AI-900 Study Guide 2026: Complete Azure AI Fundamentals Exam Prep
Everything you need to pass the AI-900 Azure AI Fundamentals exam — all 5 domains, a study plan, and what actually shows up on test day.
Quick Summary
- • AI-900 is a Fundamentals-level exam with 40–60 questions, 45 minutes, 700/1000 passing score
- • Covers 5 domains: AI concepts, machine learning, computer vision, NLP, generative AI
- • Generative AI has grown significantly in recent exam versions — do not underestimate it
- • Exam cost: $165 USD
What is the AI-900 Exam?
AI-900 is Microsoft's Fundamentals-level certification for artificial intelligence and machine learning concepts on Azure. It validates conceptual understanding of AI workloads, responsible AI principles, and how Azure AI services are used to build real-world solutions — without requiring programming or data science skills.
The exam has evolved significantly since its launch to include substantial generative AI content, reflecting the rapid growth of Azure OpenAI Service and Microsoft Copilot products. Recent versions test knowledge of large language models (LLMs), prompt engineering, and Copilot capabilities alongside traditional AI topics.
AI-900 is suitable for business professionals, developers starting their AI journey, and anyone who wants to demonstrate foundational AI literacy in Microsoft's ecosystem.
| Detail | Information |
|---|---|
| Exam Code | AI-900 |
| Credential Earned | Azure AI Fundamentals |
| Number of Questions | 40–60 questions |
| Time Limit | 45 minutes |
| Passing Score | 700 out of 1000 |
| Exam Price | $165 USD |
| Exam Level | Fundamentals |
| Prerequisites | None |
AI-900 Exam Domains & Weightings
AI-900 covers five domains. Generative AI has the highest single-domain weighting and is also the most frequently updated content area.
Domain 1: AI Concepts and Responsible AI
15–20%- • AI vs. machine learning vs. deep learning — what distinguishes each
- • Common AI workloads — prediction, computer vision, NLP, knowledge mining, generative AI
- • Microsoft's Responsible AI principles: Fairness, Reliability & Safety, Privacy & Security, Inclusiveness, Transparency, Accountability
- • Azure AI services overview — Azure AI Services (Cognitive Services), Azure Machine Learning, Azure OpenAI Service
- • AI and ethics considerations — bias, explainability, data privacy
Study tip: Memorize all 6 Responsible AI principles and be able to identify which principle applies to a given scenario. This appears across multiple questions.
Domain 2: Machine Learning on Azure
20–25%- • Supervised learning — regression (predict values), classification (predict categories)
- • Unsupervised learning — clustering (find groupings)
- • Reinforcement learning — reward-based training concepts
- • Azure Machine Learning workspace — datasets, experiments, pipelines, compute targets
- • Automated ML (AutoML) — automatic model selection and hyperparameter tuning
- • Azure Machine Learning Designer — no-code visual pipeline building
- • Model evaluation metrics — accuracy, precision, recall, F1, AUC-ROC (conceptual)
- • Feature engineering, training vs. test vs. validation data splits
Study tip: Know when to use regression vs. classification vs. clustering — scenario questions describing "predict a number" (regression), "predict a category" (classification), or "find groups" (clustering) are common.
Domain 3: Computer Vision Workloads on Azure
15–20%- • Image classification — assigning a category label to an entire image
- • Object detection — identifying and locating objects within an image (bounding boxes)
- • Semantic segmentation — classifying each pixel in an image
- • Optical character recognition (OCR) — extracting text from images and documents
- • Facial analysis — face detection, face recognition, facial attributes
- • Azure AI Vision — Image Analysis, Face API, OCR/Read API
- • Azure AI Custom Vision — custom image classification and object detection models
- • Azure AI Document Intelligence — form extraction, document analysis
Domain 4: Natural Language Processing Workloads on Azure
15–20%- • Text analysis — key phrase extraction, entity recognition, sentiment analysis, language detection
- • Speech capabilities — speech-to-text, text-to-speech, speaker recognition, translation
- • Translation — text translation, document translation, custom translation
- • Conversational AI — chatbots, Question Answering (Azure AI Language), CLU (Conversational Language Understanding)
- • Azure AI Language service — sentiment, NER, key phrase, PII detection
- • Azure AI Speech service — speech recognition, synthesis, translation
- • Azure AI Translator — real-time text translation
- • Azure AI Bot Service — building and deploying conversational bots
Domain 5: Generative AI Workloads on Azure
20–25%The most updated domain — reflects rapid changes in Microsoft's AI portfolio:
- • Large language models (LLMs) — transformers, tokens, embeddings, pre-training vs. fine-tuning
- • Generative AI capabilities — text generation, image generation, code generation, summarization
- • Azure OpenAI Service — GPT models, DALL-E, embeddings, completions API, deployments
- • Prompt engineering — zero-shot, few-shot, system prompts, prompt design principles
- • Retrieval Augmented Generation (RAG) — grounding LLMs with external knowledge
- • Microsoft Copilot products — Microsoft 365 Copilot, Copilot Studio, GitHub Copilot
- • Responsible generative AI — content filtering, safety systems, harmful content mitigation
- • Azure AI Foundry — building and deploying AI applications and agents
Study tip: Generative AI is 20–25% of the exam and growing. Do not neglect Azure OpenAI Service, prompt engineering, and Microsoft Copilot — these are now core exam topics.
Ready to test yourself?
Try 40 Free AI-900 Practice Questions
Scenario-based questions with detailed explanations. No credit card required.
Start Free Practice →How Hard is AI-900?
AI-900 is comparable to AZ-900 and MS-900 in difficulty — accessible for non-technical candidates with proper preparation. The main challenge is the breadth of AI product names and concepts to understand, combined with the rapidly evolving generative AI domain that requires staying current with Microsoft's AI services portfolio.
