Disclosure: This article is published by Upsolve AI. Where our product appears alongside others, the comparison is based on publicly available information as of November 2026.

Power BI Copilot is not one product. It is a set of chat and authoring features billed three different ways, through Fabric capacity units, a per-user Microsoft 365 Copilot license or pay-per-use credits, depending on where the question gets typed. Microsoft has also said the standalone experience will move into Microsoft 365 Copilot while the report pane stays, so the bill you plan for today may not be the one you pay next year.

What Copilot in Power BI does for authors and consumers

What you get depends on who is asking and what access they hold, per Microsoft’s overview.

Two details change how far to trust the output. For consumers, Copilot can write ad hoc DAX, and a Fabric IQ option in the pane shows the reasoning steps and DAX behind an answer. Checking is possible, but someone still has to do it.

For authors, DAX gets a syntax check, not a logic check. In DAX query view, Copilot runs the response through a parser and requests a new one if the query won’t run. A query that runs can still calculate the wrong thing.

How the standalone Copilot experience changed

The sequence is scattered across several Microsoft posts.

1. September 2025: on by default. Microsoft enabled the standalone experience for every tenant that already had Copilot on, and admins had to opt out. It is still labeled preview.

2. August 26, 2026: data agents removed. Copilot in Power BI stopped connecting to Fabric data agents, because the integration was built on an OpenAI API being retired. The agents remain available elsewhere, including Microsoft 365 Copilot.

3. 2026: consolidation announced. Microsoft acknowledged the complaint that there are too many Copilots and said standalone will soon be upgraded to the Microsoft 365 Copilot experience. The report Copilot stays supported, but in Microsoft’s words, “the investment is going to Microsoft 365 Copilot and Copilot Cowork.”

4. Today: both run side by side. A new Copilot in Power BI and Fabric preview adds Microsoft 365 context such as emails and meetings inside Power BI. It is off by default, requires a Copilot Premium license, and coexists with the old experience. A later rollout to users without that license is planned, with limits to be announced.

There’s a naming trap. Power BI answers in Microsoft 365 Copilot Chat are generally available, but Microsoft lists a Microsoft 365 Copilot Premium license as required for all users. The Copilot Chat included with eligible Microsoft 365 plans doesn’t get you there, despite the shared name.

Work on the semantic model survives this reshuffle, because every surface reads from it. Training and budgets built around one chat window might not.

What Power BI Copilot needs, and what it costs

Capacity and licensing

The requirements are a paid capacity, the Azure OpenAI tenant setting left on (it is on by default), and a supported region. Sovereign clouds aren’t supported, and neither are Private Link or closed network environments. After you buy or scale capacity, Copilot can take up to 24 hours to notice.

Three things then catch people out.

Pro shops don’t have to move their workspaces. A Fabric Copilot capacity is a capacity you designate to collect Copilot usage for assigned users, including in Pro and PPU workspaces and in Power BI Desktop. It can now sit on an F2, which lists at $156.33 a month reserved. It must be in the tenant’s home region, and it doesn’t cover Embedded license mode.

The standalone bill can land on someone else’s capacity. Without a Fabric Copilot capacity, standalone Copilot picks a workspace for billing from a partly randomized list of eligible workspaces, weighted toward those with more free capacity. Users can change it. So one team’s budget can quietly fund another team’s chat. Assigning users to a Fabric Copilot capacity overrides this.

Viewers still need licenses. On F SKUs below F64, each user viewing Power BI content needs Pro, PPU or an individual trial. F64 or larger lets Free users with a viewer role view it. Pro lists at $14 a user a month and is included in Microsoft 365 E5.

A worked cost example

Copilot consumes capacity by the token: 100 CU seconds per 1,000 input tokens, 10 for cached input and 400 for output. Microsoft’s example request is 2,000 input and 500 output tokens, which comes to 400 CU seconds.

Assumptions (ours, not Microsoft’s): 200 active users, five questions each per working day, so 1,000 a day and 21,000 over 21 working days. Each question is the size of Microsoft’s example request. Prices are US list as of November 2026. Query compute triggered by Copilot is excluded, because it bills separately to the model’s capacity.

