Microsoft Fabric in 2026: The Latest Features and Who Is Using It

Microsoft Fabric had one of its busiest months yet. At FabCon Europe in Barcelona (late September 2026), Microsoft shipped a wave of features that push Fabric from “unified analytics platform” towards “data foundation for AI agents”. Microsoft now says Fabric has more than 40,000 customers worldwide.

This guide cuts through the announcements. It covers what is new, what is actually generally available (GA) versus still in preview, who is using Fabric, and what I would do about it as a data engineer this quarter.

At a glance: the updates that matter

Feature	Area	Status
Fabric IQ in Microsoft Copilot Chat	AI & Power BI	GA
Fabric Runtime 2.0 (Spark 4.1, Delta Lake 4.2, Python 3.13)	Data Engineering	GA
Materialized lake views and custom live pools	Data Engineering	GA
Data engineering agent (built on Osmos)	Data Engineering	Preview
Result-set caching	Data Warehouse	GA
GPU-accelerated Data Warehouse	Data Warehouse	Preview
Google BigQuery mirroring	OneLake	GA
IQ sharing (share governed data with partners)	OneLake	Preview
Business events in Real-Time hub	Real-Time Intelligence	GA
Operations agents with root-cause analysis	Real-Time Intelligence	Preview
Git integration and bulk item-definition APIs	CI/CD	GA
Deployment plans (YAML)	CI/CD	Preview
Capacity overage billing, F4096/F8192 SKUs	Capacity	GA
F0 capacity with on-demand billing	Capacity	Preview announced

1. Fabric IQ: your semantic models become an AI layer

The headline announcement is that Fabric IQ is now GA in Microsoft Copilot Chat and Cowork. Business users can ask questions in Copilot and get answers grounded in your existing Power BI reports and semantic models, with permissions and row-level security respected.

Fabric IQ MCP endpoint (GA): exposes trusted Power BI insights to agents running outside Fabric.
Power BI agentic experiences (GA, some parts in preview): agents can build or change semantic models and reports, then validate their own output.
Fabric IQ ontologies (preview): define shared business concepts once and reuse your existing DAX measures, with AI-assisted creation that needs your approval.

Why it matters: the quality of your semantic model now directly decides the quality of your company’s AI answers. Clean measure names, descriptions and RLS are no longer “nice to have”.

2. Data Engineering: Runtime 2.0 and an agent that writes pipelines

Fabric Runtime 2.0 (GA): Apache Spark 4.1, Delta Lake 4.2, Python 3.13 and an improved native execution engine. Existing workspaces are not moved automatically, so you choose when to upgrade.
Materialized lake views and custom live pools (GA): declarative, refreshable transformations in the lakehouse, plus pools you can tune for your workloads. Incremental refresh for updates and deletes is in preview.
Data engineering agent (preview): built on technology from Microsoft’s January 2026 acquisition of Osmos. You describe the outcome and the agent plans, runs and validates Spark notebooks and Delta writes. Microsoft positions it for evaluation, not production, for now.
Developer tooling (preview): pipeline support in VS Code, high-concurrency support across drivers, and a notebook toolkit designed for agents.

3. Data Warehouse: faster repeat queries and GPUs

Result-set caching (GA, on by default): eligible repeated SELECT queries return cached results, which is great for dashboards that hit the same queries all day.
GPU-accelerated warehouse (preview): one-click GPU enablement for high-concurrency analytics. Microsoft quotes a 7× TPC-H result, but treat that as one benchmark, not a promise for your workload.
SQL query editor (GA): faster results grid, better IntelliSense, autosave and .sql import/export. Visual authoring with data profiles and charts is in preview.

4. Real-Time Intelligence: events and agents

Business events in Real-Time hub (GA): reusable event schemas that can trigger Activator actions.
Eventhouse improvements (GA): accelerated OneLake shortcuts, inline transforms and self-optimisation, plus new MongoDB and Salesforce CDC connectors.
Eventhouse MCP server (preview): lets AI agents query Eventhouse data through a hosted MCP endpoint.
Operations agents (preview): investigate issues, suggest root causes and run pre-approved actions. Riskier actions still need human review.

5. OneLake: more sources, open formats and sharing

Google BigQuery mirroring (GA): continuous replication of BigQuery data into OneLake. Note that it is a copy, not a shortcut.
Change data feeds and shortcut lifecycle policies (GA): incremental change consumption and storage lifecycle management.
IQ sharing (preview): share governed tables plus business context, such as instructions and ontologies, with partners and customers outside your tenant.
Open table formats (preview): Iceberg catalog federation and Iceberg writes, a Table Read API, and a hosted OneLake MCP endpoint.
New sources (preview): bidirectional zero-copy sharing with Salesforce Data 360, plus AWS Glue and Google Lakehouse Catalog sources.

6. Governance, CI/CD and cost control

Git integration and bulk item-definition APIs (GA): compare and commit, branching, and bulk import/export of item definitions.
Deployment plans (preview): ordered, repeatable rollouts defined in YAML.
Policies in Fabric (preview): attribute-based controls, for example limiting who can create certain items in production workspaces.
Capacity overage (GA): bills eligible excess usage at 3× the pay-as-you-go rate instead of throttling. It is not a hard spending cap.
F4096 and F8192 SKUs (GA) for very large workloads.
F0 capacity with on-demand billing (preview announced): starts with no pre-provisioned compute and charges for what you consume. It is not free; storage and licences still apply.
Surge protection (preview, GA coming soon): workspace-level consumption limits with exemptions for critical jobs.

Who is using Microsoft Fabric?

Microsoft reports more than 40,000 Fabric customers, alongside more than 425,000 Power BI customers. Fabric Apps, launched in June 2026, already has thousands of developers. Some of the names and results being shared:

Levi Strauss, H&M and Thomson Reuters: named by Microsoft as using Fabric to unify their data estates and ground AI in business context.
TomTom: bringing geospatial and transportation data into Fabric as data products discoverable in the OneLake catalog.
EY and S&P Global: among the partners building IQ sharing integrations.
One NZ (telecoms): moved to Fabric Real-Time Intelligence and reports teams answering inquiries nearly twice as fast.
Multicedi (Italian retail): reported cutting ETL pipeline runtimes from six hours to two.
Microsoft’s own finance team: reported processing times down by two-thirds and data generation costs cut by 50% after moving to Fabric.

Across these stories the patterns are consistent: consolidating siloed systems into OneLake with a medallion architecture, real-time operational dashboards, and governed self-service BI that now feeds Copilot and agents.

What I would do this quarter

Tidy your semantic models. Add descriptions, consistent measure names and RLS. Fabric IQ and Copilot will expose whatever is there.
Test Runtime 2.0 in a dev workspace before switching production notebooks. Spark 4.1 and Python 3.13 can surface library and behaviour changes.
Replace hand-built incremental logic with materialized lake views where it fits.
Check result-set caching on your busiest warehouse dashboards. It is on by default, so measure the gain.
Turn on surge protection and review overage settings so a runaway job cannot surprise you on the bill.
Move deployments into Git and try YAML deployment plans for repeatable releases.
Evaluate the data engineering agent on a non-critical pipeline, not production, and review everything it writes.

Release stages change quickly. Statuses here reflect announcements in late September and October 2026. Check the official Microsoft Fabric documentation and roadmap before planning production work around a preview feature.