TL;DR
Q: Why don’t my Power BI Pro users see Copilot?
A: Because Copilot doesn’t come with a Pro or Premium Per User license. It needs a paid Fabric capacity.
Q: Do I need an F64 for that?
A: No. The minimum is an F2, the smallest Fabric capacity. Designate it as a Fabric Copilot capacity, assign a security group, and everybody in that group can use Copilot, in Pro workspaces, PPU workspaces and Power BI Desktop. No report has to move.
Q: How far does an F2 go?
A: It depends on your scenario. As a rough rule: a few hundred simple Copilot questions or about 20 Copilot-built report pages per day, for about $191 per month (1-year reservation, West Europe), shared by the whole group.
Small investment. Massive boost.
Why your Pro users don’t have Copilot today
Copilot in Power BI runs on Fabric capacity units (CUs). Microsoft Learn is explicit about it: “A Power BI Pro or Premium Per User (PPU) license alone isn’t sufficient, Copilot requires organizational capacity.” Trial capacities don’t count either. You need a paid F2 or higher (or a Power BI Premium P1).
Many teams still think “capacity” means F64. That was true in the early days. It isn’t anymore.
What a Fabric Copilot capacity is (and what it isn’t)
A Fabric Copilot capacity is a normal Fabric capacity that you designate to collect the Copilot usage of a specific group of users. Normally, Copilot is billed to whichever capacity holds the content. With a Copilot capacity, the assigned users’ Copilot usage goes to their capacity instead, including usage from:
- Power BI Desktop
- Power BI in Pro and PPU workspaces (and Premium or Fabric capacity workspaces)
- Fabric Copilot experiences and data agents in workspaces on capacities smaller than F64
What it is not:
- Not a new SKU and not a new license. It’s a setting on a capacity you already pay for.
- Not a content migration. Reports and semantic models stay where they are.
- Not unlimited. It has the CUs of the SKU you bought, nothing more.
Step 1: Check two tenant settings
Who: Fabric administrator.
Where: Fabric > gear icon > Admin portal > Tenant settings > group Copilot and AI.
| Setting | Default | What to do |
|---|---|---|
| Users can use Copilot, AI Agents and other AI experiences powered by Azure OpenAI | Enabled | Keep it on, ideally for a security group first. It must be on at tenant level for the standalone Copilot and Copilot in apps. |
| Capacities can be designated as Fabric Copilot capacities | Enabled | Keep it on, restricted to the capacity admins who may do it. |
If your home region is outside the US and the EU Data Boundary (for example UK, Australia or Japan), you also need “Data sent to Azure OpenAI can be processed outside your capacity’s geographic region, compliance boundary, or national cloud instance”. In Germany or anywhere in the EU you don’t: Copilot is processed inside the EU Data Boundary.
Optional: “Users can access a standalone, cross-item Power BI Copilot experience (preview)” turns on the full-screen Copilot in the left navigation.
(Setting names as shown in the admin portal in November 2026. Microsoft Learn still uses slightly older wording, for example “Copilot in Fabric capacities”.)
Step 2: Find your home region
A Copilot capacity only works in your tenant’s home region. So find it before you buy anything.
- Sign in to Fabric.
- Open Help & Support (?) in the top bar and select About Power BI (or About Microsoft Fabric, depending on the experience you’re in).
- Read the line Your home region is ….
Step 3: Create an F2 in the home region
Who: someone with rights in an Azure subscription.
Where: Azure portal > Create Fabric capacity.
Pick size F2 and the region from Step 2. Add the capacity administrators. The portal shows the estimated monthly cost right in the size picker (F2 in West US 3: $262.80 per month pay-as-you-go).
Already have a capacity in the home region? You can use it, but think twice: Copilot then competes with your reports and refreshes for the same CUs. A small, dedicated F2 keeps Copilot away from your production workloads.
Note: after a purchase or a resize, it can take up to 24 hours until Copilot recognizes the capacity.
Step 4: Make it a Copilot capacity and assign your users
Who: capacity administrator.
Where: Fabric > gear icon > Admin portal > Capacity settings > Fabric Capacity > click the name of your F2.
Two things that cost me a few clicks:
- Click the capacity name, not the gear icon. The gear opens a short side panel without the Copilot setting.
- Make sure the capacity is running (resumed in Azure). A paused capacity can’t serve Copilot anyway.
- In the Details section, expand Copilot capacity: “Turn on this setting to designate this capacity as a Fabric Copilot capacity.”
- Under Apply to, choose Specific users or groups and add a security group. The entire organization is greyed out by design; if you really want everyone, use an all-users group.
- Select Apply. The status changes to Enabled for a subset of the organization.
That’s it for the users: no extra steps and no additional per-user Copilot license. Their existing Power BI licenses and permissions still apply, and their Copilot usage is now billed to the F2.
Two rules to know:
- One Copilot capacity per user. If a user is in several, the most recently created one wins.
- The user’s Copilot capacity takes precedence over the capacity of the workspace they work in.
How far does one F2 go? It depends on your scenario
Copilot is billed by tokens: what you type, plus what Copilot sends along (parts of your semantic model, the chat history) and what it writes back. So the honest answer is: it depends on your models and how your people use Copilot. As a rough rule of thumb for one F2 per day:
| What your users do | Roughly per day on one F2 |
|---|---|
| Simple questions (short prompt, short answer) | ~400 |
| Copilot-built report pages | ~20 |
Or any mix of both. The assumptions behind it:
- The F2 does nothing but Copilot, around the clock. That’s exactly the setup in this article: a dedicated Copilot capacity.
