AI-900 is Microsoft’s fundamentals exam for Azure AI, where preparation often hinges on understanding how much technical depth is enough and where a fundamentals-level answer is expected.
The AI-900 Microsoft Azure AI Fundamentals exam validates a candidate’s understanding of artificial intelligence concepts, Azure AI services, machine learning basics, computer vision, natural language processing, generative AI, and responsible AI principles. It is designed for people who are new to Azure AI, including students, career changers, business users, and cloud or data practitioners who want a structured entry point before moving into more technical AI roles.
Last updated: 2026. This guidance reflects the current AI-900 skills outline published by Microsoft Learn and uses the current Azure AI services terminology. Microsoft can revise exam objectives, product names, item formats, and timings, so candidates should check the official Microsoft Learn AI-900 exam page before booking.
What the AI-900 exam actually tests
AI-900 is a fundamentals exam, so the emphasis is on recognising concepts and matching Azure services to business scenarios rather than building production systems. Candidates should understand what machine learning is, how classification and regression differ, when computer vision or language capabilities are appropriate, and how responsible AI principles guide the use of AI systems.
A common preparation mistake is spending too much time on algorithm derivations, Python libraries, or deep learning mathematics. Those topics matter in advanced AI work, but AI-900 is more concerned with whether a candidate can interpret a scenario and identify the right capability or service. For example, a question is more likely to ask which Azure AI service supports sentiment analysis than to ask for the formula behind a model evaluation metric.
Microsoft’s naming changes also matter. Older learning material may refer to Azure Cognitive Services, while current Microsoft terminology uses Azure AI services. Similarly, Form Recognizer is now Azure AI Document Intelligence. Exam questions and supporting study resources may contain either older or newer wording depending on when they were updated, so candidates should learn the mapping rather than treating the names as separate products.
Exam format, scoring, and what not to expect
Microsoft Learn is the authoritative source for exam format and policies. The AI-900 exam is scored on a scale of 1,000 points, and the passing score is 700. Microsoft may vary the number of questions, time allowed, and item mix, so any fixed claim about the exact number of questions or duration should be treated cautiously unless it is confirmed on the current exam page.
Fundamentals exams typically use objective item types such as multiple choice, matching, drag-and-drop, ordering, and short scenario-based questions. Candidates should not rely on claims that AI-900 includes live Azure portal labs unless Microsoft Learn explicitly states that for the current exam. The safer assumption is that the exam tests recognition, interpretation, and service selection rather than hands-on deployment under exam conditions.
There is no penalty for guessing in the usual Microsoft scoring model, so unanswered questions are wasted opportunities. In practice, a sensible approach is to answer confidently known items first, flag uncertain items for review, and return to them after completing the first pass. Responsible AI questions can be overthought; the strongest answers usually align with definitions and applied principles such as fairness, reliability and safety, privacy and security, inclusiveness, transparency, and accountability.
The main knowledge areas to prioritise
The AI-900 skills outline is the best structure for study because it shows how Microsoft groups the exam content. Candidates should be comfortable with AI workloads and considerations, fundamental machine learning principles, computer vision workloads, natural language processing workloads, generative AI concepts, and Azure services that support those workloads.
Machine learning preparation should focus on the practical meaning of common tasks. Classification predicts categories, regression predicts numeric values, clustering groups similar items, and anomaly detection identifies unusual patterns. Candidates should also understand the basic lifecycle of training and evaluating a model, but they do not need to become machine learning engineers to pass a fundamentals exam.
For Azure AI services, the important skill is matching a scenario to the service family. Image analysis, optical character recognition, speech-to-text, translation, sentiment analysis, conversational language understanding, document extraction, search, and generative AI all appear as recognisable business capabilities. If a question describes extracting fields from invoices, Azure AI Document Intelligence is a more natural fit than a general language service. If it describes detecting objects in images, the capability points toward vision services.
Responsible AI deserves careful attention because it is easy to answer from instinct rather than from Microsoft’s stated principles. A scenario about checking a model for different error rates across demographic groups points to fairness. A scenario about explaining why a system made a recommendation points to transparency. A scenario about assigning ownership for monitoring and escalation points to accountability.
A practical two-to-four week study plan
Most beginners can prepare more effectively by studying in short, focused blocks rather than trying to memorise isolated service names. A two-week plan suits candidates with some cloud or data background. A four-week plan is more realistic for candidates new to both Azure and AI vocabulary.
- Week 1: Read the Microsoft Learn skills outline, complete the introductory AI concepts modules, and build a glossary of core terms such as classification, regression, inference, token, prompt, and responsible AI.
- Week 2: Study Azure AI services by workload area, then practise mapping scenarios to vision, speech, language, document intelligence, search, and generative AI capabilities.
- Week 3: Review weak areas through Microsoft Learn modules and use Azure sandbox or free learning environments where available to explore interfaces without adding unnecessary cost.
