Microsoft Fabric Barcelona 2026: Copilot IQ, Power BI Apps, SQL and Preview vs GA

The event, and how much Fabric has grown​

The conference is the European Microsoft Fabric + SQL Community Conference, which takes place in Barcelona, Spain from 28 September – 1 October 2026. The organizer changed the name this year, reflecting the addition of dedicated #MicrosoftSQL content alongside the established #MicrosoftFabric programme. Microsoft puts attendance at nearly 5,000, with more than 130 sessions and more than 10 tutorials. That is Microsoft's own estimate. For comparison, Microsoft's June preview blog described a record-breaking gathering of more than 4,000 data professionals in Vienna last year.

The customer numbers are the more interesting comparison. At this year's Atlanta event, Microsoft said Fabric was already serving more than 31,000 customers. In Barcelona it claimed more than 40,000. Microsoft also says SQL Server runs enterprise workloads at more than 90% of the Fortune 500. Both figures come from Microsoft and have not been independently verified.

The theme was the one Microsoft has pushed all year: AI agents only work well when they have governed data, shared business definitions and guardrails. Fabric IQ is Microsoft's name for that context layer. It combines OneLake data, Power BI semantic-model metrics, ontologies and real-time operational signals. Fabric IQ itself is not new. Microsoft had already said that at Microsoft Build 2026, we announced the general availability of Fabric IQ. What Barcelona adds is more places where Fabric IQ shows up.

Section summary: Growth is real, the headline numbers are Microsoft's own, and the pitch is still "context over models."

Fabric IQ in Microsoft Copilot: generally available, with conditions to check​

This is the announcement most likely to reach ordinary Microsoft 365 users:

• Generally available now: Fabric IQ business context in Microsoft Copilot Chat and Cowork.
• Coming soon: integration with Copilot's new Code experience, through the Frontier program.
• Cost: Microsoft says Copilot customers get Fabric AI insights "without additional AI token costs."
• Power BI tie-in: business definitions, measures and knowledge already captured in Power BI semantic models and reports can ground Copilot's answers.

For admins, the useful point is that a well-maintained semantic model now affects what Copilot tells people. The flip side is that if your semantic models are messy, Copilot will repeat that mess with confidence. Microsoft's announcement does not say which Copilot licences qualify, what tenant settings are needed, or how permissions behave when Copilot answers from a semantic model. Test with a pilot group before telling the finance department that Copilot now understands their revenue figures.

Power BI Desktop gets agentic app creation​

Microsoft calls this the next stage of Power BI. You start in Power BI Desktop from a trusted semantic model, describe the app you want in plain language, then preview it, edit it live and publish it to Fabric. Unlike reports, these apps can:

• Accept user input
• Write data back
• Keep shared state
• Support operational workflows

Each app runs on Fabric Apps, which supplies a Fabric SQL database, authentication and security. When an app outgrows its creator, it can be handed to developers or IT without a rebuild.

The licensing is the notable part. Microsoft says Apps in Power BI will enter preview "in the coming weeks" for Power BI Pro and Premium Per User (PPU) customers, as well as Fabric capacity customers. It also says Pro and PPU customers will get Fabric Apps and Fabric Database capabilities at no additional cost, capped at 1 GB per app.

That promise does not yet match Microsoft's own documentation. The current Fabric Apps (preview) page on Microsoft Learn lists these prerequisites:

• The workspace must have Fabric capacity assigned, and Fabric Apps consume capacity units from it.
• A Fabric tenant administrator must turn on the workload. The path is Fabric admin portal → Tenant settings → Fabric Apps (preview) → Enabled, applied to the whole organisation or to specific security groups. Changes can take a few minutes to propagate.
• Fabric Apps are not available in every region yet.

The Pro/PPU route may bring different rules when it ships. Until Microsoft updates the documentation, treat "no additional cost for Pro users" as a promise to check, not something you can already use.

Fabric Apps and Rayfin: from prototype toward production​

Microsoft says Fabric Apps launched in June 2026 and that thousands of developers have used it since. It is built on the open-source Rayfin SDK and CLI. According to Microsoft Learn, developers:

• Define data models as decorated TypeScript classes. Fabric Apps then generates the database schema, GraphQL endpoints, row-level authorisation rules and type-safe client methods.
• Run the full stack locally in Docker, then deploy with npx rayfin up. Projects are scaffolded with npm create @microsoft/rayfin@latest, and schema changes are applied with npx rayfin up db apply.
• Authenticate with Microsoft Entra ID (Fabric SSO) only once deployed. No other identity providers are available after deployment.

The Barcelona announcements add:

• Direct connections to Fabric warehouses, SQL databases, lakehouses and Power BI semantic models without copying data, with existing permissions kept intact
• TypeScript backend functions
• Secret Store
• PostgreSQL support
• Private-by-default access
• Built-in usage and performance metrics

Two points from the documentation are worth knowing before you build. First, the app's SQL database is read-only in the portal. Schema changes have to come from code, and editing the database directly can cause schema conflicts that break the app. Second, Microsoft says Fabric Apps may not suit apps that need complex multi-step transactions, stored procedures, or custom authentication providers. Security stays partly your job: keep secrets out of frontend assets and repositories, and grant contributors only the permissions they need.