Why candidates fail AI-900
- • Using outdated study materials: AI-900 has been updated significantly to include generative AI. Older practice tests and guides miss substantial exam content
- • Confusing ML task types: Regression vs. classification vs. clustering scenarios are consistently tricky — candidates who can't immediately categorize a task type lose marks
- • Azure AI service name confusion: Azure AI Vision, Language, Speech, Document Intelligence — candidates confuse which service handles which capability
- • Skipping Responsible AI: The 6 principles appear across multiple questions in different contexts — this is a high-return study topic
3-Week AI-900 Study Plan
This plan assumes 1–1.5 hours per day. Candidates with data science or software development backgrounds can often compress this to 2 weeks.
Week 1: AI Concepts, Machine Learning & Responsible AI
- Days 1–2: AI fundamentals — AI vs. ML vs. deep learning, common AI workloads and their use cases
- Days 3–4: Responsible AI — memorize all 6 principles, practice applying them to scenarios
- Days 5–6: Machine learning — supervised (regression/classification), unsupervised (clustering), reinforcement learning
- Day 7: Azure Machine Learning — AutoML, Designer, evaluation metrics + practice questions
Week 2: Computer Vision & Natural Language Processing
- Days 1–2: Computer vision — image classification, object detection, OCR, facial analysis, Azure AI Vision services
- Days 3–4: Azure AI Custom Vision, Document Intelligence — when to use custom vs. pre-built models
- Days 5–6: NLP — text analysis, sentiment, NER, Azure AI Language + Speech services, translation
- Day 7: Conversational AI — Bot Service, QnA Maker replacement (Question Answering), CLU + practice questions
Week 3: Generative AI & Mock Exams
- Days 1–2: LLMs — transformers, tokens, pre-training, fine-tuning, Azure OpenAI Service models
- Days 3–4: Prompt engineering, RAG, responsible generative AI, content filtering in Azure OpenAI
- Day 5: Microsoft Copilot products — Microsoft 365 Copilot, Copilot Studio, GitHub Copilot, Azure AI Foundry
- Day 6: Full 45-minute timed mock exam — identify weak domains
- Day 7: Targeted review + second mock exam. Book exam if scoring 80%+.
Best AI-900 Study Resources
1. Microsoft Learn AI-900 Learning Path (Free)
The official learning path is the best single resource for AI-900. It covers all five domains with interactive modules and includes hands-on sandbox exercises for Azure Machine Learning and Azure AI services. The generative AI modules have been updated for 2026 — make sure you are using the latest version.
2. MSCertQuiz Practice Tests
500 AI-900 practice questions across all five domains with detailed explanations. Updated for 2026 exam content including generative AI, Azure OpenAI Service, and Microsoft Copilot questions. Strong coverage of ML task type scenarios and Responsible AI principle application questions.
Start free AI-900 practice →3. Azure AI Services Documentation
For understanding what each Azure AI service does, the official documentation "What is..." overview pages are concise and accurate. Build a comparison table: Vision, Language, Speech, Document Intelligence, Custom Vision — mapping each to its capabilities. This directly reduces service confusion on the exam.
4. Microsoft AI School
Microsoft's free AI learning platform includes introductory courses on machine learning, responsible AI, and generative AI aligned to AI-900 content. Particularly useful for the conceptual AI foundations that the exam tests before diving into Azure-specific services.
AI-900 Exam Day Tips
Do
- • For ML task questions: identify the output type — numeric value (regression), category (classification), group (clustering)
- • For Responsible AI questions: map each scenario to the closest principle — Fairness is about bias, Transparency is about explainability
- • For generative AI questions: RAG = grounding with external data; fine-tuning = retraining on domain-specific data
- • Use the flag feature for generative AI questions if unsure — they may become clearer after reviewing the rest
Don't
- • Don't confuse Azure AI Vision (pre-built image analysis) with Azure AI Custom Vision (custom image models)
- • Don't mix up Azure AI Language (text analysis) and Azure AI Speech (audio) — they handle different input types
- • Don't assume "deep learning" and "machine learning" are interchangeable — deep learning is a subset of ML
- • Don't underestimate generative AI content — it's now 20–25% of the exam and many older study materials miss it
Ready to Practice AI-900?
500 questions across all 5 domains. Practice mode with explanations + timed exam simulation.
Start Free Practice →Frequently Asked Questions
Do I need programming knowledge for AI-900?
No. AI-900 tests conceptual understanding of AI and machine learning, not coding skills. You do not need to write Python, SQL, or any other programming language. The exam is designed for business professionals, IT generalists, and non-technical roles who want to understand AI capabilities and responsible AI principles.
What is the difference between AI-900 and AZ-900?
AZ-900 covers Azure cloud fundamentals — infrastructure, storage, compute, governance. AI-900 covers artificial intelligence and machine learning fundamentals — ML concepts, computer vision, NLP, and generative AI. Both are Fundamentals-level exams with no prerequisites. If your work involves AI or data, start with AI-900. If your work is more infrastructure/cloud, start with AZ-900.
Is AI-900 worth it in 2026?
Yes. AI literacy is increasingly expected across technical and non-technical roles. AI-900 provides recognized evidence of foundational AI knowledge, particularly in Microsoft's ecosystem including Azure OpenAI Service and Microsoft Copilot products. It's a quick certification (2–3 weeks preparation) with growing recognition as AI becomes embedded in enterprise technology stacks.
What comes after AI-900?
AI-900 is a strong foundation for more technical AI certifications. The natural next steps are AI-102 (Azure AI Engineer Associate) for hands-on AI solution building, or DP-100 (Azure Data Scientist Associate) for machine learning engineering. Both require programming skills that AI-900 does not.