DP-600 leads to the Microsoft Certified: Fabric Analytics Engineer Associate credential. The exam lasts 100 minutes, needs a score of 700 to pass and costs $165 in the US, with local pricing elsewhere. Like other Microsoft associate certifications it expires every year, but renewal is a free online assessment on Microsoft Learn, which is a policy I wish NVIDIA and Databricks would copy. The study guide splits the exam into three areas. Prepare data is the biggest at 45 to 50%, covering getting data in through the OneLake catalog and Real-Time hub, choosing between data stores, building star schemas in a lakehouse or warehouse, cleaning duplicates and nulls, and querying with the visual editor, SQL, KQL and DAX.
Implement and manage semantic models is 25 to 30%, which means storage modes, relationships including bridge tables and many-to-many, DAX iterators and window functions, calculation groups, field parameters, composite models, Direct Lake configuration and incremental refresh. Maintain a data analytics solution is the remaining 25 to 30%: workspace and item access, row, column, object and file-level security, sensitivity labels, Git integration, .pbip projects, deployment pipelines and the XMLA endpoint. A new version of the skills outline takes effect on 19 October 2026. Microsoft lists the changes as minor and limited to the query and analyze section, so anyone already studying does not need to start over. I like this exam a lot. Most data certifications ask you to recognise product names.
This one asks you to know when Direct Lake falls back to DirectQuery, how to structure a model so DAX stays fast, and how to put a semantic model under version control. Those are the things that separate a BI team that ships from one that rebuilds reports every quarter. The lifecycle material in particular is overdue. Plenty of Power BI developers have never used Git, and an exam that makes them learn it is doing the industry a favour. The gap that stood out to me is AI.
Microsoft's Fabric marketing leans hard on Copilot, data agents and natural language queries over semantic models. None of it is tested. You could argue that is sensible, since a clean star schema with good descriptions and sensible measures is exactly what makes those features work, and a messy model makes Copilot confidently wrong. I agree with that argument.
But it would be easy to add a small section on preparing models for AI, and its absence makes the exam feel a step behind the product. For preparation, the free Microsoft Learn path covers the scope, and the free practice assessment is worth doing early to find weak spots. The Fabric trial gives you an environment to build in. I would spend most of my time on semantic modelling and DAX, because that is where people with SQL backgrounds tend to lose marks, and where the questions are least forgiving of guesswork. Who should skip it: anyone whose company is on Snowflake, Databricks or Looker, and anyone wanting a credential that signals machine learning skill. For the latter, DP-100 or one of the AI-specific Microsoft exams is the better choice.
For everyone else working in the Microsoft data stack, this is the one I would take first.