Microsoft Certified: Azure AI Fundamentals
Azure AI Fundamentals confirms you can talk accurately about machine learning and generative systems as Microsoft delivers them: what the services do, where they fit, and what responsible use requires.
The certificate now runs on exam AI-901, which replaced the code most courses still name.
Exam facts
- Exam code
- AI-901
- Level
- Foundational
- Field
- Data & AI
- Duration
- 45 minutes
- Questions
- about 40 to 60 (Microsoft does not publish a fixed number)
- Passing score
- 700 of 1,000 (scaled)
- Languages
- English, German, French, Spanish, Italian, Portuguese, Japanese, Korean, Chinese, Arabic, Indonesian, RussianThe badges group regional variants: Chinese (Simplified and Traditional) count as one language here, so the vendor lists thirteen exam editions where this page shows twelve.
- Price
- 99 USD (read 8 September 2026)
- Validity
- never expires
- Exam delivery
- Pearson VUE (test centre or online proctored); students through Certiport
Prerequisites: None required. Microsoft expects conceptual knowledge of AI solutions on Azure, basic Python syntax and some familiarity with Azure resources.
Source: exam page at Microsoft · Price: Priced by the country where the exam is proctored: 99 USD in the United States, 76 EUR in Germany and Austria (Microsoft Learn country selector, read 2026-09-08)
How to prepare
1. The vendor's free learning path
Microsoft publishes the exam objectives and a learning path free of charge — the authoritative source for scope and weighting. Open the learning path (opens in a new tab)
2. Online courses
Courses and practice questions from the directory that target exactly this exam — details, price and the provider link (affiliate) are on the course page.
Book the exam
Delivered by Pearson VUE (test centre or online proctored); students through Certiport. Schedule with Microsoft (opens in a new tab)
Affiliate disclosure: the course links above is affiliate links — buying through them may earn us a commission at no extra cost to you. The vendor's learning path and booking links carry no commission.
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.
- No practice test is available through our partner for the current code, only for the retired one, which we would rather say than link.
- 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
| Half | Share | What it expects you to recognise |
|---|---|---|
| Generative AI on Azure | 55 to 60 % | What a foundation model is, what a prompt does, where the Azure OpenAI services sit, and what responsible use requires of the person deploying them |
| Machine learning and the classic AI workloads | 40 to 45 % | What training is, which kind of problem a model solves, and the vision, language and document services that existed before the generative wave |
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.
Check the code on everything you buy
AI-900 material is still on sale, still ranks in search, and still targets a syllabus with the emphasis inverted. The only practice product our partner sells targets the retired exam, so the record links nothing rather than something misleading. Microsoft's own learning path is current and free, and for a forty-five-minute paper with a scaled 700 to pass, it and the study guide are enough.
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.
What a question looks like
Written by us in the exam's style. It is not a real question from any question bank, and we do not publish those.
A team wants a system that answers customer questions in natural language using the company's own product documentation, which is updated regularly. Which Azure capability best matches the requirement, and what characteristic of the documentation makes that the right choice?
The paper habitually asks for the service and the reasoning together, so half-knowing the catalogue is not enough. Options that involve training a model on the documents look thorough and fall over on the word regularly, which is a conceptual point about where knowledge should live rather than a product detail.
What it costs to get and to keep
| Item | Amount | Note |
|---|---|---|
| Exam fee | 99 USD | the United States price; Microsoft sets fees by the country where the exam is proctored (read 12 September 2026) |
| Exam fee, Germany and Austria | 76 EUR | from the Microsoft Learn country selector (read 8 September 2026) |
| Practice test | not published | our practice-test partner lists only the retired AI-900 product, so there is nothing current we can point at for this code (read 12 September 2026) |
| Renewal | not published | none; fundamentals certificates do not expire (read 12 September 2026) |
How much preparation, from where you are
- You commission or govern AI work
- The intended reader, and the responsible-use material is the part that will change how you review a proposal. Microsoft's free learning path covers the syllabus adequately.
- You build with these services already
- Short, and largely about matching product names to things you do. Whether it is worth a fee depends entirely on whether somebody is counting certificates.
What passing this does not prove
- Any ability to train, tune or evaluate a model.
- Whether you can call these services from code.
- Whether a system you specified would behave acceptably on real data.
- Anything outside Azure, beyond the concepts it shares with everyone else.
Against the alternatives
- AIF-C01 — AWS Certified AI Practitioner
- The AWS counterpart at the same tier and ambition. Pick by the platform your organisation runs; holding both signals less than people hope.
- DP-900 — Microsoft Certified: Azure Data Fundamentals
- The data fundamentals paper, and arguably the more useful of the two for most non-technical readers, because every AI project fails on data long before it fails on models.
- PL-300 — Microsoft Certified: Power BI Data Analyst Associate
- A genuine associate-level step into analytics work rather than vocabulary. If you want something an employer reads as capability, go there instead.
AI-901 — quick answers
Is this AI-900?
Not any more. The certificate is unchanged but the exam behind it is now AI-901, and AI-900 has been replaced. Check the code on anything you buy as preparation, because a great deal of published material and several practice products still target the old one.
Do I need any maths?
No, and that is the clearest line separating this tier from the associate data papers. What is required is knowing what a model does, where it goes wrong, what it costs and who is accountable for it, none of which requires statistics.
Does it expire?
No. Fundamentals certificates are permanent, which removes any renewal obligation and also removes any evidence of currency. In a field moving this quickly, write the year beside it on your CV.
Why does generative AI carry the larger share?
Because that is what Microsoft's customers are buying and asking about, and the revision that produced the new code followed the market. The classic machine-learning half is still examined and still stable; the generative half is where the vocabulary is newest and where old material misleads most.
AWS AI Practitioner or this one?
Whichever platform is in the room. Both certify the same literacy at the same level. This one is shorter, never expires and is offered in more languages including German; the AWS paper runs longer, lasts three years and is withdrawing its German edition. Holding both says little more than holding one.
What comes next
Concepts this exam draws on
Glossary entries with the reason each one matters for AI-901.
- ML
Fundamental ML principles and Azure Machine Learning at concept level.
- GenAI
Generative AI and Azure OpenAI are a named domain of AI-901.
- LLMs
Language models and Azure AI Language services.
- Responsible AI
Microsoft's responsible AI principles are explicitly tested.
Last reviewed 12 September 2026 · Getting Digital