Microsoft Certified: Azure AI Fundamentals

The code change matters more than it sounds. Search results, course titles and half the advice online still say AI-900, and that exam has been replaced. Anything sold as preparation for the old code should be checked against the current study guide before you spend on it, and a practice test written for the retired version is worth less than it appears. The certificate name stayed the same, which is why the substitution is easy to miss.

• The weighting is lopsided. Generative AI now carries the larger share, with the general machine-learning material taking the rest.
• Responsible AI runs throughout rather than sitting in its own corner, an accurate mirror of how such projects get reviewed in practice.
• No coding appears. Nothing asks you to call an API or read a model definition.
• Practice tests on sale still target the retired AI-900 code; the version matters more than the rating.
• It never expires, so there is no renewal to plan for with nothing to distinguish a fresh pass from an ancient one unless the year is written beside it.

Who should sit it: people who talk about AI systems without building them, and who keep finding that a meeting has moved on while they worked out what somebody meant. For that reader it is forty-five minutes well spent. Who should not: developers and data practitioners, for whom this is a naming exercise over familiar ground, and anyone hoping it will substitute for evidence of having built something.

Two halves, unevenly cut

The proportions are the story of the exam's revision. The retired code was a machine-learning vocabulary paper with a generative section bolted on; the current one is the reverse, and a candidate who prepared on old material has the halves the wrong way round. The classic half is stable and the generative half is not, which is why the study guide is the only reliable syllabus and why anything dated before the code change should be read with the guide open beside it.

Where it sits, and which fundamentals paper first

Against its AWS counterpart the shape differs in every column: 45 minutes here against 90 there, a certificate that never expires against one that runs three years, twelve exam languages with German among them against a list that is shedding German and Italian, and a country-dependent fee against one global figure. Against Microsoft's own data fundamentals paper the choice is about order rather than merit: data first, for most readers, because every AI project fails on its data before it fails on its model, and the vocabulary here stands on the vocabulary there. Microsoft expects conceptual knowledge of AI on Azure, basic Python syntax and some familiarity with Azure resources, and requires nothing; delivery is by Pearson VUE, and Microsoft states the question count only as a band.