Copilot across Microsoft Fabric and Power BI: Overview

What it does

Copilot in Fabric is a contextual AI layer sitting across every Fabric workload, including Power BI, that turns natural-language prompts into code, queries, visuals and summaries. Power BI is one workload inside Fabric, so its Copilot is the same platform feature you meet in Data Factory, notebooks and the warehouse, just pointed at reports and semantic models.

Key facts

There isn't one Copilot. Each workload (Power BI, Data Factory, Data Engineering, Data Science, Data Warehouse, Real-Time Intelligence) has its own Copilot experience tuned to that job.
Power BI sits inside Fabric. A Power BI licence and a Fabric capacity are the same billing surface for Copilot purposes, so anything you read about Fabric Copilot capacity applies to Power BI Copilot too.
All Copilot experiences run on paid capacity. Since 28 April 2025 the floor is F2 or P1; the old F64 minimum is gone.
Copilot is grounded on your data and metadata, not on public web content. Quality of output tracks quality of your semantic model, schemas and descriptions.
Consumption is metered in Capacity Units against your capacity, billed by tokens processed, not per user.
Fabric IQ (announced at Ignite 2025) is the newer semantic/ontology layer that lets Copilot, including M365 Copilot, chat with governed Fabric and Power BI data. Treat it as a separate, mostly-preview topic.
Availability varies by feature: report creation in Power BI is broadly available, several Fabric workload Copilots and the Fabric IQ pieces are still preview.

When to use / skip

Use this doc to orient a client who's confused about why "Copilot" means five different things in their tenant. The single most useful framing: Power BI is a Fabric workload, capacity and licensing are shared, but the actual Copilot features are workload-specific and mature at very different rates. Skip trying to give one blanket answer on capability or GA status. You'll be wrong for at least three workloads.

Configuration decisions

Decide early whether Power BI Copilot and Fabric workload Copilot draw on the same capacity or a dedicated Fabric Copilot capacity for billing isolation.
Work out data residency: cross-geo processing is a tenant setting and may push prompts outside your compliance boundary.
Agree who owns semantic-model quality. Copilot output in Power BI is only as good as the model descriptions and synonyms behind it.
Separate "available today" from "roadmap" per workload before you promise anything.

Gotchas

People assume buying Copilot Studio or M365 Copilot lights up Fabric Copilot. It doesn't. Fabric Copilot needs Fabric/Power BI paid capacity.
The word "Copilot" in a Microsoft demo could be any of six different things. Always pin down which workload.
Preview features can change behaviour or pricing between visits. Check status per feature, not per platform.
Fabric IQ, Work IQ and Foundry IQ are three different "IQ" layers announced together. Don't mix them up in front of a client.

Consultant notes

Lead scoping conversations with the capacity/billing model, not the shiny features. That's where budget surprises live.
When a client says "we want Copilot for our reports", they usually mean Power BI report creation plus asking questions of data. Confirm before you cost anything wider.
Keep a one-line status tracker per workload Copilot. It changes often enough that memory will let you down.

Re-check the F-SKU floor and per-workload GA status before quoting anything. Both move.