Clarification on Microsoft Fabric Copilot Charges with F2 Capacity During Testing

Hizqeel Emmanuel 20 Reputation points
2026-03-14T15:50:01.2966667+00:00

Hello,

Title: Clarification on Microsoft Fabric Copilot Charges with F2 Capacity During Testing

Hello,

I am currently testing Microsoft Fabric Copilot using an F2 capacity and would like clarification on the cost behavior I observed.

Test Scenario:

  • I created an F2 Fabric Capacity in the West Europe region.

I used Copilot in Power BI to generate a simple report page (Sales and Customer Overview dashboard).

The test lasted approximately 10 minutes, after which I paused the capacity.

According to Azure Cost Analysis, the charges were approximately:

Fabric Capacity – Copilot and AI: ~$0.57

  **Fabric Capacity – Compute Pool:** ~$0.10
  
     **Total:** ~$0.67

Observation:

When I created Fabric capacity in East Asia, Copilot features were not available due to region restrictions, and the incurred costs were significantly lower during testing.

Questions:

Is the $0.57 Copilot charge for generating a single report page within the expected cost range for F2 capacity?

Does using Copilot trigger additional Fabric compute pool meters beyond the base capacity runtime?

How exactly is Copilot usage metered in Microsoft Fabric? Is it based on number of prompts, compute time, or capacity unit consumption?

Is there a recommended best practice for testing Copilot features while minimizing Fabric capacity costs?

Regarding Fabric capacity billing, is the F2 capacity charged at a fixed hourly rate regardless of actual usage, or is the billing calculated based on the actual compute usage time (for example, per minute or per second) when the capacity is running?

I would appreciate clarification so I can better understand the billing model for Copilot workloads and Fabric capacity usage.

Thank you.

Cost Management
Cost Management

A Microsoft offering that enables tracking of cloud usage and expenditures for Azure and other cloud providers.

0 comments
Answer accepted by question author
Siva shunmugam Nadessin 10,905 Reputation points Microsoft External Staff Moderator
2026-03-16T11:16:31.15+00:00

Hello Hizqeel Emmanuel,

Thank you for reaching out to the Microsoft Q&A forum.

Thanks for the detailed scenario – your numbers actually line up with how Copilot in Fabric is metered. Here’s the TL;DR:

  1. You’re seeing ~$0.57 of “Copilot and AI” because Copilot calls into Azure OpenAI and eats Capacity Units (CUs) from your F2 SKU.
    • Copilot consumption = (input tokens × 100 CU/1 000 + output tokens × 400 CU/1 000), smoothed over a 24 h window as a “background job.”
    • On an F2 (2 CU/hour) even a few‐thousand‐token request can consume ~0.1 CU-hours, which at F2’s per-CU rate is roughly $0.50 for a quick test.
  2. There is no extra “Compute Pool” meter for Copilot – it all comes out of your one Fabric capacity. That’s why you see two line items:
    • Fabric Capacity – Compute Pool (~$0.10): your 10 minutes of F2 running
    • Fabric Capacity – Copilot and AI (~$0.57): the CU-hours Copilot burned.
  3. Billing for F2 is simply “pay-as-you-go per second” while the capacity is running. It’s a flat wall-clock charge (you pay from Start → Pause), not “only when CPU is busy.” Copilot doesn’t add separate Azure meters or change the per-second rate.
  4. Best practices to cut costs when testing Copilot: • Only spin up (and keep running) your F2 capacity when you’re actually poking at Copilot. Pause it immediately afterward.
  5. Keep prompts tight (fewer input/output tokens).
  6. For heavier Copilot experimentation, either use a higher SKU briefly (more CUs/hour lowers $/CU) or use a split-capacity pattern (an F64 dedicated to Copilot so your Production capacity stays small).
  7. Region restrictions: if Copilot isn’t available in East Asia, the AI CU meter won’t emit, so you’ll only see the base compute cost (that’s why your East Asia tests look cheaper).

Bottom line: yes, ~$0.57 for a 10-minute, token-heavy Copilot request on F2 is expected. You’re billed per-second for running F2 (the ~$0.10), and Copilot simply consumes CUs out of that pool (the ~$0.57). Hope that clears things up!

Reference links

• How Copilot in Microsoft Fabric works – Cost of Copilot in Fabric https://learn.microsoft.com/fabric/fundamentals/how-copilot-works#cost-of-copilot-in-fabric

• Copilot in Fabric consumption – Capacity utilization type & token-based meter https://learn.microsoft.com/fabric/fundamentals/copilot-fabric-consumption#capacity-utilization-type

• Buy subscription SKUs – Pay-as-you-go vs reserved instance billing model https://learn.microsoft.com/fabric/enterprise/buy-subscription#sku-types

• Throttling & carryforward (so you know what happens if you exhaust CUs) https://learn.microsoft.com/fabric/enterprise/throttling#throttle-triggers-and-throttle-stages

Was this answer helpful?

1 additional answer

Sort by: Most helpful
Most helpful Newest Oldest

  1. AI answer

    2026-03-14T15:50:26.1+00:00

    Copilot usage and Fabric capacity billing are working as expected in this scenario, based on how Copilot and capacity are metered.

