Study guide
1.2 Resource Provisioning and Configuration
Azure AI Engineer Associate AI-102
Track your progress as you read
Finished sections get a check mark.
1 Introduction
About the Azure AI Engineer Associate (AI-102) Exam Study Strategies and Exam Tips for the AI-102
2 Domain 1: Plan and Manage an Azure AI Solution (20-25%)
1.1 Azure AI Services Overview and Selection 1.2 Resource Provisioning and Configuration 1.3 Security and Access Management 1.4 Responsible AI Principles and Implementation 1.5 Azure AI Solution Architecture Patterns
3 Content Safety and Moderation (within Plan and Manage, Domain 1)
2.1 Azure AI Content Safety Overview 2.2 Prompt Shields and Adversarial Attack Detection 2.3 Text and Image Moderation Implementation 2.4 Content Filtering in Azure OpenAI Service
4 Domain 4: Implement Computer Vision Solutions (10-15%)
3.1 Azure AI Vision — Image Analysis 4.0 3.2 Azure AI Face Service 3.3 Azure AI Custom Vision 3.4 Optical Character Recognition (OCR) 3.5 Video Analysis and the Retired Spatial Analysis
5 Domain 5: Implement Natural Language Processing Solutions (15-20%)
4.1 Azure AI Language — Text Analytics Features 4.2 Conversational Language Understanding (CLU) 4.3 Custom Text Classification and Custom NER 4.4 Azure AI Speech Services 4.5 Azure AI Translator Service 4.6 Question Answering and Conversational AI
6 Domain 6: Implement Knowledge Mining and Information Extraction Solutions (15-20%)
5.1 Azure AI Search — Fundamentals and Architecture 5.2 AI Enrichment with Skillsets 5.3 Azure AI Document Intelligence
7 Domain 2: Implement Generative AI Solutions (15-20%)
6.1 Azure OpenAI Service — Models and Deployment 6.2 Chat Completions API and Prompt Engineering 6.3 Embeddings and Vector Search for RAG 6.4 Fine-Tuning and Model Customization 6.5 Image Generation (gpt-image; DALL-E 3 Retired) 6.6 RAG After On Your Data, and AI Agents
8 Domain 3: Implement an Agentic Solution (5-10%)
3.1 Agentic Solutions: Foundry Agent Service and the Microsoft Agent Framework
9 Exam Review: Cross-Domain Topics and Advanced Practice
7.1 Service Comparison and Selection Guide 7.2 SDK and REST API Patterns 7.3 Responsible AI Across All Domains 7.4 End-to-End Solution Architectures 7.5 Exam Day Quick Reference
1.2 Resource Provisioning and Configuration
Key Takeaways Explain This Topic AI
Provision AI resources via the portal, Azure CLI (az cognitiveservices account create), PowerShell, or infrastructure-as-code (ARM, Bicep, Terraform).
The --kind flag selects the service type: AIServices is the Microsoft Foundry multi-service resource (CognitiveServices is the older multi-service kind), ComputerVision/TextAnalytics/SpeechServices are single-service, and OpenAI is standalone Azure OpenAI.
F0 (Free) has no SLA and tight rate limits; S0 (Standard) carries the 99.9% SLA — production scenarios always need S0 or higher.
A customSubDomainName is mandatory for Microsoft Entra ID token authentication; without it you are stuck on key auth.
Two keys per resource exist specifically to enable zero-downtime key rotation.
Last updated: October 2026
Quick Answer: Provision Azure AI resources through the portal, CLI, PowerShell, or IaC (ARM/Bicep/Terraform). Each resource has an endpoint URL plus two keys (or Entra ID auth). Choose F0 only for dev (no SLA, hard rate limits) and S0 for production (99.9% SLA). Set a customSubDomainName if you want token-based authentication.
Provisioning methods you must recognize
The exam shows CLI snippets and ARM/Bicep fragments and asks what they do or which flag to change. Learn the verbs and the --kind values.
Azure portal
Create a resource -> AI + Machine Learning -> pick the service (e.g. "Foundry", kind AIServices, for multi-service).
Set Subscription, Resource Group, Region , Name, and Pricing tier .
Review + Create.
