What Is the AI-200 Exam?
AI-200 (Developing AI Cloud Solutions on Azure) is Microsoft's Associate-level certification for the Azure AI Cloud Developer Associate role. Per Microsoft's official audience profile, it validates contribution to all phases of implementing AI solutions on Azure with an emphasis on back-end services: requirements, design, development, deployment, security, and monitoring.
Unlike AI-102 or AI-103 — which test calling and orchestrating Azure AI services directly (vision, language, generative AI, agents) — AI-200 tests the platform underneath: how you host the containers, store and query the vectors, wire the messaging, and secure and observe the resulting system. If AI-103 is “can you build the AI app,” AI-200 is “can you build and run the infrastructure it depends on.”
You should sit AI-200 if you are a developer or platform engineer who is proficient in Azure and third-party SDKs, Azure data management services, Azure monitoring/troubleshooting, Azure messaging/eventing, vector databases, and Python — and who deploys containerized applications on Azure as part of the job.
Exam Snapshot
Exam code
AI-200
Level
Associate
Questions
~40–60
Duration
100 minutes
Passing score
700 / 1000
Cost
$165 USD
Domains
4
Renewal
Free annual online assessment
Domain Breakdown
Straight from Microsoft's official skills-measured outline — every bullet below is a tested sub-skill, not a paraphrase.
Domain 1: Develop containerized solutions on Azure
20–25%Implement container application hosting
- • Build, store, version, and manage container images with Azure Container Registry
- • Build and run images using Azure Container Registry Tasks
- • Deploy containers to Azure App Service, including environment variables and secrets configuration
Implement container-orchestrated solutions
- • Deploy applications to Azure Container Apps — environment configuration and revision management
- • Implement event-driven scaling with KEDA (Kubernetes Event-Driven Autoscaling) in Container Apps
- • Deploy and manage applications to AKS using manifest files
- • Monitor and troubleshoot AKS and Container Apps solutions via logs, events, and end-to-end connectivity
Domain 2: Develop AI solutions by using Azure data management services
25–30% — largestAzure Cosmos DB for NoSQL
- • Connect via the SDK and run queries
- • Optimize query performance and RU consumption with indexing policies and consistency levels
- • Store and retrieve embeddings; execute vector similarity search for semantic retrieval
- • Implement a change feed processor to detect and handle new or updated items
Azure Database for PostgreSQL
- • Connect and query via SDKs
- • Model schemas and indexing strategies — table design, appropriate data types
- • Optimize query latency and reduce pgvector compute overhead
- • Configure compute, memory, and storage resources for vector workloads
- • Run vector similarity search, including RAG patterns with metadata filters
- • Implement connection optimization for throughput and latency
Azure Managed Redis
- • Implement data operations — caching, expiration, invalidation
- • Implement vector indexing to enable similarity search
Domain 3: Connect to and consume Azure services
20–25%Event- and message-based AI solutions
- • Azure Service Bus — dead-letter queue handling, messages, topics, subscriptions
- • Azure Event Grid — filters, custom events, retries
Azure Functions
- • Build serverless APIs — triggers and bindings
- • Configure and deploy function apps
Domain 4: Secure, monitor, and troubleshoot Azure solutions
20–25%Secure Azure solutions
- • Azure Key Vault — secret rotation and retrieval
- • Azure App Configuration — store and retrieve app configuration information
Monitor and troubleshoot Azure solutions
- • Trace distributed systems with OpenTelemetry SDKs
- • Write KQL queries to analyze logs and metrics
Deciding between AI-200 and the related AI-103 beta-to-GA track? Read AI-103 vs AI-200: How the Two Azure AI Associate Exams Differ.
Where Candidates Actually Lose Points
Picking the wrong vector store
Cosmos DB, PostgreSQL/pgvector, and Azure Managed Redis can all run vector similarity search. The exam tests the constraint that decides between them: Cosmos DB when you need change-feed-driven ingestion and NoSQL flexibility; PostgreSQL/pgvector when the data is already relational and you need RAG with metadata filters; Managed Redis when the vectors are a cache in front of a slower source of truth, not the source of truth itself.
KEDA scale rules vs. fixed replica counts
Container Apps questions frequently describe unpredictable, bursty, or zero-to-thousands load and expect a KEDA-based scale rule (e.g., on Service Bus queue length) rather than a fixed min/max replica count or a custom polling script.
Dead-letter handling in Service Bus
Scenarios about poison messages or repeated processing failures expect you to recognize dead-letter queue configuration and inspection — not retry loops built in application code.
Least-privilege secrets access
Key Vault questions test rotation and retrieval configuration, and expect scoped access policies or RBAC roles rather than broad Contributor-level access to the vault.
Reaching for KQL, not just app logs
Monitor/troubleshoot questions expect you to write or interpret a KQL query against Log Analytics rather than describing a generic "check the logs" answer — know basic operators like where, summarize, and render.
4-Week AI-200 Study Plan
Weighted toward Week 2 because the data-management domain alone is 25–30% of the exam — larger than any single other domain.
