Azure Mastery
Download
Guide
A practical route through Microsoft's AI Fundamentals exam: what changed since AI-900 retired, the five skills domains and their weights, a realistic three-to-four week study plan, the topics that trip people up, and what to study next.
The exam
If you started studying for AI-900 and are seeing "AI-901" everywhere now, here is what changed. Microsoft retired AI-900 on 30 June 2026, and AI-901 is its replacement — the exam candidates book going forward, not a second credential stacked on top of AI-900. It covers the same skills outline with a refreshed question pool. If you already passed AI-900, that certification stands and you do not need to sit AI-901 too. If you have not taken either yet, AI-901 is the one to prepare for.
AI-901 earns the Microsoft Certified: AI Fundamentals credential. It is pitched at both technical and non-technical candidates: no data science background or programming experience required, just basic familiarity with cloud concepts and client-server applications. That makes it a sensible starting point whether you are a project manager who needs to talk sensibly about an AI initiative or a developer about to specialise into AI-103, the AI App and Agent Developer Associate exam.
Don't expect to write Python or design a model architecture. AI-901 stays conceptual: recognise AI workload types, describe machine learning concepts in plain language, know Microsoft's responsible-AI principles, and match a scenario to the right Azure AI service. Anchor your revision to the current outline rather than an old AI-900 course referencing retired service names.
Format & domains
AI-901 is entirely multiple choice and multi-select. There is no drag-and-drop ordering, no hotspot screenshots, and no multi-question case study to manage, which makes the format itself the easy part of this particular exam.
The exam fee changes from country to country, so check Microsoft's certification page for the price where you live. The five skills domains and their weights, from the current outline, are below.
| Domain | Weight |
|---|---|
| Describe AI workloads and considerations | 15–20% |
| Describe fundamental principles of machine learning on Azure | 15–20% |
| Describe features of computer vision workloads on Azure | 15–20% |
| Describe features of NLP workloads on Azure | 15–20% |
| Describe features of generative AI workloads on Azure | 20–25% |
Generative AI is the single largest domain and the one Microsoft has expanded most in recent refreshes, so give it priority even though the other four are formally tied at 15–20% each. Computer vision and NLP reward the same underlying skill: matching a scenario to the exact Azure AI service that handles it.
Study plan
AI-901 needs considerably less time than an associate-level exam like AZ-104, since there is no hands-on configuration to practise — just concepts and vocabulary to pin down precisely. Four weeks at a gentle pace, or three if you already touch AI or ML tooling at work, is realistic for most candidates.
Identify the four common workload types — computer vision, natural language processing, document processing, and generative AI — from a scenario, then work through the six responsible-AI principles: fairness, reliability and safety, privacy and security, inclusiveness, transparency, and accountability. Questions often ask which single principle a situation violates.
Learn the difference between regression, classification, and clustering, and between features and labels and training versus validation data. Add a basic sense of the Transformer architecture, then Azure Machine Learning's capabilities: automated ML, the data and compute services, and model management and deployment.
Cover image classification versus object detection versus OCR versus facial detection, and which of Azure AI Vision or Azure AI Face handles each. Then the NLP taxonomy — key phrase extraction, entity recognition, sentiment analysis, speech recognition and synthesis — mapped to Azure AI Language for text and Azure AI Speech for spoken input and output.
Finish with the largest domain: generative AI model scenarios such as chat, code generation, and summarisation, the services involved — Microsoft Foundry, Azure OpenAI Service — and the responsible-AI considerations specific to generative AI. Then switch to sharpening: drill your weakest domain daily and sit one or two full-length simulator runs before you book.
Where people lose marks
AI-901 doesn't punish you for lacking hands-on experience — it punishes imprecise vocabulary. A handful of areas consistently cost candidates marks because two or three closely related terms blur together under time pressure.
Fairness, reliability and safety, privacy and security, inclusiveness, transparency, and accountability sound similar in the abstract, and questions test whether you can name the one principle a scenario is really about — a model that performs worse for one demographic group is a fairness problem, not a generic "ethics" issue. Learn each principle with a concrete example, not just its name.
These three get confused constantly, particularly classification versus clustering: classification predicts a known label from labelled data, while clustering groups unlabelled data by similarity with no predefined categories. Regression predicts a continuous number rather than a category. Practise labelling scenarios with the right one of the three.
Image classification, object detection, optical character recognition, and facial detection are four distinct tasks, and Azure splits them across Azure AI Vision and Azure AI Face. Questions describe a use case — identifying products in a shelf photo versus extracting a tax reference from a scanned form — and expect the specific task and service, without hesitation.
Key phrase extraction, entity recognition, sentiment analysis, and translation route through Azure AI Language, while speech recognition and synthesis route through Azure AI Speech. The trap is a scenario mentioning both text and spoken input in one sentence — read for which stage of the pipeline the question is actually asking about.
The largest domain is also the newest, and it repeats the responsible-AI theme with generative-AI-specific twists: hallucination, content filtering, and grounding responses in your own data don't apply the same way to a classic classification model. Do not assume week one's principles cover everything generative AI raises.