Azure CLI az group create --name rg-ai-prod --location eastus
- # Multi-service resource (one endpoint, many services)
- az cognitiveservices account create \
- -name my-ai-services --resource-group rg-ai-prod \
- -kind CognitiveServices --sku S0 --location eastus \
- -custom-domain my-ai-services --yes
- # Single-service Vision resource
- az cognitiveservices account create \
- -name my-vision --resource-group rg-ai-prod \
- -kind ComputerVision --sku S1 --location eastus --yes
- # Read the endpoint and keys
- az cognitiveservices account show -n my-ai-services -g rg-ai-prod --query properties.endpoint
- az cognitiveservices account keys list -n my-ai-services -g rg-ai-prod
- ARM and Bicep (infrastructure as code) {
- "type": "Microsoft.CognitiveServices/accounts",
- "apiVersion": "2023-05-01",
- "name": "my-ai-services",
- "location": "eastus",
- "kind": "CognitiveServices",
- "sku": { "name": "S0" },
- "properties": { "publicNetworkAccess": "Enabled", "customSubDomainName": "my-ai-services" }
- } resource aiServices 'Microsoft.CognitiveServices/accounts@2023-05-01' = {
- name: 'my-ai-services'
- location: 'eastus'
- kind: 'CognitiveServices'
- sku: { name: 'S0' }
- properties: { publicNetworkAccess: 'Enabled', customSubDomainName: 'my-ai-services' }
- }
- Pricing tiers and SLA
- Tier SKU Use case SLA Rate limit (typical)
- Free F0 Dev/test only None Low (e.g. ~20 calls/min)
- Standard S0 Production 99.9% Higher (service-dependent)
- Higher SKUs S1+ High throughput 99.9% Highest, per-service quotas
A common trap: a free tier often allows only one F0 instance of a given kind per subscription . If a deployment fails citing an existing free resource, that is the cause — not a region or quota problem.
On the Exam: Any production scenario requiring an availability guarantee => S0 or higher, because F0 carries no SLA. If a scenario stresses cost during development and ignores SLA, F0 is acceptable.
Endpoint and authentication components
Every resource exposes a small set of values you wire into clients:
- Component What it is Example
- Endpoint Base URL for API calls https://my-ai-services.cognitiveservices.azure.com/
Key 1 / Key 2 Two interchangeable auth keys a1b2c3... / g7h8i9...
- Resource ID ARM identifier for RBAC scoping /subscriptions/.../accounts/my-ai-services
- Custom subdomain Required for Entra ID token auth my-ai-services # Key-based auth (dev)
- from azure.core.credentials import AzureKeyCredential
- from azure.ai.vision.imageanalysis import ImageAnalysisClient
- client = ImageAnalysisClient(endpoint, AzureKeyCredential(key))
- # Entra ID token auth (production) -- requires customSubDomainName
- from azure.identity import DefaultAzureCredential
- client = ImageAnalysisClient(endpoint, DefaultAzureCredential())
A subtle exam fact: token (Entra ID) authentication only works if the resource has a custom subdomain . The regional *.cognitiveservices.azure.com shared endpoint supports key auth only. If a question says "we enabled managed identity but token auth fails," the missing custom subdomain is the usual culprit.
Zero-downtime key rotation
Two keys exist for exactly one reason: rotate without an outage.
App authenticates with Key 1 .
Regenerate Key 2 — app is unaffected (still on Key 1).
Switch the app to Key 2 .
Regenerate Key 1 — app is unaffected (now on Key 2).
On the Exam: The correct rotation sequence always moves traffic to the other key before regenerating the key in use. Regenerating the active key first causes the very outage the two-key design prevents.
Regional availability and data residency
Not every service lives in every region. Azure OpenAI model availability in particular varies by region and model version. Choose a region that satisfies data residency/compliance (EU data in EU regions), then optimize for latency (closest to users), and plan DR around Azure paired regions .
Resource naming and organization also surface in IaC questions. A multi-service resource and a single-service resource of the same kind cannot share a name within a subscription, and the customSubDomainName must be globally unique because it becomes the host portion of the endpoint URL. Tag resources consistently (environment, owner, cost-center) so Azure Policy and cost reports can group them, and keep each environment (dev, test, prod) in its own resource group so RBAC and lifecycle operations stay isolated.
Reading provisioning questions
Many Domain 1 items hand you a CLI command or a Bicep block and ask one of three things: what does it create, which one line is wrong, or which property to change to meet a new requirement. Train yourself to scan four fields first — --kind / kind (which service), --sku / sku.name (F0 vs S0, hence SLA and rate limits), location (residency and service availability), and customSubDomainName / --custom-domain (whether token auth will work). If a question adds "must support Microsoft Entra ID authentication" to an otherwise valid template, the fix is almost always adding the custom subdomain.
If it adds "must guarantee availability," the fix is changing F0 to S0. If it adds "data cannot leave Germany," the fix is the location . Recognizing which single property each new sentence targets is the entire skill these questions test.
On the Exam: Free tier limits and the one-F0-per-kind-per-subscription rule explain most "the deployment failed" scenarios. Before suspecting quotas or regions, check whether a free instance of that kind already exists.