Power BI Retires R and Python Visuals in May 2027: How to Find and Replace Them

Power BI's R and Python visuals have a retirement date​

Microsoft's Learn documentation is clear about the change. R and Python visuals in Power BI, including R-powered custom visuals, retire in May 2027. After retirement, R and Python visuals stop working in published reports in the Power BI service, including reports published before retirement. Reports that already work will break too, not just new ones.

What changes and what doesn't​

The scope is narrower than the headline suggests. This retirement applies only to R and Python visuals in interactive Power BI reports. Python scripts that you use in Power Query to populate tables aren't affected.

Some teams will be tempted to stay on an old Desktop build. That only keeps local rendering alive. Your colleagues open reports in the service, and the service is where these visuals will stop working. Microsoft's documentation doesn't give an exact day in May.

This has happened before​

The 2027 date follows an earlier, smaller cut. According to a Microsoft 365 message center notice (MC1283817) reported by PupuWeb, starting May 1, 2026, Power BI will no longer support embedding reports and dashboards that contain R or Python visuals in Embed for your customers (also known as app owns data) and Publish to web scenarios. After that date, embedded Power BI reports or dashboards that contain R or Python visuals will continue to load, but those visuals will appear blank.

Some people missed the signal. In a Fabric Community thread from late 2025, one reply argued that Microsoft had announced no retirement for Python or R visuals. The same thread was later updated with a post saying they are in fact now retiring in May 2027. If your organization took that earlier reassurance at face value, now is the time to look again.

How to find affected reports​

Microsoft gives a concrete way to search. The Fabric blog announcement says to look for the visualType values pythonVisual for Python visuals and scriptVisual for R visuals. The Learn overview adds that if you use Power BI Project (PBIP) format and Git-integrated workspaces, search your report definitions in source control to find built-in R and Python visuals across multiple reports. If your reports are under version control, a single repo-wide search for those two strings will give you a first inventory.

A practical checklist for report owners:

1. Inventory. Search PBIP report definitions for pythonVisual and scriptVisual. Include any R-powered custom visuals from AppSource.
2. Preserve. Save each visual's script along with its package and version requirements before anything gets rebuilt.
3. Pick a replacement. Microsoft's Python visuals documentation suggests a built-in Power BI visual, a custom visual, or an analysis in a Fabric notebook. The Fabric blog adds that AI coding tools can also help you create a visual with the Power BI custom visuals software development kit (SDK).
4. Watch the roadmap. The Power BI roadmap includes planned HTML and Vega core visuals for more flexible report visualizations. That could be a good landing spot for people who use Deneb-style charts.
5. Test with consumers. There is no automatic migration, so each replacement has to be rebuilt by hand and checked against the original.

For context, Microsoft's Python visual documentation already lists real limits on these visuals. Input is capped at 150,000 rows and 250 MB, calculations time out after five minutes, output renders at 72 DPI, and the image itself isn't interactive. Microsoft hasn't given these limits as its reason for retiring the visuals. They do help explain why a native or SDK-built replacement may end up better than the original.

Section summary: Power BI reports using R or Python visuals will break in the service in May 2027. Power Query scripts are safe. Search for pythonVisual and scriptVisual now.

Databricks makes Genie Ontology generally available​

Databricks' November release notes list Genie Ontology as generally available on November 9, 2026. Databricks calls it a context layer that combines knowledge defined in Unity Catalog semantics with knowledge "learned" from dashboards, SQL queries, notebooks and usage. A ranking system called OntoRank weighs authority, certification, relevance and recency when it picks context, and it respects user permissions. Genie MCP makes that context available to outside agents.

Several related pieces also reached general availability:

• Genie Search indexes sources such as Google Drive, Jira, Confluence and Workday, and indexes each document's permissions with it.
• Ontology Snippets are units of learned knowledge that stay tied to their source asset and inherit its permissions.
• Organizational directories, Unity Catalog Domains and Pages are also GA.

Databricks also reports a benchmark built with Turing in which Genie passed 91% of enterprise task runs, 1.6 times the rate of the best frontier agent, using less than half the tokens. Databricks ran the test on a simulated workspace it designed, so treat that figure as a reason to run your own evaluation, not as a result you can expect.

