Quick answer. As of September 2026, Microsoft Learn says Power BI Q&A experiences are going away in February 2027 and recommends Copilot for Power BI. According to Microsoft Learn, Copilot needs paid Fabric capacity at F2 or higher or Power BI Premium at P1 or higher.
Trial capacities and free SKUs are not supported, according to Microsoft’s Copilot for Power BI overview page, and a Fabric Copilot capacity on at least an F2 or P1 SKU is the route for Pro and Premium Per User workspaces. Copilot is already enabled by default for tenants with paid Fabric capacity, and for a capacity in the US, data processed by Copilot in Fabric stays in the capacity’s geographic region unless an admin allows otherwise, but the separate setting that lets Copilot use OpenAI as a Microsoft subprocessor, off by default, is one that Microsoft’s OpenAI as a subprocessor in Microsoft Fabric page says allows data to leave Fabric’s compliance boundary for infrastructure that may have lower compliance controls and may not comply with FedRAMP. Microsoft’s What is Copilot in Fabric? page states that Copilot “isn’t yet supported for sovereign clouds due to GPU availability”, and Microsoft Learn as read in September 2026 does not name Copilot for Power BI’s status in GCC, GCC High or DoD, so a government-cloud tenant confirms its status with Microsoft and plans a report-level fallback before February 2027.
Every morning a finance operations team opens the same dashboard and types a question into the box at the top: overdue receivables by region, this week against last. Clinical operations does the same with bed occupancy. Neither team knows the box has a name. Then the notice appears at the top of Microsoft’s Q&A pages, and the date in the IT Director’s inbox depends on which post he read. When he looks at the recommended replacement, he finds that Copilot is already on for the tenant’s paid capacity, that a capacity administrator could have switched it on anyway, and that one of its settings carries a warning his privacy office will want to read. If he waits, the question boxes disappear on a Microsoft date and the workspaces users rely on turn out not to qualify for Copilot. If he lets the defaults stand, Copilot answers from models no one certified while its capacity use competes with production reports. This guide sets out what goes away, what Copilot needs and already switches on, which settings decide where report data is processed, and what a government-cloud tenant does now.
What goes away in February 2027, and what to inventory first
The first failure is a short inventory: teams count the dashboards with a question box and miss the other places report users type questions. Microsoft’s Introduction: Use natural language to explore data with Power BI Q&A page states: “Q&A experiences are going away in February 2027.” The same page states: “We recommend using Copilot for Power BI, which offers a more advanced and integrated way to query your data using natural language.”
That notice is not confined to one page. As read in September 2026, the same February 2027 notice and the same recommendation sit on Microsoft’s Create a Q&A visual in a report in Power BI page, its Q&A for Power BI business users page and its Intro to Q&A tooling to train Power BI Q&A (preview) page. Some earlier posts gave December 2026 as the date; Microsoft’s Learn pages as read in September 2026 say February 2027, and that is the date to plan against.
Microsoft’s Q&A for Power BI business users page adds: “Q&A is available in Power BI iOS apps and in Power BI embedded analytics as well.” Microsoft’s Embed a standalone Q&A visual page, which covers Q&A in embedded analytics, carries no retirement notice as read in September 2026 and does not describe a replacement, so the owner of any application that embeds Q&A confirms its path with Microsoft directly.
This guide recommends building the inventory before choosing what replaces anything. For each place Q&A appears, record its owner, who uses it and the questions they actually ask:
- Question boxes on dashboards.
- Q&A visuals placed in reports.
- Q&A used from the Power BI mobile apps.
- Q&A embedded in an internal or customer-facing application.
- Q&A tooling that report authors used to train Q&A.
The last field recorded, the questions users actually ask, is what the replacement is tested against.
What Copilot needs before a report user can ask it anything
The second failure arrives after the question boxes are gone: the workspaces users rely on cannot run Copilot at all. Microsoft’s Copilot for Power BI overview page sets out the requirements in a table, as of September 2026:
- Capacity: paid Fabric capacity at F2 or higher, or Power BI Premium at P1 or higher.
