
In Short: My FabCon Europe 2026 Takeaways
I spent FabCon week watching the FabCon Europe 2026 keynotes and working through Microsoft's announcements. Our full FabCon Europe 2026 recap covers what was announced and the status of each item. This post is different: it's my personal view of what matters, and what we at Solv. Systems are changing in how we work with clients because of it.
The short version. Your Power BI semantic models are now something Copilot reads, capacity is no longer a single up-front decision, and Power BI delivery is code by default. So we're changing how we scope, how we size, how we build and how we govern. Here are my five takeaways, with the concrete change attached to each.
1. Your Semantic Model Is Now Something Copilot Talks To
The announcement I keep coming back to is Fabric IQ in Microsoft Copilot Chat and Cowork, which Microsoft says is generally available. Users can ask a business question in Copilot and get an answer grounded in governed Power BI reports and semantic models, with Copilot working out which model fits the question.
That changes who your semantic model is for. It used to serve report builders. Now it also serves anyone with Copilot asking questions in plain English, and the quality of those answers depends on names, descriptions, certified measures and security that most models were never built with.
What we're changing (AI readiness): every Power BI engagement we scope now includes an AI-readiness pass on the semantic models in play. Clear table and measure names, descriptions written for someone who isn't the developer, and a decision on which measures are the official ones. It's no longer a "phase two" item. Our checklist for preparing a semantic model for Copilot and AI agents is where we start.
2. The AI Governance Defaults Need a Decision, Not a Shrug
The second takeaway is about defaults, and they point in different directions. The new per-semantic-model setting that controls whether read-only users can reach a model through Copilot, data agents and the remote Fabric and Power BI MCP servers is on by default. Microsoft 365 Copilot integrated into Power BI, which is in preview, is off by default behind a new admin portal setting. And Restrict from Copilot gives administrators control over which semantic models take part in AI experiences.
My view is that "on by default" is the one that will catch people out. AI access to a model carries on until someone with write permission turns it off, and in most estates we look at, nobody has been asked to make that call.
What we're changing (governance): at go-live we now hand over an AI access register alongside the usual workspace and security documentation - every semantic model, its owner, and a deliberate yes or no on AI access. Models carrying sensitive HR or finance data get looked at first. Both layers of control are explained in our guide to Microsoft 365 Copilot in Power BI and controlling AI access to your models.
3. Capacity Stops Being One Big Up-Front Bet
F0 and on-demand billing got the most attention, and our F0 and on-demand billing explainer has been one of the most-read posts on our blog since it went up. Both are listed as coming soon, in preview "in the coming weeks". F0 is a zero-provisioned SKU with no upfront compute charge, and on F2 or higher you'll be able to move individual billing categories onto on-demand, with spend limits over a rolling 24-hour window. The pricing hasn't been published, so nobody can model the cost yet - us included.
Capacity overage, meanwhile, is generally available today: extra compute billed at three times the pay-as-you-go rate, behind what Microsoft calls "a spending threshold, not a hard spending cap".
What we're changing (scoping and sizing): we've stopped opening the capacity conversation with "F2 or F64?". We start from the shape of the workload - what's steady, what's spiky, what's still an experiment - and keep reservations for load that has proven itself. I've written up why we're changing how we size Microsoft Fabric now that F0 exists, and the step-by-step version is our pilot first, reserve later proof-of-concept playbook. One more habit: on every capacity we touch, we check the overage setting, because Microsoft's own documentation is inconsistent on whether it's on by default. The detail is in Fabric capacity overage explained.
4. Power BI Pro Teams Are About to Get Apps
This one matters most for the small and mid-sized teams we work with in South Africa and the UK. Microsoft says apps in Power BI, built on Fabric Apps, will be available in preview in the coming weeks to Power BI Pro and Premium Per User customers at no additional cost, including Fabric Database capabilities up to 1 GB per app. You start from a semantic model, describe the app, and publish something that can accept inputs and write back data.
For years, a request like "can we capture the forecast adjustment right next to the report?" meant a separate Power Apps build or a capacity conversation. That gap is closing for Pro customers. But it's announced, not available, and I won't plan a client rollout around a preview date.