Two details in the same release notes matter for Microsoft shops. Databricks lists a Beta feature that pulls metadata from Power BI semantic models so Genie can use Power BI measure definitions. It also lists a Genie One desktop app for Windows, now in Beta.

Edelbyte comes out of stealth with Kusto pedigree​

Edelbyte launched on November 11 with $70 million raised over two rounds led by Sequoia and Quiet Capital, according to SiliconANGLE. Greenoaks and Swish Ventures also took part. The founders are former Microsoft executives who built Azure Data Explorer, the engine known as Kusto. Kusto users will recognize the pitch: fast queries over huge, constantly changing datasets.

Edelbyte says its platform handles records, vectors, search, analytics and graph processing, with deployments sized for hundreds of petabytes. It also claims queries under 100 milliseconds on multi-petabyte data. Query federation lets it run queries where the data already lives, and automated index tuning keeps up as access patterns change. These are company claims, and nobody has published a benchmark that can be reproduced.

Alteryx puts governed workflows behind Gemini for Healthcare​

On November 10, Alteryx announced that its MCP data connector now works with Google Cloud's Gemini for Healthcare. The idea is that the model doesn't do the math itself. In Alteryx's example, a finance team asks Gemini which claims need reconciliation. Gemini then runs an approved Alteryx workflow that applies the reimbursement rules and returns the exceptions for a person to review.

Alteryx points to its SOC 2 Type II attestation and says it has "safeguards designed to support HIPAA compliance." Being designed to support compliance is not the same as being compliant automatically. Each deployment still needs its own review.

AWS open-sources a lake-to-graph interface​

AWS released an open-source web UI that runs locally and turns selected data-lake records into graphs in Amazon Neptune Analytics. You pick an Athena catalog and database, define the graph with SQL/PGQ, and preview rows. The tool then runs the projection queries through Athena, stages the results in S3 and imports them. Data does still move here: you're copying a slice of the lake into a graph, not querying the whole lake in place.

PuppyGraph and Qlik: graphs in place and AI for pipeline work​

• PuppyGraph raised a $10 million Series A led by Particle Future and defy.vc, which brings its total funding to $15 million. Its engine queries relationships across existing data platforms without a separate graph database. An AI-assisted schema builder proposes relationships for a person to approve, and MCP support connects outside agents.
• Qlik put the Talend Studio AI Assistant into public preview. It's built on Qlik Answers and governed through Qlik Talend Cloud. It drafts jobs from plain-language requests, helps with mapping and code, writes documentation, and summarizes execution logs to point at likely errors. Developers still decide what ships. Qlik says the preview is available from the R2026-09 patch for Premium and Enterprise customers, with general availability targeted for Q1 2027.

Hood College trains the people behind the dashboards​

Hood College's Bioscience Research and Education Center and Frederick County Workforce Services will run two AI-enhanced data analytics cohorts, one in spring 2027 and one in fall 2027. Each cohort takes 10 to 15 participants. The course starts with data and AI literacy and works up to predictive modeling. It is aimed at unemployed and underemployed people and at workers who want to advance. The $34,000 behind it is entirely federal money from the Department of Labor. The program is small, but it's a reminder that someone still has to understand what the agents are producing.

The bigger picture​

Almost every vendor this week is selling the same thing: governed, permission-aware context that keeps AI agents from making things up. The approaches differ. Databricks learns it from your existing work, Alteryx encodes it in visual workflows, and PuppyGraph and AWS map it as relationships. MCP shows up again and again as the connector between agents and data.

Microsoft's Power BI change fits the same direction. The company is moving away from arbitrary scripts that render as static images in reports, and toward native, SDK-built and Fabric-based options that are easier to govern. You can argue that this hurts the data scientists who relied on ggplot2 and matplotlib inside reports, and that argument is reasonable. Even so, the date is set, so start your inventory now.

Bottom line: if you run Power BI, search your reports for R and Python visuals before you spend time on agent-ontology plans. You have until May 2027.