- Region: the Fabric capacity must be in a supported region.
- Sovereign clouds: listed as not supported.
The same Copilot for Power BI overview page states: “Trial capacities and free SKUs aren’t supported.” Microsoft’s What is Copilot in Fabric? page states: “The Azure OpenAI Service isn’t available on trial SKUs.”
Where report users are licensed on Pro or Premium Per User, the plan stalls here unless it names a capacity. Microsoft’s Fabric Copilot capacity page describes the route: “A Fabric Copilot capacity is a Fabric capacity that you designate to collect and bill the Copilot usage of a specific group of users.” The same page states: “The Fabric Copilot capacity must reside on at least an F2 or P1 SKU.” It also states: “Fabric Copilot capacity is supported only in the Fabric tenant’s home region.”
According to Microsoft’s Fabric Copilot capacity page, users assigned to a Fabric Copilot capacity can use Copilot in Power BI on workspaces in these license modes, and in Power BI Desktop:
- Pro.
- Trial.
- Premium Per User.
- Premium capacity.
- Fabric capacity.
The same page states: “Capacities with the Embedded license mode aren’t supported.”
Timing and load matter once capacity is chosen. The Copilot for Power BI overview page states: “When you purchase new capacity or scale up capacity, it can take up to 24 hours for Copilot to recognize the change and become available for use.” Microsoft’s What is Copilot in Fabric? page states: “Furthermore, Copilot in Power BI consumes your available Fabric capacity, so you should manage its usage to avoid overconsumption that can lead to throttling and disruption of your other Fabric operations.” Microsoft’s Consumption rates and billing for Copilot in Fabric page states: “Requests to Copilot in Fabric consume Fabric capacity units (CUs).” It adds: “Copilot operations in Fabric are classified as background jobs, which allows Fabric to handle a higher volume of Copilot requests during peak hours.”
Our recommendation: decide which capacity bills Copilot before the first user arrives, and watch the capacity it shares with production reports from the first week.
Copilot may already be on: the defaults and who can change them
The third failure is assuming Copilot is off because no one in IT turned it on. Microsoft’s What is Copilot in Fabric? page states: “Copilot is enabled by default for tenants with paid Fabric capacities (F2 or higher).” The Copilot for Power BI overview page states: “Copilot is enabled by default, but your admin might turn it off.”
A second default followed. Microsoft’s Enable Fabric Copilot for Power BI page states: “As of September 2025, this setting (and therefore the Standalone Copilot experience) is enabled by default for all tenants where Copilot is already turned on.”
Tenant administrators are not the only people who can switch it on. The same Enable Fabric Copilot for Power BI page states: “This AI and Copilot setting is automatically delegated to capacity administrators; tenant administrators can’t turn off the delegation.” It continues: “This means that even if Copilot isn’t enabled on a tenant level, a capacity administrator can choose to enable Copilot for their capacity.” And the scope is wider than reports: “If you enable Copilot in Fabric, users can access Copilot across workloads.” The page adds: “Copilot access isn’t limited to a specific workload, like Power BI.”
Microsoft’s Copilot and Agent tenant settings page lists these tenant settings among those enabled by default:
- Users can use Copilot and other features powered by Azure OpenAI.
- Capacities can be designated as Copilot in Fabric capacities.
Microsoft 365 Copilot is a different product with its own controls outside the Fabric admin portal; this guide covers only the Fabric settings.
What we recommend: read the current tenant and capacity settings before planning anything else, and write down which capacity administrators hold the delegation.
Where report data is processed: the settings that decide it
The privacy office’s first question is where report data goes when a user asks Copilot about it, and the answer depends on the capacity’s region and on settings, not on Copilot alone. Microsoft’s What is Copilot in Fabric? page states: “The Azure OpenAI Service that powers Fabric Copilot is currently deployed in US datacenters (East US, East US2, South Central US, and West US) and in the EU data boundary.” Microsoft’s Privacy, security, and responsible AI use of Copilot in Fabric page states: “Data processed by Copilot in Fabric stays within your capacity’s geographic region, unless you explicitly allow data to be processed outside your region”. The same page states: “Your data isn’t used to train models and isn’t available to other customers.” It also states: “Copilot doesn’t store your data for abuse monitoring.”