1. Daily load: 1,000 × 400 = 400,000 CU seconds.

2. Capacity per day: an F4 provides 4 × 86,400 = 345,600 CU seconds, which is too small. An F8 provides 691,200, so this load takes 58% of it.

3. Bill: an F8 is $625.33 reserved or $1,051.20 pay-as-you-go.

4. Per question: $625.33 ÷ 21,000 = about $0.03.

Copilot runs as a background job, smoothed over 24 hours, so you size for your busiest day, not a morning spike. The capacity still bills on weekends.

The token assumption does the most work. Microsoft’s figure is an illustration, not an average, and multistep answers and longer prompts cost more. If each question used five times the tokens, the load would be 2,000,000 CU seconds a day. That overflows an F16 (1,382,400) and takes 72% of an F32 (2,764,800), at $2,501.33 reserved, or about $0.12 a question. Measure yours: the Capacity Metrics app shows Copilot usage by experience over 30 days.

Below F64, Pro for those 200 viewers adds 200 × $14 = $2,800, if they don’t already have it. Reserved F64 costs $5,002.67, or $4,377.33 more than a reserved F8. Divide by $14 and you need about 313 viewers who would otherwise buy Pro before F64 wins. On pay-as-you-go the figure is about 526. If E5 already covers Pro, there’s no saving to chase.

The Microsoft 365 route is 200 × $30 for the add-on = $6,000, assuming that add-on is the Premium license Microsoft’s docs name. It buys all of Microsoft 365 Copilot, not just Power BI answers, so compare it only if you’d want those seats anyway. Copilot Cowork bills usage-based Copilot Credits on top.

Two capacity risks sit outside the example:

• Chat isn’t the only consumer. Microsoft warns that subscription summaries sent to hundreds of users and AI narrative visuals on busy dashboards can exhaust a capacity.

• Shared capacity shares the failure. Once a capacity is exhausted, all operations shut down. A separate Fabric Copilot capacity takes the token load off the capacity serving your production reports.

• 
  

Copilot limits worth knowing before rollout

All but the last come straight from Microsoft’s documentation.

• The pane ignores your slicers. Copilot data questions don’t apply the filters or slicers currently affecting the report page. A user filtered to EMEA can get an answer that isn’t about EMEA.

• It answers what, not why, in English. Microsoft lists forecasting, anomaly detection, key influencers and questions about why a pattern happens as unsupported. English is the only supported language for data questions.

• It may answer from general knowledge. When a question isn’t tied to the semantic model, Copilot might respond from the language model’s general knowledge rather than your data.

• Fabric IQ turns off verified answers. With Fabric IQ on in the pane, Copilot doesn’t return verified answers. Authors who wrote them should test both modes.

• Q&A is retiring, but you can’t switch it off yet. Q&A ends in February 2027 and reports with Q&A visuals will show an error. Yet data questions use the same engine, and Prep data for AI still requires Q&A enabled. Remove the visuals, not the setting.

• Only 200 characters of each description count. Copilot reads just the first 200 characters of a measure or column description, so put the business meaning first.

• Repeat prompts return the cache. The same prompt on an unchanged model within a rolling 24 hours gets the cached response, and clearing the chat doesn’t change that.

• Approval is per model, and it’s a warning. Unapproved models show users a warning, but they can view the answer anyway. An admin setting limits standalone search to approved items. There’s no approval for a report, dashboard or app.

• Models are open to AI by default. The tenant setting that exposes Fabric data to Microsoft 365 Copilot is enabled by default, and Restrict from Copilot lets admins control which semantic models take part. A partner summary from 2toLead says that per-model setting starts in the allowed position. Row-level and object-level security still apply, but a model few people ever opened can now surface in a chat answer.

When a dedicated data agent fits better than Copilot

Often it doesn’t. If your data is already modeled in Power BI, your users are internal and sit in your Microsoft tenant, and someone owns model quality, use Copilot. Start with the report pane and an F2 Fabric Copilot capacity.

