft.com/en-us/fabric/enterprise/licenses; pricing: https://azure.microsoft.com/pricing/details/microsoft-fabric/ (not checked, link only).

Data connectivity and modelling (BI view)

• Lakehouse / Warehouse / Eventhouse / mirrored databases feed Power BI semantic models. Models use DAX.
• Direct Lake (Microsoft Learn, 2026-09-02): table storage mode that loads Delta tables from OneLake into VertiPaq memory on demand. Refresh only “frames” metadata (seconds) instead of copying data. Two variants: Direct Lake on OneLake (multiple Fabric sources, no DirectQuery fallback, can be mixed with Import tables) and Direct Lake on SQL endpoint (single source, falls back to DirectQuery for views or SQL-level RLS). Needs a Fabric F SKU or P capacity; guardrails scale with SKU. Not supported through gateways. Calculated columns on Direct Lake on OneLake are in preview (August 2026). Import still preferred for self-service analysts who cannot write to the lakehouse.
• OneLake integration can write Import-mode tables to OneLake as Delta without migration.
• Mirroring (Oracle and SAP reported GA at FabCon 2026, secondary) and shortcuts reduce ETL.

Deployment

SaaS on Azure; tenant home region must support all workloads for IQ MCP and standalone Copilot. Sovereign-cloud availability of these AI features is not verified. On-premises BI stays on Power BI Report Server (Fabric F64+ reserved capacity includes the right).

AI features as of 2026-10-05

• Copilot in Fabric: assistance in each workload; Power BI parts need F2+ or P1+ and are described in Power BI. Experiences for Data Factory, Data Engineering, Data Science, Data Warehouse and Real-Time Intelligence are preview (Microsoft Learn, 2026-08-24). Not supported on trial capacities.
• Fabric data agent: Microsoft Learn describes it as generally available (page dated 2026-05-11; FabCon coverage says GA moved to F2+ from F64+). Configurable conversational Q&A agent over up to five sources (lakehouse, warehouse, Power BI semantic model, KQL database, mirrored database, ontology, Microsoft Graph); generates NL2SQL / NL2DAX / NL2KQL, read-only, uses the caller’s permissions (RLS/CLS apply), English only, replies capped at 25 rows and 25 columns, cannot read unstructured files. Publishable to Microsoft 365 Copilot, Copilot Studio, Foundry and custom apps. Needs F2+ or P1+ with Fabric enabled and cross-geo AI settings. Git integration and deployment pipelines supported. Purview governance features for agents are in preview.
• Fabric IQ (workload still labelled preview, Microsoft Learn 2026-09-29): “business context” layer in Microsoft IQ (alongside Work IQ, Foundry IQ, Web IQ). Items: ontology (preview) (entity types, relationships, rules, metrics bound to OneLake data; can be generated from semantic models; imports RDF/OWL), Power BI semantic model, Planning, Graph, data agent, operations agent. At Build 2026 Graph in Fabric, Planning and operations agents were reported GA (secondary: Microsoft Build recaps); Learn pages still mark ontology preview.
• MCP / agents: Fabric IQ MCP (GA, read-only Power BI discovery and DAX, see Power BI); broader Fabric MCP servers overview and Skills for Fabric (open-source, MIT, microsoft/skills-for-fabric on GitHub; works with GitHub Copilot CLI, VS Code, Claude Code, Cursor, others; skills do not currently render finished Power BI reports). Fabric local MCP reported GA at FabCon (secondary). Fabric IQ with Agent 365 and GitHub Copilot: preview (Build 2026, secondary). Fabric Apps for semantic models: preview.
• Governance: Purview DLP and access restriction policies limit what agents return; outputs are logged for audit.

Strengths and weaknesses vs peers

• Strengths (largely Microsoft’s stated design, i.e. vendor view): one governed lake for BI and AI; Direct Lake avoids copying data into the model (Microsoft Learn); Power BI semantic models serve as the shared business-logic layer for agents. Competing lakehouse-plus-BI stacks: Databricks AI/BI, Snowflake, Google Looker plus BigQuery (see Looker MCP server).
• Weaknesses (opinion, not verified by independent testing): capacity-unit billing and throttling need monitoring; feature churn between preview and GA across workloads; many AI features gated by region and capacity size.

Self-learning

See Power BI learning path. Fabric-specific, all opened 2026-10-05:

• Microsoft Learn module “Introduction to end-to-end analytics using Microsoft Fabric” (free): https://learn.microsoft.com/en-us/training/modules/introduction-end-analytics-use-microsoft-fabric/.
• Fabric Analyst in a Day (FAIAD): free hands-on workshop for Power BI/Fabric analysts, linked from the Fabric overview (https://aka.ms/LearnFAIAD redirects to the Microsoft events catalogue; session availability not checked).
• Certification Microsoft Certified: Fabric Analytics Engineer Associate (exam DP-600, paid, renew yearly free; English version updates 2026-10-19): https://learn.microsoft.com/en-us/credentials/certifications/fabric-analytics-engineer-associate/.
• Skills for Fabric (GitHub): https://github.com/microsoft/skills-for-fabric.

Sources (fetched 2026-10-05)

• https://learn.microsoft.com/en-us/fabric/fundamentals/microsoft-fabric-overview
• https://learn.microsoft.com/en-us/fabric/iq/overview
• https://learn.microsoft.com/en-us/fabric/iq/connectors/fabric-iq-mcp
• https://learn.microsoft.com/en-us/fabric/data-science/concept-data-agent
• https://learn.microsoft.com/en-us/fabric/fundamentals/direct-lake-overview
• https://learn.microsoft.com/en-us/fabric/enterprise/licenses
• https://learn.microsoft.com/en-us/fabric/fundamentals/skills-for-fabric-overview
• https://learn.microsoft.com/en-us/power-bi/create-reports/copilot-introduction
• https://www.psistla.com/articles/fabcon-atlanta-2026-announcements (secondary, FabCon)
• https://www.jamesserra.com/archive/2026/06/microsoft-build-2026-announcements/ (secondary, Build)
• https://learn.microsoft.com/en-us/credentials/certifications/fabric-analytics-engineer-associate/

Open items

• The Microsoft announcement blogs for FabCon 2026 and Build 2026 were not fetched directly (403 or not tried); GA/preview claims marked “secondary” should be re-checked against Microsoft Fabric blog posts.
• Fabric GA date, Fabric pricing page contents, FabCon Europe 2026 dates not verified.
• Fabric Remote MCP: a Microsoft Q&A thread (2026) says a general “Fabric core” remote endpoint was announced at FabCon but not publicly documented; only workload-specific remote MCPs (Power BI, Fabric IQ, Eventhouse) were seen. Unconfirmed.