AI-901 has two domains in Microsoft's skills outline as of April 15, 2026. Identifying AI concepts and capabilities is 40-45% of the exam, and building small AI solutions in Microsoft Foundry is 55-60%. Give the concepts one week and the Foundry exercises two, and set aside the machine learning theory from AI-900 courses.
Exam format
| Domain | Weight | Skill groups | Plan |
|---|---|---|---|
| Identify AI concepts and capabilities | 40-45% | Responsible AI principles; AI model components and configurations; AI workloads | Week 1 |
| Implement AI solutions by using Microsoft Foundry | 55-60% | Generative AI apps and agents; text and speech; computer vision and image generation; information extraction | Weeks 2 and 3 |
AI-901 is a Fundamentals exam, so you get 45 minutes of exam time inside a 65-minute seat, and Microsoft Learn is not available as a lookup during it. You need a score of 700 or greater to pass.
The exam page expects "knowledge of Python coding syntax and programming techniques" and familiarity with Azure resources. The study guide adds that you should know your way around REST APIs, SDKs and CLIs. Most questions cover generally available features, though Preview features can appear if they are commonly used.
The exam costs USD 99 in the United States as of October 7, 2026, before tax, and the price depends on the country where you sit it.
Coming from AI-900
Microsoft retired AI-900 on June 30, 2026 and replaced it with AI-901 for the Azure AI Fundamentals certification. The AI-900 outline had a domain called "Describe fundamental principles of machine learning on Azure", worth 15-20%, covering regression, classification and clustering, features and labels, Azure Machine Learning and automated machine learning. AI-901 has no machine learning domain.
Two of Microsoft's own pages still carry AI-900 material. The "Find documentation" links at the bottom of the AI-901 study guide point to Anomaly Detector, Language Understanding, Azure Machine Learning and Azure Bot Service, none of which appear in the skills list. The schedule box on the AI-901 exam page still describes the old exam, machine learning principles included. Plan from the skills measured list itself.
AI-900 asked you to describe services. Most objectives in AI-901's heavier domain are about building something, such as "Create a lightweight chat client application by using the Foundry SDK", and the exam tests you on those objectives.
Azure subscription
The Foundry exercises in weeks 2 and 3 run in an Azure subscription. Microsoft's exercise pages note that a new subscription includes credits for the first 30 days. If you need a new one, create it at the start of week 2, so the 30 days cover the hands-on weeks.
The three-week plan
Three weeks fits if you have about five hours a week. Microsoft's course AI-901T00-A is built from two learning paths, and their module times set most of the schedule. The concept modules for domain 1 add up to 3 hours 17 minutes. The Foundry modules for domain 2 add up to 4 hours 59 minutes. The rest of each week goes to writing client code and to review.
Week 1
Work through the AI concepts for developers and technology professionals path. Its prerequisites are basic computing concepts and math.
- Introduction to AI concepts (40 minutes), which includes the responsible AI unit and an exercise that explores the workloads
- Introduction to generative AI and agents (37 minutes)
- Introduction to natural language processing concepts (30 minutes)
- Introduction to AI speech concepts (28 minutes)
- Introduction to computer vision concepts (34 minutes)
- Introduction to AI-powered information extraction concepts (28 minutes)
The path's seventh module, on retrieval-augmented generation (34 minutes), is not in the skills list. Microsoft notes that related topics may appear, so keep it for week 3 if you have time left.
By the end of the week you should be able to take a short business scenario and name its workload. The outline's workloads are generative and agentic AI, text analysis, speech, computer vision and information extraction. Within text analysis, learn the four techniques the outline names (keyword extraction, entity detection, sentiment analysis and summarization) well enough to say what each one returns for a piece of text.
Week 2
Switch to the Get started with AI applications and agents on Azure path. Its prerequisites are basic computing concepts and Python.
- Get started with AI in Azure (56 minutes). Its exercise creates a Foundry project and connects a client application to it.
