AI-200 exam guide: Azure AI Cloud Developer Associate

AI-200 (Developing AI Cloud Solutions on Azure) is the exam for Microsoft Certified: Azure AI Cloud Developer Associate, and it is the replacement for AZ-204, the Azure Developer exam that retired on 31 July 2026. Despite the name, it is a back-end cloud developer exam: containers, data services with vector search, messaging and monitoring, not model training or prompt engineering. It has been generally available since 9 July 2026 and costs US$165, or ₹4,865 in India, as of November 2026.

This guide covers what AI-200 replaced and what that means if you hold the old certification, what each of the four skill areas tests, how it differs from AI-103 (the exam for building AI apps and agents), the format and price, and how to prepare. Facts were checked on Microsoft Learn on 4 November 2026.

AI-200 at a glance

Exam: AI-200: Developing AI Cloud Solutions on Azure.
Certification: Microsoft Certified: Azure AI Cloud Developer Associate.
Status: live. The beta was announced in May 2026, the exam became generally available on 9 July 2026, and Microsoft refreshed the skills measured on 7 October 2026. No retirement date is announced.
Who it's for: back-end Azure developers who build, deploy, secure and monitor the cloud services that AI applications run on.
Prerequisites: no prerequisite certification is listed. Microsoft expects you to know Python, the Azure SDKs, data services, messaging, vector databases, containers and monitoring.
Length: 120 minutes.
Passing score: 700 on Microsoft's 1-1,000 scale.
Price: US$165, and ₹4,865 in India, as of November 2026; tax may be extra.
Delivery: Pearson VUE, online or at a test centre, in 13 languages.
Validity: one year, renewed for free with an online assessment.

AI-200 vs AZ-204: what it replaced

Microsoft retired AZ-204 (Developing Solutions for Microsoft Azure) and its certification, Azure Developer Associate, on 31 July 2026, three weeks after AI-200 went live. Microsoft's own announcement pairs the two: Azure Developer Associate is replaced by Azure AI Cloud Developer Associate. AI-200 is a new exam, not a renamed one. There is no upgrade exam and no automatic transfer, so to hold the new certification you pass AI-200.

The content overlaps more than the name suggests. Both exams are about building on Azure's platform services: compute, storage, messaging and security. What AI-200 adds is the data layer that AI applications need, above all storing and searching embeddings.

If you hold Azure Developer Associate

Your certification stays on your Microsoft Learn transcript until it expires. What you can't do is renew it, because the renewal assessment retired with the exam. Once it expires it moves to the history section of your transcript. If you want a current developer certification, plan to take AI-200 before your old one lapses; your AZ-204 knowledge of App Service, Functions, Cosmos DB and Key Vault carries straight over.

AI-200 and the AZ-400 prerequisite

Many people searching for AI-200 want to know whether it counts toward Microsoft Certified: DevOps Engineer Expert, since AZ-204 used to. As of November 2026 the DevOps Engineer Expert page lists only Azure Administrator Associate and Azure Developer Associate, so AI-200 does not count yet. If AZ-400 is your goal, the AZ-400 exam guide explains the routes that work today.

What the AI-200 exam tests

The study guide splits the exam into four areas. Microsoft gives each a weight range, and the October 2026 update kept data management the largest:

Develop containerized solutions on Azure: 20-25%.
Develop AI solutions by using Azure data management services: 25-30%.
Connect to and consume Azure services: 20-25%.
Secure, monitor, and troubleshoot Azure solutions: 20-25%.

Containerized solutions (20-25%)

This area is about getting code into containers and running it on Azure: building and storing images in Azure Container Registry (including ACR Tasks), and deploying to App Service, Azure Container Apps with KEDA-based scaling, and Azure Kubernetes Service. Expect questions on picking the right host for a workload and on scaling rules. A practice question on building images when the registry is behind a private endpoint from the AI-300 bank shows the kind of networking detail that trips people up.

Azure data management (25-30%)

The biggest area, and the one where "AI" earns its place in the title. It covers Azure Cosmos DB for NoSQL (request units, the change feed, vector search), Azure Database for PostgreSQL with the pgvector extension, and vector indexing in Azure Managed Redis. You need to know how to store embeddings, query them for similarity, and keep throughput and cost under control. A practice question on where agent thread history is stored from the AI-103 bank shows Cosmos DB playing exactly this role behind an AI app.

Connecting Azure services (20-25%)

This is classic back-end integration: Azure Service Bus for queues and topics, Azure Event Grid for events, and Azure Functions for code that runs in response to them. The questions ask which service fits a pattern (a reliable work queue, a fan-out of events, a scheduled job) and how to wire them together. A practice question on triggering retraining from Event Grid events is a good example of choosing events over polling.

