Cognitive services and Azure ML for Dataflows will be fully retired by September 15th, 2025

Beginning August 11th, Power BI users will no longer be able to create new models via cognitive services. Existing models will continue to refresh until September 15th, at which point they will no longer be supported, and Azure ML and Cognitive services will be fully retired.

Leveraging machine learnings in Power BI through cognitive services and Azure Machine learnings models is a useful feature in Power BI that enables customers to easily create and deploy machine learning models for predictive analytics. However, we have received feedback that customers want more control and flexibility over the end-to-end AutoML pipeline. The AutoML solution based on Synapse Data Science in Microsoft Fabric will help address this by offering customers a powerful and more customizable machine learning experience, with access to advanced tools and features.

With the new AutoML feature in Fabric, you can automate your machine learning workflow and get the best results with less effort. AutoML in Fabric, or Automated Machine Learning, is a set of techniques and tools that can automatically train and optimize machine learning models for any given data and task type. You don’t need to worry about choosing the right model and hyperparameters, as AutoML will do that for you.

Additionally, you are able to track and examine your AutoML runs using Fabric’s MLFlow integration and use the built-in visualization modules to generate interactive plots of your outcomes. Fabric also supports a large number of Spark and single-node model learners, ensuring that you can find the best fit for your machine learning problem.

Machine learning in Fabric is part of the Data Science experience within Fabric’s unified platform, typically powered by Synapse Data Science and integrated with Azure Machine Learning.

Key Capabilities

1. No-Code to Pro-Code Modeling – Fabric supports both drag-and-drop AutoML for business analysts and full-code notebooks (Python, Spark, etc.) for data scientists.
2. AutoML Integration – Users can train models automatically with minimal input. AutoML chooses the best algorithm, handles preprocessing, and tunes hyperparameters.
3. Notebook Support (Spark Runtime) – You can write and run machine learning code in Fabric notebooks, which are based on Apache Spark and support languages like Python.
4. Model Training and Evaluation – You can train models within Fabric using ML frameworks like scikit-learn, XGBoost, PyTorch, etc., and track experiments.
5. MLflow Integration – Fabric supports MLflow for tracking experiments, logging models, and managing lifecycle transitions (e.g., staging → production).
6. Seamless Deployment – Trained models can be deployed as real-time endpoints or batch scoring pipelines, integrated with Power BI or other Fabric workloads.
7. Security & Governance
   Built-in governance, access controls, and lineage tracking ensure models are secure and auditable across the data estate.

Use Cases

• Predictive analytics embedded into Power BI dashboards
• Churn prediction, customer segmentation, or demand forecasting
• End-to-end ML pipeline development using a unified UI

The Fabric solution will unlock even more customization and powerful ML capabilities that bring in other benefits of having your ML live in Fabric. We encourage customers to migrate their solution to Fabric as the current ML features in Power BI will be discontinued soon.

Please note

Beginning August 11th, customers will not be able to create new models via cognitive services, the UI entry point will be removed.  Your existing models that leverage AI services via dataflows will continue to refresh until September 15^th, at which point there will no longer be any refresh support for those dataflows that leverage AI services.

Next steps

To learn more about the new solution, refer to the Use Azure AI services in Fabric) documentation.

This post was originally published on this site