Settings that allow data out of the region are off by default. Microsoft’s Copilot and Agent tenant settings page lists these as disabled by default, and describes each as applying only to customers whose capacity’s geographic region is outside of the EU data boundary and the US:
- Data sent to Azure OpenAI can be processed outside your capacity’s geographic region, compliance boundary, or national cloud instance.
- Data sent to Azure OpenAI can be stored outside your capacity’s geographic region, compliance boundary, or national cloud instance.
- Data sent to OpenAI as a Microsoft Subprocessor can be processed outside your capacity’s geographic region, compliance boundary, or national cloud instance.
Microsoft’s Copilot for Power BI overview page states: “If your tenant or capacity is outside the United States or EU data boundary, Copilot is disabled by default unless your Fabric tenant admin enables the tenant setting called Data sent to Azure OpenAI can be processed outside your tenant’s geographic region, compliance boundary, or national cloud instance in the Fabric admin portal.” Microsoft’s Enable Fabric Copilot for Power BI page states: “The cross-geo setting stays disabled and doesn’t autodelegate to capacity administrators.”
For a US tenant, the setting that changes where data can go is a different one. The same Copilot and Agent tenant settings page lists “Users can use Copilot, AI Agents and other AI experiences powered by OpenAI as a Microsoft Subprocessor” as disabled by default. Microsoft’s OpenAI as a subprocessor in Microsoft Fabric page states: “Enabling this feature will allow data to egress outside of Fabric’s compliance boundary to a third-party system or infrastructure that may have lower compliance controls (for example, may not comply with FedRAMP or Supply Chain Risk Management).”
The same OpenAI as a subprocessor in Microsoft Fabric page states: “OpenAI models delivered through OpenAI as a subprocessor in Copilot experiences are currently excluded from country/region-specific processing commitments when applicable.” And it states where the decision sits: “The tenant administrator is responsible for ensuring features meet their organization’s security requirements and compliance obligations.”
Whether a given configuration is acceptable under FedRAMP, HIPAA, GLBA or a contract clause is a determination for the organization’s privacy, compliance and contracting owners; this guide describes Microsoft’s settings and caveats and makes no such determination. Where the question is evidence of who used Copilot on which items, that belongs to Microsoft Purview for Microsoft Fabric: The Audit Evidence Checklist.
Our recommendation for a US tenant: leave the cross-region settings off unless a capacity sits outside the US and the EU data boundary, and treat the OpenAI-subprocessor setting as a privacy-office decision, taken in writing with Microsoft’s caveat on the record.
Copilot answers only as well as the semantic model behind it
The fourth failure is a correct setting and a wrong answer: Copilot turned on over models no one prepared. Microsoft’s Copilot for Power BI overview page states: “When a question is related to data in the semantic model, Copilot uses the semantic model to answer the question.” It continues: “Otherwise, it answers from the large language model’s (LLM) general knowledge.” And it warns: “Without preparation, Copilot can misinterpret the data and return generic or inaccurate results.”
That same Copilot for Power BI overview page states: “Right now, three chat-with-your-data experiences exist for business users: the standalone Copilot experience, the Copilot pane for reports, and Copilot in apps (scoped to curated app content and supports verified answers by app authors).” Microsoft’s Copilot and Agent tenant settings page states: “Tenant admins can default Copilot search to be limited to items that are marked as approved for Copilot.” Microsoft’s Privacy, security, and responsible AI use of Copilot in Fabric page states: “Copilot responses can include inaccurate or low-quality content, so make sure to review outputs before you use them in your work.”
Model work does not have to wait for the capacity decision. A GCC High estate whose move from Premium P1 to Fabric is still open can certify and approve models now; the move itself is covered in Power BI Premium P1 to Microsoft Fabric in GCC High: A Readiness Guide.