- Week 4: Take timed practice questions, review every missed item against the skills outline, and rehearse exam-day tactics such as flagging uncertain answers and managing time.
When Microsoft Learn screenshots or Azure portal screens differ from current interfaces, candidates should focus on the concept behind the task. Azure services change names, blades move, and portal designs are refreshed, but the exam objective usually remains tied to the capability. The useful question is not “where was this button in the module?” but “which service or principle is Microsoft testing here?”
Some candidates benefit from guided preparation when they have limited time or prefer structured instruction. The Readynez AI-900 Azure AI Fundamentals course can be used in that context, especially when a learner wants exam preparation aligned to the skills outline rather than a broad introduction to AI.
How to reason through AI-900 scenario questions
Scenario questions are often easier when the candidate separates the business need from the distracting detail. A simple method is to identify the capability first, then identify the constraint, and only then choose the Azure service or principle. Capability means the AI task, such as analysing images, extracting text, translating speech, classifying documents, or generating responses. Constraint means the condition that affects the answer, such as privacy, transparency, latency, cost, or human review.
Consider a question where a company wants to extract vendor names, invoice numbers, totals, and dates from scanned invoices. The capability is document extraction, not general chat or image classification. The best fit would be Azure AI Document Intelligence because the business problem is structured information extraction from documents.
Consider another question where a support team wants to understand whether customer comments are positive, negative, or neutral. The capability is sentiment analysis within natural language processing. A candidate who recognises the workload can avoid being distracted by unrelated options such as speech recognition or computer vision.
A responsible AI scenario might describe a loan approval model that performs less accurately for one group of applicants than another. The capability is less important than the principle. The issue is fairness, and the right answer will usually involve evaluating and mitigating unequal impact rather than simply increasing model complexity.
Where AI-900 fits with other Microsoft fundamentals exams
AI-900 is a good starting point when the goal is to understand AI concepts and Azure AI services. AZ-900 is better suited to candidates who first need broad Azure platform knowledge, including subscriptions, governance, compute, networking, and storage. DP-900 is the more relevant fundamentals exam for people whose immediate goal is data concepts and Azure data services; candidates choosing that route may find Microsoft Azure training useful as a broader foundation and can compare it with a data-first path such as DP-900.
After AI-900, the natural next step depends on the role. People who want to design and implement AI solutions need deeper hands-on skills with Azure AI services, search, orchestration, monitoring, and integration. People moving toward analytics or data engineering may instead strengthen data platform skills before returning to applied AI.
Using practice questions without learning the wrong lesson
Practice questions are useful when they teach reasoning, but they can be harmful when used as memorisation drills. AI-900 is broad enough that candidates who memorise question wording may be exposed by small changes in scenario phrasing. A better review habit is to write down why each wrong option is wrong.
For service-selection questions, candidates should ask whether the answer addresses the exact workload. For responsible AI questions, they should ask which Microsoft principle is being tested. For machine learning questions, they should ask whether the problem is asking for a prediction type, a training concept, or a model evaluation idea.
Timed practice is also important because fundamentals questions can feel deceptively simple. Candidates often lose time by debating between two plausible services when one word in the scenario points clearly to the workload. Flagging those items and moving on protects time for the rest of the exam.
FAQ
What score is needed to pass AI-900?
The passing score for AI-900 is 700 out of 1,000. Microsoft uses scaled scoring, so candidates should focus on the published skills measured rather than trying to calculate a raw percentage from practice tests.
Does AI-900 include live Azure portal labs?
Candidates should not assume AI-900 includes live Azure portal labs unless the current Microsoft Learn exam page says so. Fundamentals exams usually use objective item types such as multiple choice, matching, drag-and-drop, and scenario-based questions.
How long does it take to prepare for AI-900?
A candidate with basic cloud or data knowledge may prepare in about two weeks of focused study. Someone new to Azure and AI terminology should plan closer to four weeks, with time for Microsoft Learn modules, scenario practice, and review of responsible AI principles.
Is coding required for the AI-900 exam?
AI-900 does not require production coding skills. Candidates should understand what Azure AI services do, how common AI workloads differ, and how responsible AI principles apply to business scenarios.
Should candidates take AI-900 before AZ-900 or DP-900?
It depends on the goal. AI-900 is the right starting point for AI concepts and Azure AI services, AZ-900 is stronger for general Azure platform basics, and DP-900 is better for data concepts and Azure data services.
Building confidence before the exam
The strongest AI-900 preparation stays close to the Microsoft Learn skills outline, uses current Azure AI services terminology, and practises scenario reasoning rather than memorising disconnected facts. Candidates who can identify the workload, recognise the constraint, and apply the appropriate responsible AI principle are usually studying at the right level for the exam.
A practical next step is to review the official skills outline, choose a two-to-four week schedule, and decide whether self-study or guided training fits the available time. Learners planning a broader Microsoft certification path can also consider Unlimited Microsoft Training, and anyone unsure about the right route can contact Readynez for guidance.
