Just passed the AI-900 yesterday with an 875. Took me about three weeks of studying, mostly evenings after work, 45 minutes to an hour a day. I've been in data analytics for two years but had basically no machine learning background before starting, so I wasn't sure what to expect difficulty-wise.
The exam is 40–60 questions with a 60-minute limit, which sounds comfortable but some questions are multi-select and case-study format so don't assume you can breeze through. Content breaks down roughly into AI workloads and considerations (~15%), fundamental ML principles (~20%), computer vision (~15%), NLP (~15%), and Azure AI services (~35%). That last section is the heaviest and the one I almost underestimated.
Microsoft Learn has free learning paths and they're genuinely useful for the Azure-specific content — Cognitive Services, Azure Machine Learning Studio, which service does what. I also used a paid Udemy course but the free material covered about 85% of what I needed. Don't skip the hands-on labs if you have time; recognizing service names in context on the exam is easier when you've actually navigated the portal.
Worth noting the exam can be taken online proctored now. I did it at home with no issues — just clear your desk of papers and disconnect any second monitors ahead of time or you'll waste 20 minutes on setup with the proctor.
The Azure free tier is enough to run all the AI-900 relevant labs. I used it for about two weeks and it didn't cost anything. Worth setting up just to see the interfaces even if you don't go through every single exercise.
I'm studying for this right now. How important is it to actually use the Azure portal — can you get by just memorizing service descriptions? I don't currently have a subscription set up.
Nice score. I passed with an 820 two months ago. The Azure AI services section tripped me up more than I expected — specifically questions about when to use Form Recognizer versus Document Intelligence versus Custom Vision. Know the use cases for each before you sit.