We recommend approving for Copilot only the semantic models you would certify for reporting, testing each against the questions from the Q&A inventory, and starting users in Copilot in apps, where app authors can supply verified answers.
If your tenant is in a government cloud
The government-cloud reader faces a different failure: waiting for February 2027 to learn whether Copilot is offered at all. Microsoft’s What is Copilot in Fabric? page states: “Copilot isn’t yet supported for sovereign clouds due to GPU availability.” Microsoft’s Copilot for Power BI overview page lists sovereign clouds as not supported in its requirements table.
Microsoft’s OpenAI as a subprocessor in Microsoft Fabric page states: “Access to OpenAI-operated models isn’t currently available for use in government clouds (GCC, GCC High, DoD) or sovereign clouds.” That sentence names government clouds and sovereign clouds as two sets, while the Copilot requirement says only “sovereign clouds”. As read in September 2026, no Microsoft Learn page we read names the status of Copilot for Power BI in GCC, GCC High or DoD, and Microsoft’s Power BI for US government customers page does not mention Copilot. Microsoft 365 Copilot is a different product, and its position in a government cloud says nothing about Power BI.
A GCC, GCC High or DoD tenant therefore asks Microsoft for Copilot for Power BI’s status in its cloud, in writing, and plans as if the answer could be no. The fallback below is this guide’s planning, not Microsoft’s recommendation:
- Take the high-use questions from the Q&A inventory.
- Rebuild each as a curated visual, report page or bookmark on the certified semantic model.
- Retire the question box or Q&A visual once its replacement is in use.
- Set a date to re-check Microsoft’s status for your cloud.
Where Power BI itself should run, in the service or on a report server, is a separate hosting decision this guide does not cover.
Build the fallback now, for the questions users actually ask, so the retirement date changes nothing they depend on, whatever Microsoft says about your cloud.
When this guide is the wrong starting point
Some estates reach the retirement with a different decision in front of them, and this guide should not be forced onto it:
- Q&A use is negligible. If the inventory finds few users, retire the question boxes and visuals and defer any Copilot decision until there is demand for it.
- The estate has no paid capacity and no plan to buy any. Copilot does not run on Pro or Premium Per User licensing alone, so the decision is a capacity purchase, which this guide does not price.
- The semantic models are not certified. Turning Copilot on first produces the generic or inaccurate answers Microsoft’s Copilot for Power BI overview warns about, so model work comes before Copilot.
- Q&A is embedded in a customer-facing application. Microsoft’s embedded Q&A page does not state its path, so that owner starts with Microsoft, not with this guide.
- The real question is which cloud or server Power BI should run on, or the move from a P SKU to an F SKU. Those decisions come first and have their own guidance.
What to ask of whoever helps you move
Whoever does the work, internal team or outside help, hold the plan to evidence rather than assurances:
- Show the Q&A inventory, with each use’s owner, its users and its replacement.
- Show the current state of every Copilot and AI tenant and capacity setting, and who holds the capacity-administrator delegation.
- Show which capacity bills Copilot and how its consumption is watched alongside production reports.
- Show which semantic models are approved for Copilot, and why each one qualified.
- Show the privacy office’s decision on the OpenAI-subprocessor setting, in writing.
- For a government tenant, show Microsoft’s written status for your cloud and the fallback for each high-use question.
- Hand the configuration and the models to the internal team to run.
A plan that cannot show these is not ready, whoever wrote it. An i3solutions engagement does not produce managed-service ownership, a replacement for the internal team, open-ended scope expansion, or vendor lock-in.
How i3solutions answers
i3solutions implements Power BI as enterprise analytics infrastructure. That practice is described on the Enterprise Power BI Development & Integration Services for Defensible Analytics page. i3solutions has deployed Power BI inside a GCC High tenant and inside Azure Government for a federal customer. i3solutions has delivered Microsoft Fabric work for clients.