- Get started with generative AI and agents in Azure (58 minutes)
- Get started with text analysis in Azure (45 minutes)
- Get started with speech in Azure (47 minutes)
Several of the labs in this path have you work in the Foundry portal, and the Python client code sits in the modules' reading units. After each of the later modules, copy its Python sample into your own editor, point it at your own deployment and run it. Then change one thing, such as the prompt or the input file.
Week 3
- Get started with computer vision in Azure (50 minutes)
- Get started with AI-powered information extraction in Azure (43 minutes), which uses Azure Content Understanding in Foundry Tools
The path's last module, on Foundry IQ (38 minutes), is also outside the skills list. Treat it like the retrieval-augmented generation module.
Spend the rest of the week on review. Write the client code for your two weakest skill groups again without the samples open. Then take Microsoft's free practice assessment, which for AI-901 runs on AI Skills Navigator and needs a sign-in. The AI-901 study guide links to it. Go back to the module and exercise for any skill group whose questions gave you trouble.
What each skill group asks
Responsible AI
The outline asks you to describe considerations for each of Microsoft's six principles, which are fairness, reliability and safety, privacy and security, inclusiveness, transparency, and accountability. The responsible AI unit in Introduction to AI concepts takes about 3 minutes, so most of the practice is yours. For each principle, name one design decision in an AI app that would serve it.
Models and configuration
How generative AI models work is in the week 1 module Introduction to generative AI and agents. Choosing a model and choosing its deployment options and parameters both sit in the week 2 module Get started with generative AI and agents in Azure, whose objectives include identifying the right model for a task with the Foundry model catalog.
That module's unit on generative AI models lists deployment types such as standard, global batch and regional provisioned throughput, and explains that a Tokens Per Minute allocation sets how much traffic a deployment can handle before it is throttled. Its unit on using a model covers the playground parameters temperature, max output tokens and system instructions. A system prompt sets the assistant's behavior, tone and guardrails, and the user prompt is the end user's request. Be able to say what each of these changes, because the outline asks for both deployment options and configuration parameters.
Building in Foundry
Each build skill area in domain 2 has a lab and a Python sample in the second learning path. Several objectives ask for a "lightweight application", and the samples are where that code is.
| Skill area | What the lab has you do | Python sample |
|---|---|---|
| Generative AI apps and agents | Deploy and explore a model in the Foundry portal, then use it in an agent with knowledge tools (35 minutes) | Units "Using a generative AI model" and "Creating an agent" |
| Text analysis | Use a generative model in the portal for common language tasks, then explore Azure Language features (20 minutes) | Unit "Create a client application that analyzes text" |
| Speech | Create a speech-capable agent with Azure Speech Voice Live (25 minutes) | Unit "Speech recognition" |
| Computer vision and image generation | Use generative models in Foundry to work with visual data (30 minutes) | Unit "Multimodal models for image analysis" |
| Information extraction | Use Azure Content Understanding to analyze documents and extract information (25 minutes) | Unit "Extract information from documents" |
Booking the exam
Before you book, open the exam sandbox once so the question screens and review controls are familiar. Book when you can write the Python client for every domain 2 skill area without the sample open, explain what each call does, and have gone back over every skill group you struggled with on the practice assessment.
If the first attempt does not go your way, Microsoft's retake policy makes you wait 24 hours before the second attempt and 14 days between later ones, with at most five attempts in 12 months. Once you pass, there is nothing to renew, since Fundamentals certifications do not expire.
Checking your weak areas
If you want a second read on your weak skill groups before you pick a date, CramHQ's free AI-901 assessment takes about 15 minutes, needs no credit card, and shows which areas are weakest when you finish. If it shows more gaps than your review week can cover, the AI-901 Pass Plan costs USD 19.99 once for 12 months of access, with repair tasks ordered by how many exam points each gap costs, five practice tests and a 40-question timed simulation.