How to practise
Watching videos leaves you able to recognise a term, not recall it, and AI-901 asks for precise vocabulary against the clock. Two habits close that gap. Begin with retrieval practice: answer from memory before the explanation loads, since the effort of dragging the right term back is what fixes it. Then layer on spaced repetition: return to each domain just as it starts to fade rather than covering it once and never coming back — the responsible-AI principles you nailed in week one are exactly the ones you'll blur with generative AI's considerations by week four if you never revisit them.
Third, full-length simulation. A timed, mixed-domain run at the exam's real 45-minute pace is the only way to find out whether you can sustain quick recall across 40 to 60 questions back to back. Run at least one before you book, and review every wrong answer rather than just noting the score.
Azure Mastery is built around exactly this loop for AI-901. Practise with over 380 exam-style questions mapped to the five current domains, get an on-device readiness score from the Exam IQ engine, follow an adaptive study plan that targets your weak spots, and let knowledge-decay alerts tell you what to revise before you forget it. Come exam week, the Practice tab gives you a timed run at the exam's real length and question count, all of it available offline and without creating an account.
Download Azure Mastery — freeExam day
If you are sitting online, run the system check the day before and clear your desk down to the surface, because the proctor will have you pan the camera around the room and put away anything within arm's reach. Arrive early whichever way you sit; a rushed start eats into the concentration precise recall needs.
With 45 minutes for 40 to 60 questions, you have well under a minute and a half per question, so keep moving rather than dwelling on any one of them. Use the mark-for-review flag on anything that makes you hesitate between two related terms, and come back to it once you have banked the easier marks. Unlike associate-level exams, AI-901 has no multi-question case studies, so every question stands alone and can be revisited freely before you submit.
Read each option carefully — the wrong answers are usually real Azure AI services that are simply the wrong fit, so precision matters more than quick elimination. The score is scaled: you need 700 out of 1000, not every question right, so an unfamiliar scenario is not a disaster. Flag it, make your best call, and keep your pace.
Keep going
AI-901 is a launchpad, not a destination — it gives you the vocabulary the next exams assume. If your goal is building generative AI applications and agents, AI-103 (AI App and Agent Developer Associate) is the natural next step: hands-on and SDK-aware, with Python or .NET work against Azure OpenAI and Azure AI Foundry, and a largest domain covering Foundry agents, tool calling, and retrieval-augmented generation with Azure AI Search.
There is no longer a fork in the road here: AI-103 replaced the retired AI-102 (AI Engineer Associate) on 30 June 2026, so it is now the single Associate-level credential for AI developers rather than one of two. Budget two to three months after AI-901; the jump from recognising terms to writing working code against the SDKs is a real one.
Not sure AI is the right track for you, or wondering how it compares with Azure's infrastructure and security paths? Which Microsoft certification should you get first? walks through the Fundamentals, Associate, and Expert tiers across every Microsoft track.
Frequently asked
They cover the same skills outline, but Microsoft retired AI-900 on 30 June 2026 and AI-901 is its replacement — the exam you book going forward. If you already passed AI-900, that certification still stands and you do not need to sit AI-901 as well. If you have not yet taken either, AI-901 is the current exam.
AI-901 is a Fundamentals-level exam, so it stays conceptual rather than hands-on. There is no code to write and no model to configure. The difficulty comes from precision: you need to recognise which AI workload type a scenario describes, which of the six responsible-AI principles it touches, and which specific Azure AI service fits — computer vision questions in particular expect you to distinguish classification, detection, OCR, and facial detection cleanly. Most candidates with some general IT or cloud background pass after three to four weeks of focused study.
Three to four weeks of steady study is realistic for most candidates, and it can compress to two if you already work with machine learning or generative AI tools. If AI concepts are genuinely new to you, allow six to eight weeks so the vocabulary across all five domains has time to settle, especially the responsible-AI principles and the ML terminology in the second domain.
You need 700 out of 1000 to pass. The score is scaled rather than a raw percentage, so it does not map directly onto a fixed number of correct answers, and harder questions carry more weight. Treat 700 as a comfortable target to clear rather than a line to scrape over.
No. AI-901 is a Fundamentals certification, and unlike Associate-level certifications such as AZ-104 or AI-103, Fundamentals credentials do not expire and carry no renewal requirement. Once you pass, the credential stands even as the underlying skills outline continues to update.
The developer path now runs through a single exam. AI-103 (Microsoft Certified: AI App and Agent Developer Associate) is the natural next step for building generative AI applications and agents, and it expects hands-on Python or .NET SDK work against Azure OpenAI and Azure AI Foundry. It replaced the retired AI-102 (AI Engineer Associate) in June 2026, so AI-103 is where AI-901's vocabulary leads once you are ready to write code.
Practise with over 380 exam-style AI-901 questions across all five domains, with an on-device readiness score, an adaptive study plan, and a full-length simulator. Free to start, works offline.
Download Azure Mastery — free iPhone & iPad · Free to start · No account required