What we're changing (scoping): write-back requests now get a different first question. If the first screen of the app is a report, we look at Fabric Apps. If it's a form or a multi-step process, Power Apps is still the mature, generally available answer. For smaller teams I've gone into this further in why pay-per-use changes the "Fabric is too expensive" conversation for small teams.
5. Power BI Delivery Is Code by Default Now
The quieter announcement with the biggest effect on our day-to-day: Power BI Projects (PBIP) are generally available, and PBIR is now the default report format. Add the Git and CI/CD updates in Fabric - compare and commit, selective branching and bulk import/export APIs generally available, deployment plans and file-level commits in preview - and Power BI work can be reviewed and deployed like any other code.
What we're changing (delivery): every new Power BI build starts as a PBIP in a Git repository on day one, with pull requests for changes. Before we touch an existing client report, we check the known limitations - sensitivity labels aren't supported with Power BI projects, for one - because editing and saving a legacy report converts it to PBIR, and in practice that's one-way. The details are in what PBIP and PBIR change for your Power BI files, and how it reshapes a whole project is in how FabCon 2026 changes the way we'll build Power BI projects.
What I'm Not Changing Yet
Not everything announced at FabCon changes how we work today. The data engineering agent is in preview. Workspace-level surge protection is "generally available soon". F0 and on-demand billing are coming soon with no published price. Microsoft's own feature summary says "Availability and timelines are subject to change", and I take that literally.
So a few rules stay firm. We don't advise anyone to cancel a reservation on the strength of an announcement. We don't put preview features on a client's production roadmap until they appear in the tenant and the documentation lands. And where Microsoft's posts use different status words for the same thing, we plan around the cautious one.
What we are doing is getting ready: gathering capacity metrics so the on-demand comparison is quick once prices arrive, and tidying semantic models now, because Copilot is already reading them.
Where Solv. Systems Comes In
If you want to know which of the FabCon announcements change your plans and which can wait, that's exactly the conversation we're having with clients right now. Our Microsoft Fabric consulting team can review your semantic models, AI access settings and capacity against the FabCon Europe 2026 announcements, and give you a short list of what to act on first.
Sources and Further Reading
- FabCon and SQLCon 2026 in Barcelona: Building the data foundation for Microsoft Copilot and agents - Arun Ulag
- Fabric September 2026 Feature Summary
- Power BI September 2026 Feature Summary
- FabCon Europe 2026: The Announcements That Matter and What They Mean for You - Solv. Systems
- Microsoft Fabric F0 and On-Demand Billing: What Changes and When - Solv. Systems
Frequently asked
In Nick de Vrye's view: semantic models are now an AI surface because Fabric IQ in Microsoft Copilot Chat and Cowork is generally available; AI access defaults need a deliberate decision per model; capacity is no longer one up-front bet now that F0 and on-demand billing are coming soon; Power BI Pro customers are due to get apps in preview; and Power BI delivery is code by default with PBIP generally available and PBIR the default format.
No. Both were announced at FabCon Europe 2026 as coming soon, in preview in the coming weeks, and Microsoft says availability and timelines are subject to change. Pricing for F0 and the on-demand multiplier has not been published, so the cost cannot be modelled yet. Capacity overage, by contrast, is generally available now.
The per-semantic-model setting that lets read-only users reach a model through Copilot, data agents and the remote Fabric and Power BI MCP servers is on by default, so AI access continues until a user with write permission turns it off. Microsoft 365 Copilot integrated into Power BI, which is in preview, sits behind a separate admin portal setting that is off by default.
We would not cancel a reservation on the strength of an announcement. F0 and on-demand billing are coming soon with no published price, and a reserved F SKU is still likely to be cheapest for steady, all-day load. Gather capacity metrics now so you can compare quickly once Microsoft publishes the rates.
Not yet. Microsoft says apps in Power BI, built on Fabric Apps, will be available in preview in the coming weeks to Power BI Pro and Premium Per User customers at no additional cost, including Fabric Database capabilities up to 1 GB per app. Until the preview appears, Power Apps remains the mature, generally available option for form-led and multi-step processes.
Yes, for new builds. Every new Power BI project starts as a Power BI Project (PBIP) in a Git repository from day one. For existing reports we check the known limitations first, such as sensitivity labels not being supported with Power BI projects, because editing and saving a legacy report converts it to PBIR.


