Is Microsoft killing Power BI?
For most of its life Power BI has been a reporting tool, and Microsoft is now moving it down a layer, to the semantic layer where the meaning of the numbers gets defined.
It's a fair question to ask after last week's FabCon in Barcelona. After listening to sessions and reading blogs like Power BI's next chapter, I felt something was missing: the report, which for ten years was more or less what Power BI was. It barely features in how Microsoft talks about the future. The attention goes to apps built from natural language, models drafted by agents, and answers that turn up in Copilot Chat, Outlook or Excel without anyone opening Power BI at all.
So, is Microsoft killing Power BI? No, but it is changing what Power BI is for. For most of its life Power BI has been a reporting tool, and Microsoft is now moving it down a layer, to the semantic layer where the meaning of the numbers gets defined. Whatever sits on top of that can be almost anything: a report, an app, a Copilot answer or an Excel sheet.
I think that's the right move for self-service analytics where a pixel perfect dashboard is not a requirement, and I also think it makes one thing far more important than it was a year ago: semantic governance.
What Microsoft announced
As usual, Microsoft arrived at FabCon Europe with far more announcements than anyone could take in at once. I won't go through all of it, but these are the four announcements that stood out to me:
- The semantic layer as the headline. Microsoft says more than 22 million Power BI semantic models are used every day, which makes Power BI "the world's most widely adopted semantic layer". Semantic views extend that idea into OneLake, so governed metrics can be defined where the data lives and reused beyond Fabric.
- Apps instead of (or next to) reports. Power BI Desktop is getting a natural-language app building experience. You describe the application, it's generated from a semantic model, and it can take input, write data back and run workflows. Pro and Premium Per User customers get it in preview at no additional cost.
- Definitions become a little more portable. Microsoft is backing Apache Ossie, a vendor-neutral standard for exchanging semantic metadata, and working with Snowflake on cross-platform conversion. It also wants DAX to become an Ossie-recognized query language. But portable isn't the same as open. The Ossie Microsoft converter carries Power BI measures across as DAX text and never translates them into SQL. It also can't check that DAX offline: its own documentation says to "use a separate DAX parser" for that, or to deploy the model into a Fabric workspace so Microsoft's engine can compile it. The definition travels, but in practice only Microsoft's engine can run it.
- Copilot everywhere. Fabric IQ is generally available in Microsoft Copilot Chat and Cowork, the Fabric IQ MCP server is GA so developers can use it in "external MCP clients", and the M365 Copilot experience is coming into Power BI itself.
The September feature summary fills in the rest. Agentic authoring for models and reports went GA. A remote authoring MCP server lets people build semantic models without Power BI Desktop. Power BI Projects (PBIP) went GA, so model definitions live as text files in Git. Visuals can be shared live in Outlook, and Export to Excel went from 150,000 to 500,000 rows.
Why the reporting layer can be anything now
Read those announcements together and a pattern shows up. Almost every new feature is about getting the semantic model out of the Power BI report and into some other surface: an app, a Copilot answer in Teams, a visual in an email, a spreadsheet, an agent working through MCP. Microsoft isn't trying to make everyone open Power BI more often. It's trying to make every surface ask Power BI for the numbers.
That means the front end is becoming interchangeable. Your CFO might see revenue in a Power BI report, a Fabric app, a Copilot summary or a chart that Claude drew from the model over MCP, and from Microsoft's point of view all four are fine, as long as the definition underneath is theirs. Microsoft putting the semantic model at the centre is, of course, also a very good commercial position when you already have 22 million of them.
Why this makes semantic governance the main event
For a long time, governing Power BI mostly meant governing reports: you certified the ones people could trust, looked at who was opening them, and retired the ones nobody used. That worked because the report was where people met the numbers. Now the same numbers reach people through an app, a Copilot answer or an Excel export, and the report is only one of many places they turn up. What stays the same across all of them is the definition underneath. So governance has to move from the report to the meaning behind it, which is what I call semantic governance.
As a linguist I'm probably more attached to that word than most, since semantics is simply the study of meaning. In practice it comes down to a few things. For every concept the business runs on, someone owns the definition and one version is certified. You can see where that definition is used and where other versions have crept in. And every answer is traceable back to the definition it was built on.
Keeping metrics in the reporting layer was always bad practice, of course, but it happened, and it still happens all the time. As long as the report was the only place people saw the numbers, the damage stayed more or less contained. With where the future is going that's no longer true.
1. A definition now reaches far more people than it used to. A measure that used to sit on one report page can now reach a CFO through a Copilot answer, and they may never see the report it came from or the caveats that went with it. If that definition was wrong, or simply different from the one Finance uses, the mistake travels along with it. Copilot in Power BI can produce incorrect answers to data questions. To me that's a bigger worry than the old problem of dashboards nobody opens. An unused dashboard is wasteful, but at least you can easily monitor its usage. A wrong answer in Copilot looks just as convincing as a right one, and unless you can trace it back to the definition it used, chances are nobody notices until the number has already been repeated in a meeting.
2. Exchanging definitions is not the same as agreeing on them. I'm very glad to see Ossie. A neutral format for semantic metadata is something the industry has needed for years. But it solves a translation problem, not a meaning problem. Power BI and Snowflake will be able to swap a definition of revenue. They won't tell you that Finance's revenue and Sales' revenue are two different numbers, or which one the board saw. It’s like in linguistics: a shared alphabet doesn’t guarantee that two people mean the same thing by the same word.
3. Every platform now has its own semantic layer. Most enterprises don't run only Power BI. They run it next to Tableau, Qlik, Databricks Genie and a lot of SharePoint files, and each of those ships its own layer of meaning. Microsoft has no particular reason to reconcile its definitions with a competitor's, and neither does anyone else. Add agents that build models in minutes with minimal human input, and the number of definitions grows faster than anyone can review them.
To be fair, Microsoft is clearly thinking about governance. The September release added a control to block read-only users from reaching semantic models through Copilot and MCP, Fabric added Policies and DLP access restrictions, and Fabric IQ answers now carry citations and sensitivity labels. These are real improvements in governing access. What they don't cover yet is governing meaning, especially across tools.
This is the work we do at Datalogz. Datalogz Control Tower sits above Power BI and the other tools, keeps the inventory, shows what depends on each model across old and new surfaces, maps business concepts across the estate and flags where definitions start to drift apart, then routes the fix to the right owner. We deliberately don't build a semantic layer of our own, because the reconciling role only works if it's independent of the tools it reconciles.
So, is Microsoft killing Power BI?
No. It's moving Power BI down a layer, from the place where people look at numbers to the place where numbers get their meaning. That's a more strategic place to be, which Microsoft definitely knows. But it also means that organisations need to start governing the definitions in the semantic model and make sure they are owned, certified and consistent, whichever surface your users end up preferring. The organisations that do well with agentic Power BI will be the ones that treat the agreement on meaning as the product, and the reports, apps and copilots as the packaging.
Written by Anouk Gorris, Head of Product at Datalogz
References
- Mohammad Ali, Power BI's next chapter: The evolution of business intelligence, Power BI Updates Blog, September 2026
- Arun Ulag, FabCon and SQLCon 2026 in Barcelona: Building the data foundation for Microsoft Copilot and agents, Azure Blog, 28 to 29 September 2026
- Power BI September 2026 Feature Summary, 29 September 2026
- Microsoft Fabric September 2026 Feature Summary, 29 September 2026
