Copilot for Fabric on F-SKU capacity (F2 floor April 2025)

What it is

Microsoft Fabric Copilot is the AI-assisted authoring and analytics layer embedded across Fabric's workloads — Power BI, Data Factory, Synapse Data Engineering, Data Science, and Real-Time Intelligence. It supports natural-language report creation, DAX measure generation, notebook code completion, pipeline authoring, and conversational data exploration. What distinguishes it from most enterprise Copilot rollouts is that its availability is gated at the capacity level rather than the per-user licence level: Copilot only surfaces inside a workspace when that workspace is backed by Fabric capacity of at least the F2 tier, or by a Power BI Premium capacity (P1 or above — P1 is equivalent to F64). A workspace running on the free trial capacity type simply does not expose Copilot features, regardless of what licences individual users hold. Microsoft has since added a Fabric Copilot capacity: usage from Pro, Premium-per-user and trial-licensed workspaces can be billed to a designated F2-or-higher or P1 capacity, so the paid-capacity requirement attaches to the capacity that is billed rather than to every workspace.

Why it matters

• Fabric capacity billing scales in powers of two (F2 through F2048), so F2 is the cheapest paid tier — and since April 2025 Copilot unlocks from F2 upward (before that, only from F64/P1).
• Prompts and semantic-model metadata go to Azure OpenAI Service endpoints within the tenant's Azure geography by default, and are not used to train foundation models.
• Tenant admins can disable Copilot entirely via the Fabric Admin Portal if data-sovereignty policy requires it.

Key points

• Copilot for Fabric is gated at the capacity level, not per user: it only appears in workspaces backed by at least an F2 capacity (or a Power BI Premium P1+ capacity; P1 ≈ F64).
• Until April 2025 Copilot required F64/P1; since 30 April 2025 it is available on every paid F SKU from F2 — trial capacity is excluded.
• The F2 floor is roughly $263 per month in West Europe (check the current price list) — cheap for an enterprise, but a real gate for a departmental pilot.
• When a client compares SAC and Power BI on AI features, ask which Fabric capacity they actually run before assuming Copilot is available.
• Outside the US and France, Copilot is disabled by default at the tenant level unless an admin explicitly enables cross-region processing of prompts and metadata by Azure OpenAI as a Microsoft subprocessor — a genuine default-governance difference from Joule's centrally configured masking/filtering.
• The commercial model has no SAP equivalent: Joule prices what an agent does (AI Units per action), Copilot for Fabric prices what the workspace can do (a capacity-tier floor) — model both explicitly in a TCO comparison, never assume a shared unit of cost.
• Copilot for Fabric is primarily a single-turn, workload-embedded assistant (DAX generation, pipeline authoring, Q&A), not a multi-step agent orchestrator — for genuine agentic workflows on Microsoft's stack, the relevant comparison is a Microsoft Foundry agent, not Copilot itself.
• Tenant admins retain a hard kill switch in the Fabric Admin Portal, separate from the data-residency setting, for organisations that need to disable Copilot entirely regardless of capacity.

Terms used on this page

F-SKU

Microsoft Fabric's capacity unit tier, scaling in a doubling sequence from F2 through F2048; the compute unit that prices all Fabric workloads including Copilot.

Data residency setting (Fabric)

A tenant-level toggle controlling whether prompts and metadata sent to Azure OpenAI as a Microsoft subprocessor may be processed outside the capacity's geographic region — off by default outside the US and France.

Fabric Admin Portal kill switch

The tenant-wide control, separate from the data-residency setting, that lets an administrator disable Copilot entirely regardless of capacity tier or licensing.

OneLake shortcut

A Fabric feature making an external or mirrored data source appear as a virtual folder inside a Fabric Lakehouse without copying it into Fabric storage — the mechanism underlying BDC Connect for Fabric (C167).

Capacity-level feature gating

A licensing pattern where a feature's availability depends on the compute tier backing a workspace rather than on individual user licences — Copilot for Fabric's model, in contrast to Joule's per-action AI Unit metering.

Azure OpenAI as a Microsoft subprocessor

The specific legal/technical framing Microsoft uses for how Copilot's underlying model calls are routed through Azure OpenAI Service on the customer's behalf — the term that appears in the tenant setting governing data residency.

Sources

1. Microsoft Learn — Copilot for Microsoft Fabric overview
2. Microsoft Azure pricing — Microsoft Fabric F-SKU
3. Microsoft blog — Copilot for Fabric April 2025 F-SKU expansion
4. Microsoft Fabric — documentation
5. Microsoft Learn — Enable and configure Copilot in Microsoft Fabric
6. Microsoft Learn — Privacy, security and responsible AI use of Copilot in Fabric
7. Microsoft Learn — Copilot for Microsoft Fabric and Power BI: FAQ
8. Microsoft Learn — OpenAI as a subprocessor in Microsoft Fabric
9. Microsoft Learn — Fabric Copilot Capacity for usage billing
10. Microsoft Learn — Configure Fabric data agent tenant settings
11. Microsoft Q&A — Clarification on Microsoft Fabric Copilot charges with F2 capacity during testing
12. Microsoft Learn — What's new in Microsoft Fabric (September 2026: Fabric IQ in Microsoft 365 Copilot GA; Copilot worldwide)
13. CIO.com — Microsoft updates Fabric for agentic transformation (29 Sep 2026 announcements, trade press)