How to Prepare for AI-200: Check the Prerequisites, Then Plan the Weeks
Published August 18, 2026 · Facts last verified against the official sources on August 18, 2026
Microsoft names seven skills an AI-200 candidate should already hold, and Python is one of them. Check yourself against that list before you plan any weeks. The nine official learning paths hold 32 hours and 4 minutes of reading, and we counted 24 modules and 194 units in them. Twenty-four of those units are exercises you perform rather than read.
This article is about preparing. For the exam's format, its four areas and who it is aimed at, read our AI-200 exam guide. If you are choosing between this certification and AI-103, the other credential that replaced the retired AI-102, we compare them separately. And because this credential is new, what Microsoft has published about it so far is a shorter read.
Start with Microsoft's audience profile, not with a calendar
Most preparation articles hand you a week-by-week schedule. A schedule written by a stranger cannot know what you already do for a living, and for this exam that is the thing that decides everything else.
The certification page publishes an audience profile, and for a Microsoft page it is unusually specific. It is a list of skills you should hold before you begin, not a list of topics the exam will teach you. It is short enough to check yourself against in a minute. Here it is, in our order, with what each item means for a plan:
- Python programming. The exercises and the SDK material are written for developers who already write code. This is the single item that most often means "not yet" rather than "not this month".
- Implementing containerized applications on Azure. A whole exam area sits on this. If you have never built an image and pushed it to a registry, that gap will slow every week that follows.
- Vector databases. You need the idea of an embedding, and of searching by similarity instead of exact match. Three separate services then show you three ways to store one.
- Azure data management services. Azure Database for PostgreSQL, Azure Managed Redis and Cosmos DB for NoSQL. Those three are the services this exam builds on.
- Azure messaging and eventing. Queues, topics and events as design choices — not just as services you have heard of.
- The Azure SDKs, and the third-party SDKs used alongside them. Connecting to a service in code, and knowing where the connection settings and credentials come from.
- Azure monitoring and troubleshooting. Reading logs and traces to find out why something failed, rather than only checking that it started.
That list is the honest gate on whether you are ready to start. If you do not write Python, and have never deployed a container, AI-200 is not a few weeks away. That is not a judgement about ability. It is what happens when an exam assumes a job you have not done yet, and no study calendar can supply the missing part. Learn the missing skill first, on a real subscription, and come back to the certification afterwards. We would rather tell you that than sell you a schedule.
How to use the list: count how many of the seven describe work you have actually done, not work you have read about. That number decides your plan, and we come back to it below.
The official path: nine learning paths, 24 modules, 194 units
Microsoft's course page for AI-200T00-A lists the official training as nine learning paths, and gives the instructor-led version a duration of five days.
Microsoft publishes a duration for each of those paths in its own learning catalogue. Those are the minutes below. The module and unit counts beside them are ours, counted while building our own AI-200 material — Microsoft publishes the paths, not this table.
| Learning path | Minutes | Modules | Units |
|---|---|---|---|
| Implement container application hosting on Azure | 140 | 2 | 15 |
| Deploy and manage apps on Azure Container Apps | 227 | 3 | 27 |
| Deploy and monitor applications on Azure Kubernetes Service | 212 | 3 | 21 |
| Develop AI solutions with Azure Cosmos DB for NoSQL | 252 | 3 | 24 |
| Develop AI solutions with Azure Database for PostgreSQL | 294 | 3 | 27 |
| Enhance AI solutions with Azure Managed Redis | 222 | 3 | 21 |
| Integrate backend services for AI solutions | 263 | 3 | 25 |
| Manage application secrets and configuration for AI solutions | 153 | 2 | 16 |
| Observe and troubleshoot apps on Azure | 161 | 2 | 18 |
| Total | 1,924 | 24 | 194 |
That total is 32 hours and 4 minutes. Read it as a floor, not as an answer. It is the time Microsoft budgets for moving through the pages. It does not include doing the 24 exercises properly. It does not include revision. And it excludes everything you have to learn first, if the audience profile above does not yet describe you. A reader who treats 32 hours as the preparation time and books the exam has been misled by a true number.
Those 194 units are not 194 lessons. Every module opens with an introduction, ends with a module assessment and a summary, and carries exactly one exercise. That is 24 of each, so 96 of the 194 units are fixed furniture. The teaching sits in the remaining 98.
