The AI-901 (Azure AI Fundamentals) tests two things: identifying AI concepts, worth 40 to 45% of the exam, and building AI solutions with Microsoft Foundry, worth 55 to 60%. Below is every skill Microsoft lists for the exam, in plain English, with what each area is worth and what to actually study. The skills were last updated on 15 April 2026, when the AI-901 replaced the older AI-900.
The AI-901 at a glance
| Area | Weight | What it covers |
|---|---|---|
| Identify AI concepts and capabilities | 40-45% | Responsible AI, AI models, and common AI workloads |
| Implement AI solutions by using Microsoft Foundry | 55-60% | Building generative AI apps, agents, text/speech, vision, and information extraction |
Pass mark: 700 out of 1000. It is a beginner-level exam with no prerequisites, but the audience profile now expects “knowledge of Python coding syntax” and familiarity with “REST APIs, SDKs, and CLIs.” Source: the official AI-901 study guide.
Domain 1: Identify AI concepts and capabilities (40-45%)
The conceptual half. No building here, just understanding what the technologies are and when to use them. It breaks into three groups.
Responsible AI: the six principles
Microsoft tests six principles of responsible AI. Know what each one means and a real example of it:
- Fairness: the system treats all groups of people equitably.
- Reliability and safety: it behaves consistently and fails safely.
- Privacy and security: it protects data and respects consent.
- Inclusiveness: it works for people of all abilities and backgrounds.
- Transparency: people can understand how and why it makes decisions.
- Accountability: people, not the model, remain responsible for outcomes.
This group looks soft, so people under-revise it, then meet it repeatedly on the exam.
AI models and how to choose one
- How generative AI models work at a high level.
- Picking an appropriate model based on its capabilities.
- Choosing deployment options and configuration parameters for a model.
Common AI workloads
Recognise the main workload types and the techniques behind them:
- Generative and agentic AI: producing content, and agents that take actions.
- Text analysis: keyword extraction, entity detection, sentiment analysis, and summarisation.
- Speech: speech recognition (speech to text) and speech synthesis (text to speech).
- Computer vision: image analysis and image-generation models.
- Information extraction: pulling structured data out of text, images, audio, and video.
Domain 2: Implement AI solutions by using Microsoft Foundry (55-60%)
The bigger, hands-on half, and the part that did not exist on the old AI-900. You are expected to have actually built small things in Microsoft Foundry, not just describe them.
Generative AI apps and agents
- Write effective system and user prompts.
- Deploy a model and interact with it in the Foundry portal.
- Build a lightweight chat client using the Foundry SDK.
- Create and test a single-agent solution in the Foundry portal, and a small client app for an agent.
Text and speech solutions
- Build a lightweight app that includes text analysis.
- Respond to spoken prompts using a deployed multimodal model.
- Build a lightweight app using Azure Speech in Foundry Tools.
Computer vision and image generation
- Interpret visual input in prompts using a deployed multimodal model.
- Create new visual outputs using generative models.
- Build a lightweight app with vision capabilities.
Information extraction
- Extract information from documents and forms using Azure Content Understanding.
- Extract information from images, audio, and video.
- Build a lightweight app with information-extraction capabilities.
How is the AI-901 scored?
You need 700 out of 1000 to pass. Microsoft does not publish a fixed question count, but expect roughly 40 to 60 questions in about 45 minutes, mixing multiple-choice and multi-response items. There is no penalty for a wrong answer, so never leave a question blank. Check the official AI-901 exam page for current pricing and to book, since fees vary by region.
How should you split your study time?
Weight your revision the way the exam is weighted: more than half of it is the Microsoft Foundry implementation half. A rough split that works for most people:
- One pass on the concepts (Domain 1) to lock in responsible AI and the workload types.
- The bulk of your time in Foundry (Domain 2): deploy a model, send prompts, and build a tiny chat client and agent so the tasks feel familiar.
- Timed mocks to find which half is weaker and drill it.
Most candidates are comfortable on the concepts and lose marks on the Foundry tasks, simply because reading about them is not the same as having done them once. If you are deciding how hard this will be for you, see how hard the AI-901 really is.
Practise the way the exam is weighted
PassTheVoid mirrors the official 40-45 / 55-60 split, so you are not over-practising concepts and under-practising Foundry. You get a free timed mock, a per-domain breakdown that shows whether to revise concepts or Foundry next, and offline study. It is free and ad-free, with a one-time Pro unlock for the full question bank and unlimited mocks.