A dedicated agent like Upsolve earns its cost in four situations.

Customer-facing analytics. This is the clearest case. In app-owns-data embedding, your end users are unlicensed, and Microsoft has no plans for a Copilot replacement for embedded Q&A there. Its suggestions are a SaaS embedding pattern, narrative visuals, or building on the Fabric IQ MCP server, which has no natural-language answering tool: your agent composes the DAX. Upsolve’s embedded agent signs in your users through your app, with each seeing only their own data. To price the Microsoft route, see our Power BI Premium vs Embedded breakdown.

Data that isn’t in one semantic model. Copilot answers from Power BI reports and semantic models, so data has to be modeled there first, through import, DirectQuery or a composite model. Upsolve works across Snowflake, BigQuery, Databricks, Redshift and Postgres at the same time, and doesn’t need an existing semantic layer to start.

Context you test like code. Microsoft advises you to compare responses with known results, by hand. Upsolve treats golden assets as test cases (tests are in beta), scores answers with LLM-as-judge, and keeps full traces of each answer. The difference is how you catch a regression, not a claim about accuracy.

Users outside Microsoft’s surfaces. Upsolve answers in Slack, Teams and any MCP client.

Upsolve is priced by plan and credits rather than per user. Team lists at $2,000 a month with 10,000 credits and adds row-level security, RBAC and embedding, and a standard analytical question takes 5-10 credits. Weighing a full replacement instead? See our Power BI alternatives and the wider AI analytics tools roundup.

Power BI Copilot vs Microsoft 365 Copilot vs Upsolve

As of November 2026, US list prices.

Frequently Asked Questions

Is Copilot available for Power BI?

Yes, in the Power BI service, Desktop and mobile, with access to a paid F2 or P1 capacity or higher. The report pane is generally available. Standalone Copilot, Copilot in apps and web modeling are in preview.

What is the difference between Power BI and Copilot?

Power BI is the BI platform: semantic models, reports, dashboards and apps. Copilot is the generative AI layer on top. It reads from your model, so a messy model produces weak answers.

How to set up Copilot for Power BI?

Get a paid capacity, in a supported region, then confirm the Azure OpenAI tenant setting is on. Assign the workspace to that capacity, or assign users to a Fabric Copilot capacity. In Desktop, pick a Copilot-enabled workspace on first use. Then prepare the model with Prep data for AI and mark it Approved for Copilot.

Does Power BI Pro include Copilot?

No. A Pro or PPU license alone isn’t sufficient. The cheapest route for a Pro shop is an F2 set up as a Fabric Copilot capacity, at $156.33 a month reserved. By our arithmetic, that handles 432 of Microsoft’s example requests a day.

How much does Power BI Copilot cost?

On capacity, you pay for the SKU, not per seat: $156.33 a month reserved for an F2, up to $5,002.67 for an F64. Microsoft’s example request of 400 CU seconds is roughly $0.02 at the $0.18 per CU hour implied by pay-as-you-go list pricing. Viewers below F64 also need Pro at $14.

Which AI tool is best for Power BI?

For report readers on a well-modeled semantic model, the report pane. For people who want Power BI numbers next to their email and documents, Microsoft 365 Copilot. For customer-facing analytics, or data spread across warehouses, look at a dedicated data agent.

Serving analytics to your own customers, or working across several warehouses? See Upsolve’s pricing or book a demo.

Sources

1. Microsoft Learn: Copilot for Power BI overview. https://learn.microsoft.com/en-us/power-bi/create-reports/copilot-introduction

2. Microsoft Learn: Copilot in Power BI integration overview. https://learn.microsoft.com/en-us/power-bi/create-reports/copilot-integration

3. Microsoft Learn: Enable Fabric Copilot for Power BI. https://learn.microsoft.com/en-us/power-bi/create-reports/copilot-enable-power-bi

4. Microsoft Learn: Ask data questions with Copilot and Fabric IQ. https://learn.microsoft.com/en-us/power-bi/create-reports/copilot-ask-data-question