One module of the 24 matched nothing. We could not tie Implement event messaging with Azure Managed Redis to any bullet on the published skills list. So we teach it in our Study Guide, and we wrote no exam questions from it. Read it once. Do not give it a week.
Three numbers, and only two of them mean anything
Microsoft's own surfaces give three different sizes for this training. They are worth separating, because two of them describe real things and the third is one of those two in different units:
- 32 hours and 4 minutes — the nine learning paths added up, as the catalogue publishes them. This is self-paced reading.
- Five days — the length of the instructor-led course, on the course page.
- 120 hours — the
duration_in_hoursvalue the catalogue returns for that same course.
The third is not a third estimate. It is the five days counted as raw calendar hours: 5 × 24 = 120. You can check that without taking our word for it. The same catalogue reports 96 for both AZ-104 and SC-500, and both of those are four-day courses. So there are two real numbers here, not three: about 32 hours of self-paced material, or five days in a classroom.
Twenty-four exercises, one in every module
Twenty-four exercises across 194 units is about one unit in eight, and they are spread evenly: one per module, with no exceptions. That ratio is the number we would show anyone planning AI-200 time.
Hands-on matters more here than on a fundamentals exam, for a reason printed on the exam's own page. The certification page says the exam is proctored, gives you 120 minutes, and warns that it may include interactive components. The skills list is written in verbs — build, deploy, configure, query, trace — and those verbs are what the exercises rehearse. Reading about a KEDA scale rule teaches you the words for it. Doing the exercise teaches you which field it goes in and what breaks when it is wrong. Skipping the 24 exercises does not remove an eighth of the work. It removes the part where you find out whether the other seven-eighths went in.
The free exam sandbox is worth an hour, and it is worth being clear about what it is. Microsoft links it from both the study guide and the certification page. It shows you the question types inside the same interface the real exam uses, and the subject matter in it is not AI-200. So use it early, so that nothing about the screen is new on the day, and do not record that hour as practice.
There is also no official practice assessment for this exam yet. The certification page states: "The Practice Assessment for this exam is not currently available." It adds that these usually arrive within eight weeks of an exam leaving beta and becoming generally available. That removes the readiness check most candidates build their last two weeks around, which is a scheduling problem worth solving early rather than discovering in week ten.
The reading is not shared out the way the marks are
The study guide publishes a weight band for each of the four exam areas, and the bands are not equal. Put our counted units beside them, and the shape of a plan follows.
We assigned each of the 24 official modules to exactly one exam area while building our question bank. Microsoft publishes no such mapping, so the pairing is ours, and so are the unit counts. The minutes are Microsoft's.
| Exam area | Minutes | Units | Share of the reading | Exam weight |
|---|---|---|---|---|
| Containerized solutions on Azure | 579 | 63 | 30% | 20–25% |
| AI solutions with Azure data management services | 768 | 72 | 40% | 25–30% |
| Connecting to and consuming Azure services | 263 | 25 | 14% | 20–25% |
| Security, monitoring and troubleshooting | 314 | 34 | 16% | 20–25% |
The area names are shortened here to fit the table; the study guide prints them in full.
Two rows behave differently from the other two. Service Bus, Event Grid and Functions are worth up to a quarter of the exam. Microsoft publishes 263 minutes for them: four hours and 23 minutes, or 14% of the reading. That is three modules, with three exercises between them. Containers are the reverse — 579 minutes, nearly a third of the reading, for a band that may be no larger. Follow the reading list evenly and you spend most of a working day on an area worth about a fifth of your score. You spend a single afternoon on an area worth as much.
That is the sentence to plan a Saturday around. Four and a half hours of official material stands behind up to a quarter of the marks. So read those 263 minutes slowly, do the exercises instead of skimming them, and come back to that path a second time. It is the short path, and finishing it quickly is the mistake.
So we would give those three connect-and-consume modules at least as many hours as the eight container modules, and spend most of those hours building rather than reading. The data-services area is the largest band on the exam and the largest block of material. That is the one place where following the reading order is exactly right.
One caution, because this is our arithmetic rather than Microsoft's. Comparing unit counts with weight bands assumes questions are spread evenly across a blueprint, and Microsoft does not say that. Use the table to rank your revision. Do not use it as a fact about the exam.
So how long does it take?
Microsoft publishes no preparation time for AI-200. Neither the study guide nor the certification page suggests one. What Microsoft publishes is the size of the material, and those are two different things.
