AI-900 Exam Explained: What It Covers, How Hard It Is and How to Pass
AI-900 is the easiest Microsoft exam that still requires study. It is a fundamentals exam, so there is no configuration, no case study and no command syntax. What it does require is a clear map of Azure's AI services and the vocabulary of machine learning, because the questions are almost all "which service" and "which type of model" decisions dressed in short business scenarios.
Most candidates pass on the first attempt with one or two weeks of preparation. The ones who do not usually made one of two mistakes: they assumed general AI knowledge was enough and never learned the Azure product names, or they skipped the responsible AI principles because they seemed soft. This guide covers the format, the five skill areas, the questions that catch people, and a short plan.
AI-900 at a glance
| Exam | AI-900: Microsoft Azure AI Fundamentals |
| Certification earned | Microsoft Certified: Azure AI Fundamentals |
| Time | About 45 minutes of exam time inside a longer appointment |
| Questions | Roughly 40 to 60, multiple choice, multiple response, drag and drop and hot area |
| Passing score | 700 on a 1 to 1000 scale |
| Fee | $99 USD; varies by country, and Microsoft regularly issues free vouchers through training events and Learn challenges |
| Validity | Fundamentals certifications do not expire |
| Prerequisites | None. No coding and no prior Azure certification required |
What AI-900 actually tests
The skills outline lists five areas. Microsoft adjusts the exact weights when it updates the exam, but each area sits between roughly 15 and 25 percent, with generative AI and machine learning principles slightly heavier in the current version. Treat all five as required.
Describe AI workloads and considerations. The kinds of problems AI solves (prediction, anomaly detection, computer vision, natural language processing, knowledge mining, document intelligence, generative AI) and the six responsible AI principles: fairness, reliability and safety, privacy and security, inclusiveness, transparency, accountability. Expect a scenario and a question about which principle it illustrates.
Describe fundamental principles of machine learning on Azure. Regression versus classification versus clustering, features and labels, training and validation data, the common evaluation metrics, deep learning at a conceptual level, and how Azure Machine Learning supports the workflow through automated machine learning, the designer and compute targets. This is the most "academic" area and the one where non-technical candidates need the most time.
Describe features of computer vision workloads on Azure. Image classification, object detection, optical character recognition, facial detection and analysis, and which Azure AI service does each: Azure AI Vision, Azure AI Face, Azure AI Custom Vision, and Document Intelligence for forms and receipts.
Describe features of natural language processing workloads on Azure. Key phrase extraction, entity recognition, sentiment analysis, language detection, speech to text and text to speech, translation, and conversational language understanding. Again the question is which service: Azure AI Language, Azure AI Speech, Azure AI Translator.
Describe features of generative AI workloads on Azure. Large language models, prompts and completions, the difference between a foundation model and a fine-tuned one, Azure OpenAI Service, Azure AI Foundry (the studio experience), Copilot as a product family, and the responsible use considerations that come with generative models. This area was expanded in the recent updates and now carries a real share of the exam.
What catches people
Service names. Microsoft has renamed its AI services more than once. The exam uses the current names (Azure AI Vision, Azure AI Language, Azure AI Speech, Azure AI Document Intelligence, Azure OpenAI Service). Old study material uses Cognitive Services, Form Recognizer, LUIS and Text Analytics. Learn the mapping, because the exam will not.
Regression versus classification versus clustering. Predicting a number is regression, predicting a category is classification, grouping without labels is clustering. The scenarios are short and the wrong answers are the other two.
Responsible AI. A scenario about a hiring model that treats candidates differently is about fairness; one about explaining a loan decision is about transparency; one about a system that fails safely is reliability and safety. Six principles, learn all six with an example each.
Generative AI terms. Tokens, prompts, grounding, embeddings, content filters, fine-tuning. These are new to many candidates and they now appear regularly.
Try three real AI-900 practice questions
From our AI-900 bank, unedited, with explanations behind the toggle.
How long to study
- Already working with Azure or data: 8 to 15 hours over one week. Mostly vocabulary and service names.
- Technical background, new to AI and Azure: 20 to 30 hours over two weeks.
- Non-technical background: 30 to 45 hours over three to four weeks, with extra time on the machine learning principles area.
A short plan
Days 1 to 3: workloads and responsible AI. Learn the workload types and the six principles with one example each.
Days 4 to 7: machine learning principles. Regression, classification, clustering, features and labels, the evaluation metrics, and a walk through Azure Machine Learning's automated ML in the free tier so the terms have a picture attached.
Days 8 to 10: vision and language. Open the Azure AI services in the portal or the Foundry playground and run one demo of each: analyse an image, extract text from a receipt, detect sentiment in a paragraph, translate a sentence.
Days 11 to 12: generative AI. Use a playground with a language model, change the temperature and the system prompt, and read the content filtering settings.
Days 13 to 14: practice. Timed sets from our bank, then review by area. Our timed AI-900 exam scores you by skill area, which shows the last gap.
Exam day and after
Pearson VUE or online proctoring. The questions are short; pace is rarely a problem. Read the scenario for the one clue that selects the service or the model type, and answer. There is no penalty for guessing. A pass earns a certification that does not expire, and a fail can be retaken after 24 hours at full price, so the exam is low risk compared with almost any other.
Is AI-900 worth it?
As a first Microsoft certification, or as a structured way to learn what the Azure AI catalogue actually contains, yes. It is not a job qualification on its own, but it is the fastest route to speaking accurately about AI services, and it is the natural step before AI-102, the engineer-level exam that builds on the same map.
Start with the free AI-900 exam dumps, read the AI-900 certification guide for the skills outline, and sit the exam when practice scores are above 85 percent.