FabCon Europe 2026 highlights: Microsoft Fabric gets easier to adopt, deploy and operate

FabCon Europe 2026 brought another long list of announcements for Microsoft Fabric. But rather than repeat every item from the September feature summary, I want to focus on the changes that could have the biggest practical impact when building and running Fabric solutions.
The common thread is clear: Microsoft is removing more of the friction between a successful proof of concept and a production-ready data platform. F0 capacities lower the barrier to getting started. Deployment Plans give teams more control over deployments. Apache Ossie points towards a more open semantic layer. New T-SQL syntax reduces repetitive code and Data Factory gains better orchestration and monitoring. MCP servers also make Fabric capabilities and governed business context easier to expose to AI agents through a standardized protocol.
Microsoft’s September 2026 Fabric update contained more than 150 announcements and enhancements across the platform. In this blogpost, I will walk through the announcements that stood out to me and explain why they matter for Fabric architects, engineers and teams preparing to move workloads into production.

F0 capacities: a lower entry barrier for Fabric

The upcoming zero-provisioned F0 capacity introduces on-demand billing without an upfront compute commitment. That makes it interesting for evaluations, development and workloads with irregular demand. Supported workloads are billed on demand, so teams can start small while they learn how much capacity they actually need.  F2 and higher capacities can also use on-demand billing for selected workload categories. Stable workloads stay on provisioned capacity, while occasional peaks are billed separately.

Why this matters

  • Start evaluating Fabric without immediately committing to fixed compute.
  • Match development and experimental costs more closely to actual use.
  • Separate variable workloads from predictable shared-capacity consumption.
  • Apply CU-hour spending limits over a rolling 24-hour window.

The real change is the consumption model, not simply a smaller SKU. F0 and the broader on-demand options are announced as coming soon.

Deployment Plans: CI/CD that understands the full solution

Deploying a complete Fabric solution involves more than moving items between workspaces. Some items must be deployed or executed in a specific order.

Deployment Plans, now in preview, add that orchestration layer. Teams can arrange items in ordered deployment sets and run actions before or after deployment. Deployments plans work with the following deployment tools: git integration, Fabric deployment pipelines and supported REST APIs.

Fabric deployment plan canvas
Source:
link

A plan could deploy a lakehouse, run a notebook and then deploy a dependent warehouse. If a step fails, dependent steps stop. This captures dependencies and release logic in Fabric instead of a separate manual runbook. This feature is launched as a public preview.

Apache Ossie: towards an open semantic layer

Microsoft also committed to Apache Ossie, an open-source Apache project for exchanging semantic metadata across analytics, AI and BI platforms.

Semantic models hold measures, dimensions, relationships and business definitions. Recreating that context in every platform adds work and can lead to inconsistent definitions.

Apache Ossie universal semantic layer
Source:
link

Ossie aims to make that context portable. One highlighted scenario converts the same Ossie document into both a Snowflake Semantic View and a Power BI semantic model.

Power BI semantic model definition in Ossie
Source:
link

This matters for AI agents too. Access to tables is not enough, agents need governed business meaning. Ossie is still in an early stage, but it is an important step towards a more portable semantic layer.

GROUP BY ALL and ORDER BY ALL

Fabric Data Warehouse is also getting useful T-SQL improvements in preview: GROUP BY ALL, ORDER BY ALL, QUALIFY and FROM … SELECT.

GROUP BY ALL and ORDER BY ALL in Fabric Warehouse
Source:
link

GROUP BY ALL groups by every selected non-aggregated column. ORDER BY ALL applies the same idea to sorting.

These are small additions but they reduce boilerplate and make analytical SQL easier to read and maintain. QUALIFY provides a more direct way to filter window-function results, while FROM … SELECT offers a data-flow-oriented alternative for structuring a query.

Pipeline run conditions: dependencies based on state

Pipeline dependencies should follow the state of the data, not only the clock. Pipeline-level run conditions, now in preview, let downstream pipelines wait for an upstream result. Available conditions include Custom, Data availability and Window coverage.

Microsoft Fabric pipeline run conditions
Source:
link

This should reduce brittle schedules, unnecessary reruns and custom coordination logic. You can test this feature already, however, still in preview.

MCP servers: bringing Fabric into the agent ecosystem

AI agents were a recurring FabCon theme. Through Model Context Protocol, Fabric can expose capabilities to compatible clients without a separate custom integration for each one.

Fabric Core MCP Server is now generally available as a cloud-hosted endpoint. Clients authenticate with Microsoft Entra ID and can search OneLake, manage Fabric items and view capacity information. Existing access controls and audit logs still apply.

Fabric IQ MCP: governed business context for agents

Fabric IQ MCP Server, also generally available, has a different role. Core MCP focuses on the platform; Fabric IQ MCP exposes governed business data.

AI applications can discover Power BI reports and semantic models, inspect metadata and execute DAX queries against authorized models. Row-level and object-level security remain enforced.

The distinction is useful: one server exposes platform operations, while the other exposes governed business context.

Data agents are MCP servers too

Published Fabric data agents can also act as MCP servers. MCP Tasks, now generally available, support longer-running work without keeping the connection open. Clients can check status, resume work and provide extra input when requested.

Fabric Data Agent MCP server configuration
Source:
link

For me, MCP is one of the announcements to watch. Fabric is exposing platform operations, semantic data and data-agent capabilities through one open protocol, reducing one-off integration work.

Other announcements worth watching

A few other announcements worth watching:

• Fabric Policies in preview for more granular administrative control.

• DLP Restrict Access becoming generally available.

• Copy job support for change data capture and SCD Type 2 becoming generally available.

• Mirroring for Google BigQuery and SharePoint Lists becoming generally available.

• Fabric IQ, data agents and ontologies continue to expand as Microsoft builds out its agentic and conversational analytics experiences.

Closing: Fabric is becoming a true enterprise-grade platform

FabCon Europe 2026 was not about one headline feature. The bigger story is how the updates work together.

F0 lowers the entry barrier. Deployment Plans improve releases. Ossie supports semantic portability. T-SQL and pipeline updates reduce daily friction. MCP connects Fabric with agents and AI applications.

Together, these updates address the questions that appear after a proof of concept: cost, deployment, interoperability and reliable operations.

Microsoft Fabric is still evolving quickly but its focus is becoming increasingly practical. The conversation is no longer only about what teams can build in Fabric. It is also about how they can deploy, govern and operate those solutions with confidence.


Niels De Swaef

Niels De Swaef is a Data Engineer passionate about using technology to solve complex problems and improve people’s lives. He has experience with the Microsoft Azure data stack, including Synapse, Databricks, Azure Data Factory and CI/CD pipelines. Niels is also a lifelong learner who is always looking for new ways to use data to make a positive impact.