- • Build and version images with Azure Container Registry; run ACR Tasks
- • Deploy the same image to App Service, Container Apps, and AKS via manifest — compare the operational model of each
- • Configure a KEDA Service Bus scale rule on a Container Apps job and test scale-to-zero
- • Break something on purpose (bad manifest, wrong revision) and practice diagnosing it from logs and end-to-end connectivity checks
- • Cosmos DB: connect via SDK, tune an indexing policy, compare consistency levels, then store embeddings and run a vector similarity query
- • Build a change feed processor that reacts to new/updated Cosmos DB items
- • PostgreSQL: design a schema with pgvector, run vector similarity search, then build a metadata-filtered RAG query
- • Azure Managed Redis: implement caching with expiration/invalidation, then add vector indexing for similarity search
- • Compare all three side by side — this comparison is the single highest-leverage study activity for AI-200
- • Service Bus: queues, topics/subscriptions, and dead-letter queue handling end to end
- • Event Grid: custom events, event filters, and retry behavior
- • Azure Functions: build a serverless API with an HTTP trigger and a Service Bus binding, then deploy the function app
- • Key Vault: rotate a secret, scope retrieval with least privilege
- • App Configuration: externalize settings out of a function app or container
- • Instrument a service with OpenTelemetry SDKs and write KQL queries against the resulting logs/metrics
- • Two full-length mock exams (100 minutes each); review every miss and re-test the weakest domain the next day
Frequently Asked Questions
What is the AI-200 exam?+
AI-200 — officially Developing AI Cloud Solutions on Azure — is Microsoft’s Associate-level exam for the Azure AI Cloud Developer Associate certification. It targets the back-end engineering behind AI solutions: containerized hosting, AI-ready data services, event- and message-driven integration, and securing/monitoring the result. Skills measured were last updated 2026-04-15.
How is AI-200 different from AI-102 or AI-103?+
AI-102 and AI-103 test how you call and orchestrate Azure AI services directly — vision, language, generative AI, agents. AI-200 zooms out to the platform underneath: Container Apps and AKS hosting, Cosmos DB/PostgreSQL/Redis as vector stores, Service Bus and Event Grid integration, and securing the whole stack. Many teams need both roles; AI-200 is the backend/platform-engineering half.
What is the single largest domain on AI-200?+
Develop AI solutions by using Azure data management services at 25–30% — the only domain in that range. It covers Cosmos DB for NoSQL (vector search, change feed, RU tuning), Azure Database for PostgreSQL with pgvector (schema design, RAG patterns, connection optimization), and Azure Managed Redis (caching and vector indexing). Under-preparing this domain is the most common reason candidates fail.
How many questions are on AI-200 and what is the passing score?+
AI-200 runs 100 minutes with roughly 40–60 questions, scored on Microsoft’s 0–1000 scale where 700 is a pass. That is not a flat 70% — scenario questions in the data-management domain typically carry more weight than single-fact recall.
Do I need Python for AI-200?+
Yes. The official audience profile lists Python programming as a required proficiency alongside Azure SDKs. Expect to read short SDK snippets against Cosmos DB, PostgreSQL, and Service Bus client libraries — you are not asked to write large programs, but you must recognize correct client configuration and query patterns.
What is the hardest part of AI-200?+
Choosing the right data store for a vector-search scenario. Cosmos DB, PostgreSQL with pgvector, and Azure Managed Redis can all store embeddings and run similarity search, and the exam consistently presents a scenario where two of the three look plausible. The deciding constraint is usually consistency/latency requirements, existing relational schema, or cache-vs-source-of-truth positioning — not raw vector-search capability.
How long should I study for AI-200?+
Candidates already comfortable with Container Apps, AKS, and an Azure SDK language should plan 4–6 weeks, weighted toward the data-management domain. Candidates newer to Azure application development should plan 10–12 weeks, starting with the official Microsoft Learn path before layering on scenario practice.
What are the prerequisites for AI-200?+
None are enforced at registration, but Microsoft’s official audience profile expects proficiency in Azure and third-party SDKs, Azure data management services, Azure monitoring and troubleshooting, Azure messaging and eventing, vector databases, Python, and containerized application deployment on Azure.
Is AI-200 worth taking alongside AI-103?+
For most backend/platform engineers, yes. AI-103 proves you can build the application layer against Azure AI Foundry; AI-200 proves you can build and secure the infrastructure that layer runs on — containers, vector data stores, messaging, observability. See our AI-103 vs AI-200 comparison for a full side-by-side.
How is MSCertQuiz different from other AI-200 practice questions?+
AI-200 is a new exam and most "practice questions" circulating online are generic Azure-developer trivia that never mentions Cosmos DB vector search, pgvector, Azure Managed Redis vector indexing, or KEDA autoscaling by name. MSCertQuiz has 500 AI-200 questions built directly against the four official domains, with architectural explanations for every answer.
Ready to test this against real scenarios?
500 AI-200 practice questions across all four domains — 40 free, no card required.