AI-901 vs AI-900: what changed?
AI-901 replaced AI-900 when Microsoft retired AI-900 on June 30, 2026. Both lead to the same credential, Microsoft Certified: Azure AI Fundamentals, but the exam behind it was rebuilt rather than refreshed. If you are studying from AI-900 material, a meaningful part of it is now wrong.
The domains, side by side
AI-900 — five areas, all descriptive:
| Area | Weight |
|---|---|
| Describe Artificial Intelligence workloads and considerations | 15–20% |
| Describe fundamental principles of machine learning on Azure | 15–20% |
| Describe features of computer vision workloads on Azure | 15–20% |
| Describe features of Natural Language Processing workloads on Azure | 15–20% |
| Describe features of generative AI workloads on Azure | 20–25% |
AI-901 — two areas, one of them hands-on:
| Area | Weight |
|---|---|
| Identify AI concepts and capabilities | 40–45% |
| Implement AI solutions by using Microsoft Foundry | 55–60% |
The three changes that matter
1. The verb changed from “describe” to “implement”. Every AI-900 domain began with describe. On AI-901 the majority of the score sits in a domain that begins with implement. You are expected to have deployed a model in the Foundry portal, written a prompt that works, built a lightweight chat client with the Foundry SDK, and created and tested a single agent.
2. Python is now assumed. Microsoft’s audience profile for AI-901 states you need “knowledge of Python coding syntax and programming techniques” and familiarity with “REST APIs, SDKs, and CLIs”. AI-900 explicitly said data science and software engineering experience were not required. That is a real change in the bar, and the most common thing people get caught out by.
3. Classic machine learning is gone. AI-900’s “fundamental principles of machine learning on Azure” domain — regression, classification, clustering, training and evaluation, the Azure Machine Learning designer — has no counterpart in the AI-901 outline. If you were revising confusion matrices and train/test splits, you can stop.
What survived
Plenty, in compressed form. Responsible AI is still there, and still examined as six principles: fairness, reliability and safety, privacy and security, inclusiveness, transparency, accountability. Computer vision, text analysis, speech and information extraction all still appear — but now inside the two new domains, and framed as things you build with Foundry rather than services you list.
The centre of gravity also moved toward generative and agentic AI, matching the shift Microsoft made at associate level when AI-102 became AI-103.
Already passed AI-900?
Your certification stands. Microsoft Certified: Azure AI Fundamentals does not sit on the annual renewal track that applies to associate, expert and specialty certifications, so there is no renewal assessment pushing you toward AI-901.
Whether to take AI-901 anyway comes down to what you want it for. If the badge was the point, you already have it. If you want the current material — Foundry, agents, prompts, Content Understanding — then AI-901 teaches genuinely different content, and studying for it is worthwhile even though the credential name on your profile will not change.
Where your old knowledge lands
| AI-900 topic | Status on AI-901 |
|---|---|
| Responsible AI principles | Still examined, still six principles |
| AI workload types | Still examined, inside the concepts area |
| Computer vision services | Reframed as building vision into a Foundry app |
| NLP / text analysis | Reframed as text analysis and speech in Foundry |
| Generative AI basics | Expanded, and now partly hands-on |
| Regression, classification, clustering | Gone |
| Azure Machine Learning designer | Gone |
| Training, evaluation, confusion matrices | Gone |
| Agents, prompts, Foundry SDK | New |
| Content Understanding | New |
Roughly speaking: your conceptual knowledge transfers, your classic machine learning knowledge does not, and the hands-on half is new ground for everyone.
Is AI-901 harder than AI-900 was?
Different rather than uniformly harder, but harder for most candidates in practice. AI-900 was genuinely approachable for non-technical people — Microsoft said so explicitly, stating that data science and software engineering experience were not required. AI-901 removed that sentence and replaced it with an expectation of Python.
If you are a developer, AI-901 may well feel easier than AI-900 did, because implementing something concrete is easier than memorising the names of services you never touch. If you came to AI-900 from a business or project role, AI-901 asks noticeably more of you.
If you are studying now
Use AI-901 material only. The two exams share a name and a certification, which makes older courses, videos and question banks look current when they are not. Anything that spends a module on the Azure Machine Learning designer was written for the retired exam.