Start from the floor. The official reading is 32 hours and 4 minutes. At an hour a day, and reading only, that alone is about four and a half weeks — before you have done a single exercise, sat one practice question, or revised anything. Any plan shorter than five weeks is not a plan for this material.
The ceiling is the part no published figure can give you, because it depends on how much of the audience profile already describes you. So here is our estimate, formed while building our AI-200 question bank from these same 24 modules, and tied to the count you made earlier:
- All seven items describe your current work. Eight to ten weeks at about an hour a day, with the hour weighted toward the exercises rather than the reading.
- Five or six of the seven. Twelve to sixteen weeks. Spend the first of those weeks on the missing skills, outside the learning path, before you start the path itself.
- Fewer than five of the seven. We will not give you a week count. A schedule cannot stand in for experience the exam assumes you already have. Build the missing skills on a real subscription first, and the plan above will then apply to you.
You can check that the first band is sane against Microsoft's own figure. Eight to ten weeks at an hour a day is 56 to 70 hours, and 32 of those are the official reading. That leaves roughly half the plan for the exercises, the revision and the practice questions, which is about the split we would want.
Those ranges are still our opinion. They are not measurements, and no honest article can measure your week. We give numbers at all because "it depends" is not a plan.
The last two weeks, without a practice assessment
Stop taking in new material with roughly two weeks left. Spend the time on the choices this blueprint keeps asking you to make, and on the clock.
The choices worth drilling are the ones where two services or two settings both look reasonable:
- An indexing policy against a consistency level, when the question is really about Request Unit cost.
- A pgvector indexing change against more compute, when the question is really about latency.
- Azure Service Bus against Azure Event Grid.
- Azure Key Vault against Azure App Configuration.
- KEDA scaling inside Container Apps against scaling the environment around it.
Then the clock. The session runs 120 minutes, it is proctored, and interactive components take longer than multiple choice. So practise against a timer instead of settling your pace on the day. AI-200 is offered in English only. If English is not your first language, the study guide says you can ask for an additional 30 minutes when an exam is not offered in your preferred language. That request is worth making in advance rather than on the morning.
One thing the reading cannot do is tell you whether it worked. Our AI-200 Complete Bundle is the practice exams, the Study Guide and the Quick Recap together, built from these same 24 official modules. Every question carries a link to the Microsoft Learn unit it came from, so a wrong answer sends you to the page that settles it.
The certification renews every year
Passing is not the end of the commitment, and a study plan is the right place to learn that rather than a reminder email a year later. Microsoft's associate certifications last a year, and renewing them is free. You get a six-month eligibility window for the renewal assessment. That assessment is unproctored and open book, it is shorter than the exam, and you can retake it as often as you need to before your certification expires. A pass moves the expiry out by a year.
So budget an hour a year after the exam, not only the weeks before it. That is the cheapest part of the whole plan, and the one people forget until an email tells them the deadline is close.
The week ranges, the study order, the module-to-area mapping and the unit counts above are ours, formed while building AI-200 material from these same official modules. The minutes are Microsoft's, taken from its own learning catalogue on 2026-08-18, and they measure reading rather than preparation. Microsoft publishes no preparation time for this exam. Independent study material. ciply.io is not affiliated with, endorsed by, or sponsored by Microsoft.
Practice AI-200 with real, original questions
Every answer explained and linked to the official page behind it.
Don't get surprised mid-prep.
If Microsoft changes AI-200 while you study, we email you what changed and what it means for your prep.
Only exam-change emails. No marketing. Unsubscribe any time.
Official sources used in this article
- Microsoft Certified: Azure AI Cloud Developer Associate (certification page — audience profile, exam length, languages, practice assessment status)
- Study guide for Exam AI-200: Developing AI Cloud Solutions on Azure (the four exam areas and their weightings)
- Course AI-200T00-A: Develop AI cloud solutions on Azure (the nine learning paths and the course duration)
- Exam scoring and score reports (the 1–1,000 scale, the 700 pass mark, unscored questions)
- Microsoft Certification renewal (annual free renewal, the six-month window, open-book assessment)
- Microsoft Learn catalog API — the nine AI-200T00 learning paths with their published durations
- Microsoft Learn catalog API — the course entries for AI-200T00, AZ-104T00 and SC-500T00 (the duration_in_hours field)