TL;DR: You can show an external user a Power BI report that was built with Copilot's help — but external viewers cannot use Microsoft Copilot or Fabric Data Agents themselves unless they hold a paid Microsoft license (Power BI Pro/PPU or a Microsoft 365 Copilot license). Microsoft's AI features run under a licensed user's identity, which excludes the unlicensed external clients most agencies and SaaS teams serve. To give external viewers a "ask your dashboard a question" experience, you need a capacity-based embedding layer that handles the AI outside per-user licensing. That is exactly the gap embedded portals like DataTako close — and the upcoming DataTako agentic AI brings natural-language Q&A to external portal users without a Pro license.
The short answer, with the nuance that matters
There are two very different questions hiding inside "can I share AI Power BI reports externally?":
- Can I share a report that I built using Copilot? Yes. Once a report or semantic model exists, how it was authored is irrelevant to sharing. The output is a normal Power BI report.
- Can my external users use AI features — Copilot, Q&A, Data Agents — on that report? That is where Microsoft licensing gets in the way.
Most teams asking this question mean the second one. They want clients, not colleagues, to type "why did revenue drop in March?" and get an answer. And under native Microsoft licensing, that is currently very hard to deliver to external, unlicensed users.
Why Microsoft's AI features stop at the licensed-user boundary
Since 28 April 2025, Copilot in Power BI no longer requires the old F64 capacity floor — it runs on any paid Fabric capacity from F2 upward. That was a big democratization for internal teams.
But "available on F2" is not the same as "available to anyone." Microsoft's AI surfaces — Copilot in Power BI and Fabric Data Agents — operate under the identity of a licensed user. In practice that means each person interacting with the AI needs either:
- a Power BI Pro or Premium Per User (PPU) license, or
- a Microsoft 365 Copilot license (around $30/user/month) for Data-Agent-style experiences surfaced through Copilot Studio or M365 Copilot.
For an internal analytics team, fine. For the external audiences that BI agencies, SaaS products and customer portals serve, this model breaks down fast. You are not going to buy a Pro or M365 Copilot license for every client contact, and your clients are not going to buy one to read their own dashboard. This is the same wall that already exists for viewing Power BI externally — AI just makes it more expensive. (We unpack the underlying licensing in Power BI licensing: Pro vs Premium vs Embedded vs Fabric and which Power BI license is best for external sharing.)
The existing fix for viewing: capacity-based embedding
The established way to share Power BI with external users without per-viewer licenses is Power BI Embedded using the "App owns data" model on a Fabric capacity. Your application — not each end user — owns the connection, so you pay for capacity, not per head. We cover this in depth in What is Power BI Embedded? and the complete 2026 Power BI Embedded guide, and the licensing-free sharing pattern in share Power BI reports with external users without extra Pro licenses.
This is why portals exist: they take the capacity-based embedding model and add branding, authentication, row-level security and sub-organisations so each client only sees their own data. If you are weighing the options, 4 ways to share Power BI reports with external users and Power BI guest user access: pros, cons and alternatives lay out the trade-offs.
What's still missing: AI for those external viewers
Embedding solves viewing without per-user licenses. It does not, on its own, hand your external users a Copilot-style "ask a question" box — because, as above, Microsoft's AI runs under a licensed identity.
So the real 2026 question becomes: how do you give an unlicensed external viewer a natural-language, ask-your-data experience inside a branded portal?
That requires an agentic AI layer that:
- lives in the embedding/portal layer, so it serves viewers on capacity, not per-seat licenses;
- respects row-level security and tenant isolation, so a client can only ever query their own data;
- is white-label and multi-tenant, so it appears as part of your product, branded per client;
- can reason across multiple reports and datasets, not just one semantic model.
This is precisely what the upcoming DataTako agentic AI is built to do: let external, portal-based viewers ask their dashboards questions in plain language — branded as yours, isolated per (sub)organisation, and without a Power BI Pro license per user. (More on the distinction between Microsoft's tools in Power BI Copilot vs Fabric Data Agents, and on the category itself in what is agentic AI for analytics.)
A quick decision guide
- Internal team, everyone licensed? Native Copilot on F2+ is fine.
- You just need external users to view reports, no AI? Capacity-based embedding / a portal removes per-user Pro costs. See embedded analytics.
- You need external users to ask questions of their data, in your branding, without licensing each one? That's the embedded agentic-AI use case — the gap native Microsoft licensing leaves open.
FAQ
Can external users use Power BI Copilot?Not without a paid license. Copilot runs under a licensed user's identity (Power BI Pro/PPU or, for Data-Agent experiences, a Microsoft 365 Copilot license). Unlicensed external viewers can't invoke it natively.
Does Copilot still require F64 capacity?No. Since 28 April 2025 Copilot in Power BI is available on any paid Fabric capacity from F2 upward. Fabric Data Agents, however, carry additional per-user Copilot licensing requirements.
Can I share a report I built with Copilot to external users?Yes. How a report was authored doesn't affect sharing. The limitation is on external users using AI features themselves, not on distributing the finished report.
How do I give clients an "ask your dashboard" experience without buying each one a license?Use an embedded, capacity-based portal with its own AI layer rather than native Copilot. This serves AI to viewers on capacity instead of per-seat licenses, with row-level security per tenant.
Is this secure for multi-tenant client data?It can be, when the portal enforces row-level security and tenant isolation so each client can only query their own data.