OneLake: IQ sharing, Salesforce and BigQuery​

• IQ sharing (preview): shares governed data and business context across teams, customers and partners. At launch it supports governed tables and files, Markdown agent instructions and RDF ontologies. Native support for Fabric IQ ontologies and Power BI semantic models is only on the roadmap. EY, S&P Global, TomTom, Financial Fabric and xMentium are building integrations. TomTom's geospatial and transport data products will be discoverable in the OneLake catalog.
• Salesforce Data Cloud 360: two-way integration, in preview.
• Google BigQuery mirroring: generally available.
• Dynamics 365 Business Central: "deeper integration," with no further details given.

Data engineering agent and Real-Time Intelligence​

The Fabric data engineering agent is in preview and uses technology from Microsoft's acquisition of Osmos. Engineers set the goal and the guardrails. The agent then plans, carries out and checks long-running work such as platform migrations, lakehouse modernisation, performance tuning, complex ETL and data preparation. Tasks inherit workspace permissions, so authorised colleagues can watch progress and add input. The same task can be started or steered from Fabric, GitHub Copilot, VS Code, Codex or Claude Code.

That last point is notable: Microsoft is openly supporting rival coding agents. Microsoft published no benchmarks, though. "Autonomous migration" is a phrase that should come with a rollback plan.

In Real-Time Intelligence, Copilot can now build Real-Time Dashboards from a description. It can also help run eventhouses: creating tables, setting policies, managing schemas and inspecting cluster state. Schema recommendations and optimisation features are aimed at performance and cost.

Ontologies, agents and an open semantic standard​

• A new agentic experience builds ontologies from semantic models, OneLake data and context documents. You can write rules in natural language and bring Power BI measures into the ontology.
• Microsoft restated its support for Apache Ossie (incubating), a vendor-neutral standard for exchanging semantic metadata that it is working on with Snowflake and others.
• Fabric data agents now integrate more closely with Fabric IQ ontologies. The "build agent with AI" mode can draw on semantic models, documentation, glossaries and data dictionaries.
• Operations agents: root-cause analysis is generally available, and pre-approved actions are in preview.

Observability and on-demand billing: news for Fabric admins​

Observability in Fabric (preview) brings together workspace monitoring, the monitor hub and operations agents:

• Workspace monitoring can collect telemetry from multiple workspaces into one eventhouse. It can send data elsewhere through custom endpoints and lets you set retention and caching policies.
• The monitor hub adds built-in views, alerts and AI insights. Activator-powered job alerts report job failures and successes, and a new capacity view shows consumption and remaining headroom.

On cost, on-demand billing extends the pay-per-use model from Autoscale Billing for Spark to Fabric workloads generally, scaling compute up and down with demand. Fabric zero-provisioned lets you start with Fabric and OneLake without committing to capacity upfront. Microsoft gave no price list and no workload-by-workload eligibility. Until it does, you can't compare this against reserved capacity with any confidence.

Databases: Hub, agents, Hyperscale and vectors​

Database Hub was introduced in Atlanta and is now a preview control plane for health, risk, governance and estate-wide visibility. Hyperscale auto-pause suits dev, test and intermittent workloads, which are the databases that tend to run up the bill overnight. Generally available DiskANN indexes mean teams can run semantic search, recommendations and RAG inside the database engine instead of standing up a separate vector store.

For sovereign and edge deployments, SQL Server on Azure Local and its disconnected variant are now generally available. They target sites where cloud connectivity is limited, restricted or deliberately cut off. Disconnected operation is a serious infrastructure project, not a switch you flip. Check the hardware, agreement and support requirements in Microsoft's Azure Local documentation before you promise anything to a regulated customer.

What to do this week​

• Sort items by status. GA items (Fabric IQ in Copilot Chat/Cowork, BigQuery mirroring, DiskANN, 192 vCores, SQL Server on Azure Local) can go into your planning now. Previews belong in test environments.
• Check your semantic models. Copilot and Fabric data agents will now rely on them.
• Review tenant settings before users find Fabric Apps on their own.
• Watch for the Pro/PPU app preview and confirm its capacity and region terms when it arrives.
• Put Hyperscale auto-pause on dev/test databases once you're comfortable with preview terms.

Windows Forum's earlier coverage made the right point before the event: attendees need to separate what is shipping and supportable now from previews and roadmap language. Barcelona delivered some of each, and the GA items are the ones you can plan around today.

Update: Microsoft clarifies SQL previews and adds SSMS announcements (September 29, 2026)​

Microsoft’s September 29 SQLCon post clarifies two statuses in the Barcelona announcements. Higher-density Hyperscale elastic pools are in preview, not merely announced. The company also says both 160-vCore and 192-vCore Premium-series Hyperscale options are generally available. For teams planning a migration, a new Azure Arc workflow is in preview; it brings assessment, provisioning, migration, validation and cutover into one experience.

Microsoft also specifies where its database-agent preview starts: SQL agents in Database Hub and PostgreSQL agents in Visual Studio Code. Support for both agents in both places is planned, not available at the preview’s outset. Admins evaluating the agents should check the particular database and tool combination they need.

For developers, Microsoft says Agent Mode for GitHub Copilot in SQL Server Management Studio is generally available, along with SQL Projects, Schema Compare and SQL Formatter in SSMS. SQL Formatter is also generally available in Visual Studio Code. These are separate from the Database Hub agent preview and can be assessed on their own deployment schedule.