Power BI Streaming Datasets Retirement: What to Use Instead

Navy Solv Systems title card reading 'Power BI Streaming Datasets Are Retiring' with a fading signal waves motif.

In Short: Power BI Streaming Datasets Have a Deadline

Microsoft announced on 15 October 2024 that it is retiring real-time streaming in Power BI. That covers every Power BI streaming dataset pattern: push semantic models, streaming semantic models, PubNub streaming semantic models and the streaming data tiles on dashboards. The date in the announcement is 31 October 2027 - from then, creating new real-time semantic models is no longer supported and existing ones are retired.

As of today, 28 September 2026, that is roughly 13 months away. Microsoft's replacement is Real-Time Intelligence in Microsoft Fabric, which needs Fabric capacity. If you have anything writing rows into Power BI today, now is the time to find it and plan the move.

What Is Changing, and When

The original announcement planned to block creation from October 2024. After customer feedback Microsoft kept creation enabled during a sunsetting period, and the current timeline is:

From 31 October 2024: the sunsetting period began. You can still create and use real-time semantic models.
From 31 October 2027: creation of new push, streaming and PubNub models and streaming data tiles is no longer supported, and existing real-time semantic models are retired.

Two caveats. Microsoft says the 2027 date is "subject to change" while it works with customers on migration. And the Learn page for real-time streaming currently stresses the creation cut-off and says existing models are "unaffected", which reads differently from the blog. The Azure Stream Analytics docs are blunter: from 31 October 2027 you cannot create jobs with the Power BI output, and existing jobs using it are stopped. Our advice is to treat 31 October 2027 as the hard stop, and check the current position before you lean on any extension.

What Is Affected (Including Push Datasets)

The question we are asked most is whether push datasets fed by the REST API are in scope. They are. Push semantic models are named explicitly, and they are what most custom integrations create. In practice the retirement touches:

Push semantic models created with the PostDataset REST API and filled with PostRows, including the "pushStreaming" hybrid
Streaming semantic models created in the service UI with an API endpoint, including ones with historic data analysis switched on (which makes them push models too)
PubNub streaming semantic models, which read a PubNub channel straight into dashboard tiles
Azure Stream Analytics jobs using the Power BI output, which creates push-and-streaming models under the hood
Custom streaming data tiles pinned to dashboards

Ordinary Import, DirectQuery and Direct Lake semantic models are not affected. Nor is automatic page refresh, which is part of the answer.

A note on Power BI Desktop

Searches for "Power BI Desktop streaming dataset" usually come from people trying to build one in Desktop. You cannot: these models are created in the service or via the API. Desktop can connect live to an existing push model, but a streaming-only model cannot back a report at all. For anything new, build in Desktop against a DirectQuery source that supports automatic page refresh. Our guide to Direct Lake, Import and DirectQuery covers the trade-offs.

How to Find What You Have

Most organisations do not know how many streaming datasets they own. They were often set up by one developer, years ago, behind one dashboard. Build the inventory before you plan anything:

Scan the tenant. A Fabric administrator can call the Admin GetDatasetsAsAdmin API and filter on addRowsAPIEnabled set to true. The content provider types RealTimeInPushMode, RealTimeInStreamingMode and RealTimeInPubNubMode separate the three kinds.
Trace the writers. For each model, find what feeds it: a Stream Analytics job, an Azure Function, an IoT script, a Power Automate flow or a third-party tool. Some vendors, Alteryx among them, have already published notes about their Power BI output tools.
Trace the readers. List the dashboards, tiles and reports on top of each model, and whether anyone still opens them. Workspace usage metrics help here, and a workspace register that records owners makes this far quicker.
Classify the need. For each one, note the real latency required. Many "real-time" dashboards are viewed twice a day and would be fine on a 15-minute refresh.

That last step usually shrinks the job. Some models can be retired, some become ordinary scheduled models, and only the truly live ones need Real-Time Intelligence.