5. Microsoft Learn: Prepare your data for AI. https://learn.microsoft.com/en-us/power-bi/create-reports/copilot-prepare-data-ai

6. Microsoft Learn: Optimize your semantic model for Copilot, including the 200-character limit. https://learn.microsoft.com/en-us/power-bi/create-reports/copilot-evaluate-data

7. Microsoft Learn: Consumption rates and billing for Copilot in Fabric. https://learn.microsoft.com/en-us/fabric/fundamentals/copilot-fabric-consumption

8. Microsoft Learn: Fabric Copilot capacity. https://learn.microsoft.com/en-us/fabric/enterprise/fabric-copilot-capacity

9. Microsoft Learn: Understand Microsoft Fabric licenses and capacity. https://learn.microsoft.com/en-us/fabric/enterprise/licenses

10. Microsoft Learn: Copilot in Power BI and Microsoft Fabric (preview). https://learn.microsoft.com/en-us/fabric/iq/connectors/copilot-power-bi-fabric

11. Microsoft Learn: Fabric IQ in Microsoft 365 Copilot Chat. https://learn.microsoft.com/en-us/fabric/iq/connectors/microsoft-365-copilot-overview

12. Microsoft Learn: Get started with the Fabric IQ MCP server. https://learn.microsoft.com/en-us/fabric/iq/connectors/fabric-iq-mcp

13. Microsoft Azure: Microsoft Fabric pricing. https://azure.microsoft.com/en-us/pricing/details/microsoft-fabric/

14. Microsoft: Power BI pricing. https://www.microsoft.com/en-us/power-platform/products/power-bi/pricing

15. Microsoft: Microsoft 365 Copilot enterprise pricing. https://www.microsoft.com/en-us/microsoft-365-copilot/pricing/enterprise

16. Power BI Updates Blog: Standalone Copilot in Power BI will be turned on by default in September. https://community.fabric.microsoft.com/blog/fbc_pbiupdatesblog/standalone-copilot-in-power-bi-will-be-turned-on-by-default-in-september/5174102

17. Power BI Updates Blog: Bringing Power BI Insights to Every Copilot User. https://community.fabric.microsoft.com/blog/fbc_pbiupdatesblog/bringing-power-bi-insights-to-every-copilot-user/5309360

18. Power BI Updates Blog: Power BI Q&A retirement reminder, February 2027 timeline update. https://community.fabric.microsoft.com/blog/fbc_pbiupdatesblog/power-bi-qa-retirement-reminder-february-2027-timeline-update/5365841

19. Fabric Updates Blog: Retirement of Fabric data agent integration in Copilot in Power BI. https://community.fabric.microsoft.com/blog/fbc_fabricupdatesblogs/retirement-of-fabric-data-agent-integration-in-copilot-in-power-bi/5328344

20. 2toLead: Fabric IQ in Copilot, Restrict from Copilot for Power BI (partner summary of the FabCon Europe 2026 announcements). https://www.2tolead.com/insights/fabric-iq-copilot-restrict-from-copilot

21. Upsolve AI: Pricing. https://upsolve.ai/pricing

22. Upsolve AI docs: Embed Upsolve in your product. https://docs.upsolve.ai/embedding

23. Upsolve AI docs: Golden Assets. https://docs.upsolve.ai/ai-agent-builder/golden-assets

24. Upsolve AI docs: Agent Observability. https://docs.upsolve.ai/ai-agent-builder/agent-observability

25. Upsolve AI docs: Use Upsolve from Claude and other AI assistants (MCP server). https://docs.upsolve.ai/ai-agent-builder/deploy-agents/mcp-server

26. Microsoft Learn: Fabric IQ in Copilot Cowork, generally available. https://learn.microsoft.com/en-us/fabric/iq/connectors/cowork-overview

27. Microsoft Fabric Community: Fabric Analytics at FabCon Europe, including Restrict from Copilot. https://community.fabric.microsoft.com/blog/-/Bringing-governed-analytics-into-the-flow-of-work-Fabric/5368918