For the Q&A retirement, the work is delivered as a project, with specialists embedded in your team where it needs them: inventorying Q&A use, readying and approving the semantic models Copilot may answer from, configuring the Copilot tenant and capacity settings and the capacity that bills Copilot, and building the report-level fallback for the questions users depend on. All i3solutions Power BI and analytics developers, architects, and consultants are 100% U.S.-based. i3solutions has been a Microsoft Partner with 30 years of enterprise technology delivery experience and has delivered 1,500+ Microsoft engagements across aerospace and defense, financial services, and health sciences.
Key Takeaways
- Microsoft’s Learn pages say Power BI Q&A experiences are going away in February 2027 and recommend Copilot for Power BI.
- Inventory every question box, Q&A visual, mobile and embedded use before choosing a replacement for any of them.
- Copilot needs paid F2 or P1 capacity or better; Pro and Premium Per User users reach it through a Fabric Copilot capacity.
- Copilot is on by default where capacity is paid, and a capacity administrator can enable it even when the tenant has not.
- For a US capacity, the setting that changes where data can go is the OpenAI-subprocessor setting, off by default and carrying Microsoft’s own FedRAMP caveat.
- Approve for Copilot only the models you would certify, and review what it returns.
- A government-cloud tenant confirms its status with Microsoft in writing and builds a report-level fallback now.
Frequently Asked Questions
When is Power BI Q&A retired?
According to Microsoft Learn, Power BI Q&A experiences are going away in February 2027. Other dates circulated in earlier posts, but the Microsoft Learn Q&A pages as read in September 2026 give February 2027, and that is the date to plan against.
What replaces Power BI Q&A?
Microsoft recommends Copilot for Power BI. According to Microsoft Learn, Copilot needs paid Fabric capacity at F2 or higher or Power BI Premium at P1 or higher, and trial capacities and free SKUs are not supported. Where the inventory finds little Q&A use, retiring the question boxes and visuals may be enough.
Can Pro or Premium Per User users use Copilot in Power BI?
Yes, through a Fabric Copilot capacity. According to Microsoft Learn, a Fabric Copilot capacity on at least an F2 or P1 SKU, in the tenant’s home region, collects and bills the Copilot usage of the users assigned to it, and they can then use Copilot in Power BI on Pro and Premium Per User workspaces and in Power BI Desktop. Without paid capacity somewhere in the tenant, those users have no route to Copilot.
Is Copilot in Power BI turned on by default?
Where capacity is paid, yes, unless an admin has turned it off. According to Microsoft Learn, Copilot is enabled by default for tenants with paid Fabric capacities at F2 or higher, and as of September 2025 the standalone Copilot experience is enabled by default where Copilot is already turned on. Microsoft Learn also states that a capacity administrator can enable Copilot for a capacity even if it is not enabled at the tenant level.
Does Copilot in Power BI send our data outside the US?
For a capacity in the US, not by default. According to Microsoft Learn, the Azure OpenAI Service behind Fabric Copilot is deployed in US datacenters and in the EU data boundary, and data processed by Copilot stays within the capacity’s geographic region unless you explicitly allow otherwise. According to Microsoft Learn, the separate setting that lets Copilot use OpenAI as a Microsoft subprocessor is disabled by default, and enabling it allows data to leave Fabric’s compliance boundary for infrastructure that may have lower compliance controls and may not comply with FedRAMP.
Is Copilot for Power BI available in GCC or GCC High?
Microsoft Learn does not say by name. According to Microsoft Learn, Copilot isn’t yet supported for sovereign clouds due to GPU availability, and OpenAI-operated models are not currently available in government clouds (GCC, GCC High, DoD). As read in September 2026, no Microsoft Learn page we read names Copilot for Power BI’s status in GCC or GCC High, so a government tenant confirms its status with Microsoft in writing and builds a report-level fallback before February 2027.
Planning the Decision
If report users still type questions into Q&A, start with the inventory and a read of the current Copilot settings, then decide the capacity, the approved models, the privacy office’s position on the OpenAI-subprocessor setting and, in a government cloud, the fallback, well before the retirement date.