Migration Options: Old Pattern to New

Microsoft's migration guidance points to Fabric Real-Time Intelligence. The mapping we use:

REST API push into a Power BI model becomes an application sending events to an Eventstream custom endpoint, landing in an Eventhouse (KQL database).
Stream Analytics with the Power BI output keeps the Stream Analytics logic and switches its output to an Event Hub pointed at the Eventstream custom endpoint - or retires the job and moves processing into Eventstream.
Streaming tiles on a dashboard become a Real-Time Dashboard built from KQL queries, or a Power BI report in DirectQuery against the Eventhouse with automatic page refresh.
Data alerts on dashboard tiles become Activator rules, which can email, post to Teams, run a pipeline or trigger Power Automate.
PubNub feeds need something to bridge the channel into Eventstream or Event Hubs before the pattern above applies.

The Eventhouse is the heart of this. Unlike a streaming model's one-hour cache, it keeps history you can query properly - our guide to Eventhouse and KQL explains why it suits telemetry-scale data. For alerting and anomaly detection on top, see real-time AI with Fabric Activator.

One detail on refresh rates: on shared (Pro) capacity, automatic page refresh bottoms out at 30 minutes. On Fabric capacity, DirectQuery reports can refresh as often as every second, within the minimum your capacity admin sets.

Capacity and Licensing Implications

This is where budgets get surprised. Streaming datasets worked on Power BI Pro. Eventstreams and Eventhouses are non-Power BI Fabric items, and they need a workspace on an F capacity (or a time-limited Fabric trial). Pro and Premium Per User licences do not provision Fabric capacity.

So the migration may be your organisation's first Fabric capacity purchase. Check current F SKU prices in our Microsoft Fabric pricing guide, and size for continuous ingestion rather than bursty report queries - an always-on Eventhouse consumes capacity around the clock. Our Fabric capacity sizing guide covers how to test before you commit. Remember that below F64, report viewers still need Pro licences.

A Migration Plan That Beats the Deadline

Now to end of 2026: inventory, classify, and switch off what nobody uses. Stop building new streaming datasets, even though creation still works.
Early 2027: stand up a Fabric capacity and pilot one high-value stream end to end - Eventstream, Eventhouse, dashboard, Activator rule.
Spring to summer 2027: migrate the remaining streams in waves, running old and new side by side until the numbers match.
By September 2027: decommission the old models and Stream Analytics outputs, leaving a margin before 31 October.

Done well, you end up with more than a like-for-like swap: history you can query, alerts that act, and live data that stops decisions running on stale numbers. Teams using AI agents can go further with the Fabric RTI MCP server.

Where Solv Systems Comes In

The work breaks into four parts: a tenant inventory, a classification of each stream, a capacity estimate, and the migration itself. If your real-time estate grew up one dashboard at a time, our Power BI consulting team can map it and move it to Fabric before the deadline arrives.

Sources and Further Reading

Frequently asked

Microsoft's retirement announcement sets 31 October 2027: from that date creation of new real-time semantic models is no longer supported and existing ones are retired. Microsoft has said the date is subject to change and the Learn pages word the impact on existing models differently, so check the current position - but plan as if 31 October 2027 is the end.

Yes. Push semantic models are explicitly in scope alongside streaming semantic models, PubNub streaming semantic models and streaming data tiles. Anything that creates or writes to a push model through the Power BI REST API - including Azure Stream Analytics jobs using the Power BI output - is part of the retirement.

No - streaming and push models are created in the Power BI service or via the REST API, not in Desktop. Desktop can connect live to an existing push model, but streaming-only models cannot be used in reports at all. For new real-time reports in Desktop, use DirectQuery to an Eventhouse with automatic page refresh.

Real-Time Intelligence in Microsoft Fabric: Eventstream to ingest and transform events, Eventhouse (KQL database) to store and query them, Real-Time Dashboards or Power BI DirectQuery reports with automatic page refresh to visualise them, and Activator for alerts and actions.

Yes. Eventstreams and Eventhouses are non-Power BI Fabric items, which need a workspace on an F capacity (or a Fabric trial). Pro or Premium Per User licences alone do not provision a Fabric capacity.

A Fabric administrator can call the Admin GetDatasetsAsAdmin API and look for addRowsAPIEnabled set to true, and for the content provider types RealTimeInPushMode, RealTimeInStreamingMode and RealTimeInPubNubMode. Then trace each one back to the process that writes to it.