    1. Expected size of the Copilot charge

    Copilot in Fabric is billed by token consumption, not by time or by “number of reports.” The documented consumption rates are:

    • Input prompt: 100 CU-seconds per 1,000 tokens
    • Output completion: 400 CU-seconds per 1,000 tokens

    Approximately 1,000 tokens ≈ 750 words. A single Copilot interaction that generates a full report page can involve multiple prompts and completions under the hood (for example, understanding the model, generating measures, visuals, and descriptions). The resulting “Copilot and AI” meter charge can therefore be noticeably higher than the base compute charge, even for what appears to be a single user action.

    The specific dollar amount (~$0.57) depends on the price per 1,000 tokens and per CU-second for the region and SKU, which isn’t detailed in the context. However, the pattern of seeing a distinct “Copilot and AI” line item that is larger than the “Compute Pool Capacity Usage CU” for a short test is consistent with the documented model where Copilot is billed separately by token usage.

    1. Whether Copilot triggers additional Fabric compute meters

    Copilot consumption is:

    • Measured and surfaced as “Copilot in Fabric” in the Fabric Capacity Metrics app.
    • Billed via the “Copilot and AI” meter in Azure Cost Management.

    In addition:

    • Any downstream actions that Copilot triggers (for example, DAX queries, data refreshes, email subscriptions) are billed separately through their standard Fabric pathways, not as Copilot.
    • This means Copilot can indirectly cause usage on other meters (such as Data Warehouse, Data Movement, etc.) if it initiates operations that use those services.

    So yes, using Copilot can result in additional Fabric compute meters beyond the Copilot meter itself, but those are billed as normal Fabric usage, not as Copilot. The “Compute Pool Capacity Usage CU” line you see is the standard capacity usage meter; Copilot usage is in addition to that and appears under “Copilot and AI.”

    1. How Copilot usage is metered

    Copilot in Fabric is metered by tokens, which are then translated into capacity unit (CU) seconds:

    • Unit of measure: tokens (input and output), reported as CU-seconds per 1,000 tokens.
    • Rates:
      • Input: 100 CU-seconds / 1,000 tokens
      • Output: 400 CU-seconds / 1,000 tokens

    Key points:

    • Copilot usage is not billed per “number of prompts” directly; each prompt/response pair consumes tokens.
    • All Copilot CU consumption is processed as background capacity operations to smooth demand and avoid spikes.
    • Billing for Copilot is driven only by token consumption. Any non-Copilot operations that run as a result are billed via their own meters.

    In the Metrics app, Copilot usage appears as “Copilot in Fabric” and can be filtered and broken down by experience (for example, Power BI report vs semantic model).

    1. Best practices to test Copilot while minimizing cost

    From the documented guidance:

    • Use the Fabric Capacity Metrics app to monitor Copilot usage:
      • Filter on the “Copilot in Fabric” operation.
      • Use the Item history page to see which items and experiences are consuming Copilot CUs.
    • If testing causes throttling or high background usage, consider temporarily scaling capacity up only when needed and then scaling back down. This is a reactive measure to handle short-term spikes.
    • For larger environments or heavy Copilot testing, consider a split-capacity strategy:
      • Place Copilot workloads on a separate capacity (for example, a dedicated F64+ Copilot capacity) so Copilot consumption does not impact other workloads.
      • Copilot can work against items in other capacities while its own consumption is isolated to the Copilot capacity.

    For small-scale testing on F2, practical cost-minimizing steps include:

    • Keep the capacity running only during active tests, then pause it.
    • Limit the number and length of Copilot interactions (each interaction consumes tokens).
    • Use the Metrics app frequently to understand which actions are driving Copilot consumption.
    1. How F2 Fabric capacity is billed (time granularity)

    For Fabric capacity in Azure billing:

    • The capacity itself (for example, F2) is billed via the capacity usage meters (such as “Compute Pool Capacity Usage CU”) that represent the available capacity.
    • These meters appear in Microsoft Cost Management as part of the Fabric capacity resource.

    The context describes the meters and how they appear on the bill but does not specify the exact time granularity (per second, per minute, or per hour). It does state that:

    • Capacity usage is tracked as Capacity Units (CUs).
    • Copilot and other workloads consume CU-seconds against that capacity.

    The “Copilot and AI” meter is in addition to the base capacity meters and is driven by token-based CU consumption, not by the simple runtime of the capacity.

    In summary:

    • Copilot is billed by token-based CU consumption (input/output tokens → CU-seconds → “Copilot and AI” meter).
    • Copilot can indirectly drive additional usage on other Fabric meters when it triggers downstream operations.
    • Copilot usage appears as “Copilot in Fabric” in the Metrics app and “Copilot and AI” in Azure Cost Management.
    • Capacity itself is billed via its own meters representing available capacity; Copilot charges are separate and additive.

    References:

    AI-generated content may be incorrect. Read our transparency notes for more information.

    Was this answer helpful?

Your answer

Answers can be marked as 'Accepted' by the question author and 'Recommended' by moderators, which helps users know the answer solved the author's problem.