Security, monitoring and troubleshooting (20-25%)

The last area covers Azure Key Vault and Azure App Configuration for secrets and settings, managed identities, and observability with OpenTelemetry and KQL queries. Two practice questions here cover the security ideas well: one on keyless authentication with a managed identity, and one on customer-managed key requirements in Key Vault. For monitoring, a practice question on diagnostic settings and log retention covers routing logs to Log Analytics.

The "AI" in the name: what's not on the exam

The name misleads a lot of candidates. There is no domain on training models, writing prompts or building agents in Microsoft Foundry. AI-200 tests the cloud plumbing that an AI application depends on: where it runs, where its data and vectors live, how its parts talk to each other, and how it's secured and watched. If you came looking for a generative AI developer certification, AI-103 is the one you want.

That also makes AI-200 a sensible exam for developers who have no interest in AI features themselves. If you build back-end services on Azure, most of the syllabus is your daily work, with vector search added.

AI-200 or AI-103: which one to take

The two associate exams sit side by side, and the split is clean:

What you build: AI-103: the AI app or agent itself, in Python with Microsoft Foundry. / AI-200: the cloud services under it (containers, databases, messaging, security).
Biggest area: AI-103: generative AI and agents (30-35%). / AI-200: Azure data management (25-30%).
What it replaced: AI-103: AI-102, retired 30 June 2026. / AI-200: AZ-204, retired 31 July 2026.
Format and price: both run 120 minutes, need 700 to pass and cost US$165 (₹4,865 in India) as of November 2026.
Practice questions here: AI-103 has a full bank at AI-103 practice questions; AI-200 has no bank yet, so this guide links neighbouring questions instead.

Choose by role. If you write the code that calls models and agents, take AI-103; the AI-103 exam guide covers it, and the AI-103 cert page lists its tests. If you build and run the services those apps sit on, take AI-200. Full-stack developers on Azure often end up taking both, usually AI-103 first if the job title says "AI engineer", AI-200 first if it says "back-end" or "cloud developer". For the wider picture, including AI-901, AI-300 and the expert exams, see the Microsoft AI certification path.

Format, price and renewal

AI-200 runs for 120 minutes. Microsoft doesn't publish a question count for it; its general guidance is that most exams have between 40 and 60 questions, and this one may include interactive components. You need 700 out of 1,000, and because the score is scaled, that isn't the same as 70% of the questions.

The price is US$165 in the United States and ₹4,865 in India as of November 2026, set by the country where you take the exam, with tax on top in some places. You book through Pearson VUE and choose either online proctoring or a test centre.

The certification lasts one year. Renewal is free: you pass an online assessment on Microsoft Learn before it expires. If you let it lapse, you take AI-200 again.

How to prepare for AI-200

A preparation plan that works for most back-end developers:

Read the official study guide for AI-200 on Microsoft Learn and mark each skill as known, rusty or new. The weights above tell you where to spend time.
Take the free course. Microsoft's AI-200T00 course has free self-paced modules on Microsoft Learn covering all four areas.
Build the data layer by hand. Store embeddings in Cosmos DB for NoSQL and in PostgreSQL with pgvector, and run similarity queries against both. This is the part AZ-204 veterans know least, and it's the biggest area.
Wire up the integration services. One small project that drops messages on Service Bus, reacts to Event Grid events with a Function, and runs in Container Apps covers two areas at once.
Test yourself. The free practice assessment now runs on AI Skills Navigator, and the exam sandbox shows you the question formats. Here, the AI-300 practice questions cover containers, identity and monitoring on Azure in depth, and the AI-300 cert page explains that exam. A practice question on least-privilege workspace identities is a good check of your RBAC knowledge.

Someone with AZ-204-level experience can usually be ready in three to five weeks, most of it spent on the data and container areas. Someone new to Azure development should allow longer and build every service at least once.

Frequently asked questions

What did AI-200 replace?

AI-200 replaced AZ-204 and its certification, Azure Developer Associate, which retired on 31 July 2026. It's a new exam with its own certification, Azure AI Cloud Developer Associate, so there's no automatic transfer from the old one.

Is AI-200 harder than AZ-204?

It covers much of the same ground (App Service, Functions, Cosmos DB, Key Vault, messaging) but adds containers on Container Apps and AKS and a large vector-data section. For an experienced Azure developer, the new data material is the hard part; the rest is familiar.

Should I take AI-200 or AI-103?

Take AI-103 if you build AI apps and agents in Python with Microsoft Foundry. Take AI-200 if you build the back-end services they run on: containers, databases with vector search, messaging and monitoring. Both cost the same and run the same length.

How much does AI-200 cost in India?

₹4,865 as of November 2026, against US$165 in the United States. Microsoft sets the price by the country where you take the exam, and tax may be added.

Do I need Python for AI-200?

Yes. Microsoft lists Python and the Azure SDKs among the skills it expects, and the exam assumes you can read and write code that uses them.

Is AI-200 still in beta?

No. The beta ran from May 2026, and AI-200 has been generally available since 9 July 2026. Its skills measured were last updated on 7 October 2026.