Copilot on F2: what small Fabric teams should check first

Until now, Copilot in Microsoft Fabric has been a big-capacity feature. If your workspaces did not sit on an F64 or larger, you watched the demos and moved on. Today at FabCon in Las Vegas, Microsoft announced that Copilot and AI capabilities will be enabled for all paid Fabric SKUs, from F2 and up, within the coming weeks.

That is a real change for the people I work with most. Small teams, single-department setups and consultancies tend to run F2, F4 or F8, because that is what the budget allows. For them, the question shifts from "can we get Copilot?" to "can our capacity afford it?".

What was announced

The announcement sits in Arun Ulag's FabCon 2025 post on the Microsoft Fabric blog. The wording is short: Copilot and AI capabilities will be enabled for all paid SKUs in Fabric, and customers on F2 and above can use them, including Fabric data agents.

Microsoft says the change is driven by customer feedback about the productivity impact of Copilot, and that broader access would benefit more teams. The rollout is described as happening "within the coming weeks", so do not expect it to be live on your capacity this afternoon.

Two other FabCon items make the timing interesting. Data agents (formerly AI skills) can now be connected to Azure AI Foundry, and AI functions arrived in preview, letting you apply LLM-powered transformations such as summarisation and classification to OneLake data with a single line of code. Both are AI capabilities that small capacities can now start to look at.

Paid means paid

The key word is "paid". The documentation is clear that the Azure OpenAI Service behind Copilot in Fabric is not available on trial SKUs. You need a paid F SKU (F2 or higher) or a Power BI Premium P SKU.

So if you have been evaluating Fabric on a trial capacity and planning to test Copilot there, that will not work. You need a paid capacity, even a small one, and your workspace has to be assigned to it.

Region matters too. The Azure OpenAI Service behind Copilot in Fabric runs in US datacenters and in the EU data boundary. For Danish customers on a capacity in an EU region, that is the easy case. If your capacity sits outside the US or the EU, Copilot is disabled unless your Fabric admin enables the tenant setting that allows data to be processed outside your capacity's geographic region.

How Copilot uses your capacity

This is the part that matters most on a small SKU. Copilot is not a separate meter. Every request consumes capacity units (CUs) from the same capacity that runs your pipelines, notebooks and semantic model refreshes.

The consumption is based on tokens:

• Input prompts: 100 CU seconds per 1,000 tokens.
• Output completions: 400 CU seconds per 1,000 tokens.

Microsoft's own example is a request with 2,000 input tokens and 500 output tokens. That costs 400 CU seconds, or about 6.67 CU minutes.

Copilot operations are classified as background jobs, which means their usage is smoothed over 24 hours instead of hitting your capacity all at once. Microsoft calculates that an F64 can run over 13,824 such requests per day before the capacity is exhausted.

Scale that same example down and the picture for small teams looks like this (my arithmetic, using Microsoft's example request):

• F2: roughly 430 requests per day.
• F4: roughly 860 requests per day.
• F8: roughly 1,700 requests per day.

Those numbers assume the capacity does nothing else. In real life your F2 is already busy with refreshes and pipelines. Real prompts also vary a lot in size, and a long conversation with schema context sends far more tokens than the example. Treat the figures as an upper bound, not a budget.

Tenant settings to check

Copilot and agents are controlled by the Copilot and AI group in the Fabric tenant settings. The main switch is "Users can use Copilot and other features powered by Azure OpenAI". It can be managed at both tenant and capacity level, and it can be scoped to security groups.

Things worth checking with your Fabric admin:

1. Is the Azure OpenAI setting on, and for whom? Scoping it to a small group first is sensible.
2. Is the setting delegated to capacity admins? If so, check the capacity-level setting too.
3. If your capacity is outside the US or the EU, decide deliberately whether to allow cross-region processing.
4. Look at the setting for designating capacities as Copilot capacities, so you know how usage and billing could be consolidated.

What this means for you

For a small team, I would not switch Copilot on for everyone the day it lands. A capacity that is throttled because a dozen people asked Copilot to rewrite their DAX is not a good outcome, and on an F2 the margin is thin.

A sensible approach:

1. Wait until the rollout reaches your capacity, then enable Copilot for a pilot group only.
2. Open the Fabric Capacity Metrics app and find the operation named "Copilot in Fabric".
3. Run your normal workload for a week with the pilot using Copilot as they would in daily work.
4. Compare the Copilot share of CU usage with your headroom before you widen access.
5. If the numbers are tight, decide whether Copilot is worth an F4 or F8, rather than letting it squeeze your refreshes.

Microsoft's documentation recommends exactly this incremental approach: pilot, measure in the Metrics app, then extrapolate before you scale.

Takeaway

After two decades in data, I am glad to see a feature like this move down to the price points that small teams can actually buy. Access is no longer the blocker. Capacity is, so the work now is to measure before you roll out.

Are you on an F2, F4 or F8, and which Copilot experience would you test first on it?

Sources

• FabCon 2025: Fueling tomorrow's AI with new agentic capabilities and security innovations in Fabric (Microsoft Fabric blog)
• Consumption rates and billing for Copilot in Fabric (Microsoft Learn)
• Copilot and Agent tenant settings (Microsoft Learn)
• What is Copilot in Fabric? (Microsoft Learn)
• Enable and configure Copilot in Microsoft Fabric (Microsoft Learn)