- “Simple question” = Microsoft’s own example request in the Copilot billing docs. My demo tenant confirmed the order of magnitude: Copilot requests from Power BI Desktop came in close to that example.
- “Report page” = a single user-reported example from Microsoft Q&A: one simple Copilot-built page on an F2. Not a benchmark, but a useful order of magnitude. A page is much heavier than a question, because Copilot reads your model schema and writes a lot of report metadata.
- Bursts are smoothed. Copilot usage is spread over the following 24 hours, which takes the edge off a busy morning. Accumulated usage can still lead to throttling or rejected requests when you consistently ask for more than the F2 has.
Your numbers will be lower with large semantic models, long chats, many report pages, or when Copilot shares the capacity with refreshes and reports. They scale with the SKU: an F4 carries twice as much, an F8 four times.
So don’t trust my table, measure your own: start with a pilot group and look at the Metrics app after one or two weeks (see below).
The price
Fabric capacity list prices (USD, Azure Retail Prices API, checked November 4, 2026):
| Region | Pay-as-you-go F2 per month (730 h) | 1-year reservation F2 per month |
|---|---|---|
| West Europe / Germany West Central | ~$321 | ~$191 |
| Sweden Central | ~$277 | ~$165 |
| East US | ~$263 | ~$156 |
For example, spread over a pilot group of 50 users, a reserved F2 in West Europe is about $3.80 per user and month. Small investment. Massive boost.
A word on pausing: a running capacity costs the same whether Copilot is busy or idle. If you pause it, the smoothed usage still outstanding is added to your bill right away. That’s why a 10-minute F2 test with Copilot shows a separate “Copilot and AI” charge. Pausing a Copilot capacity doesn’t make the AI usage free.
Monitor: Capacity Metrics app
Open the Microsoft Fabric Capacity Metrics app, go to Item History, pick your capacity and set the Operation slicer to Copilot In Fabric. You see 30 days of Copilot CU usage by workspace, item and date. Start with a pilot group, watch utilization for one or two weeks, then decide: keep the F2, or move to F4/F8 when the data says so.
The catch
- Home region only. The Copilot capacity must sit in your tenant’s home region.
- Not for everything. Fabric AI functions aren’t supported, and workspaces in Embedded license mode are excluded.
- Up to 24 hours until Copilot recognizes a new or resized capacity.
- One Copilot capacity per user.
- No “entire organization” switch. Assign users or groups.
- Heavy usage throttles. When accumulated usage exceeds what the F2 can carry, Copilot requests are delayed or rejected. That’s what the Metrics app is for.
Sneak peek: why Copilot matters even more with Fabric Apps (aka modern report, aka report app)
This section is my personal opinion, based on public information.
In October I wrote that the report canvas will be replaced as the default front end by Fabric Apps, aka modern report, aka report app (Unpopular opinion: Power BI reports will be replaced by Fabric Apps). Here’s why that makes Copilot more important, not less.
What’s public:
- Fabric Apps is in preview. Microsoft’s Power BI documentation says Fabric Apps “enables coding agents to create custom web apps for exploring and presenting data”, and: “Reports remain a core analytics experience. Fabric Apps complement reports.”
- The June 2026 Power BI update introduced “Fabric Apps for Semantic Models (Preview)”, described as “an AI-first app development path on top of semantic models”.
Why Copilot matters even more:
- Building becomes AI-first. With reports you drag fields onto a canvas. With a modern report you describe what you want, and an agent builds it. Today’s Learn recipe does exactly that with GitHub Copilot in VS Code. Note: GitHub Copilot is a separate product with its own license and billing; it’s not part of the Fabric Copilot capacity. AI isn’t an add-on anymore; it’s how the thing gets made.
- AI inside the app. Everyone who reads asks questions with Copilot, and report apps like my Finance Cockpit have an “Ask why” button backed by a Fabric data agent. Data agents consume Fabric capacity like Copilot does, and the Copilot capacity supports data agents on capacities smaller than F64.
- The semantic model is the foundation for all of it. Copilot in Power BI, data agents and the agent that builds your modern report all read the same semantic model. Descriptions, synonyms and hidden helper columns pay off three times.
My takeaway: an F2 Copilot capacity is the cheapest way to get your team used to working with AI on top of your semantic models today. That’s exactly the skill the next generation of Power BI front ends will ask for.
Takeaway
- Copilot in Power BI needs capacity, but not a big one.
- One F2, designated as a Fabric Copilot capacity, unlocks Copilot for a whole group of Pro, PPU and Desktop users.
- Four steps: tenant settings, home region, F2, designate and assign a group.
- Roughly a few hundred simple questions or about 20 report pages per day, depending on your scenario, for about $191 per month reserved.
- Start with a pilot, watch the Metrics app, scale on evidence.
Small investment. Massive boost.
Sources
- Fabric Copilot capacity
- Copilot consumption, usage, and billing in Fabric
- Copilot and Agent admin settings
- Find your Fabric home region
- Manage your Fabric capacity
- Copilot for Power BI overview
- Overview of Copilot in Fabric
- How Copilot in Microsoft Fabric works
- Pause and resume your capacity
- Microsoft Q&A: Copilot charges with F2 capacity during testing
- Azure Retail Prices API (Microsoft Fabric, checked November 4, 2026)
- What is Fabric Apps (preview)?
- Power BI data in Fabric Apps (preview)
- Power BI June 2026 update
- Create a semantic model app with GitHub Copilot
Numbers are list-price planning estimates. Copilot consumption rates can change; check the Learn consumption page before you size production.