Skip to content

AI Citadel Governance Hub sizing guide

This guide provides recommendations for sizing the AI Citadel Governance Hub based on various deployment scenarios. Proper sizing ensures optimal performance, scalability, and cost-efficiency.

The approach is based on T-Shirt Sizing Methodology, which categorizes deployments into Small, Medium, and Large sizes based on expected workloads and user counts.

NOTE: Below estimates are based on typical usage patterns and may vary based on specific customer requirements and workloads. It is advisable to conduct a detailed assessment for precise sizing.

Governance Hub Environment Sizing

As a centralized supplemental landing zone for AI workloads, the AI Citadel Governance Hub can be sized based on the number of environments required by the customer. Common configurations include: • One central for production and non-production • Two with one for production and one for non-production • Other: based on customer requirements (like have 3 or more environments)

T-Shirt Sizing Categories

Small: development and experimental

No SLA – development only. Can be multiple environments per customer. Full network isolation supported (Experimental, low load) https://azure.com/e/b23669e7f59948c488608ecab4b7c1fa

Medium Classic: non-production with SLA (or minimum production)

Minimum-production or non-production with SLA. Can be multiple environments per customer. Full network isolation supported (uses APIM Premium) (Estimated to handle ~300-500 PTU) https://azure.com/e/7a539b8ef73a4b5e87b5e60c079f0835

Medium v2: non-production with SLA (or minimum production)

Minimum-production or non-production with SLA. Can be multiple environments per customer. Full network isolation supported (uses APIM Standard V2) (Estimated to handle ~200 PTU per APIM unit) https://azure.com/e/b44716c27c054c4eb979248744bc341d

Large:

Multi-zone production setup with SLA. Full network isolation supported (Estimated to handle ~1,000-1,500 PTU) https://azure.com/e/fed48a9999804396a7e60d31550f24bf

Considerations for sizing

When determining the appropriate size for the AI Citadel Governance Hub, consider the following factors:

  • Above calculation excludes the sizing for the agentic applications (which would be sized separately based on their own requirements)
  • LLM and other AI services which should be sized based on the expected usage patterns and workloads.
  • Medium and Medium V2 tiers can be scaled out by adding additional APIM units to handle increased load as per customer requirements.

SKUs implementation

Based on the selected T-Shirt and environment, updating the SKUs in the deployment main.bicepparam under the COMPUTE SKU & SIZE section:

// ============================================================================
// COMPUTE SKU & SIZE - SKUs and capacity settings for services
// ============================================================================
param apimSku = readEnvironmentVariable('APIM_SKU', 'StandardV2')
param apimSkuUnits = int(readEnvironmentVariable('APIM_SKU_UNITS', '1'))
param eventHubCapacityUnits = int(readEnvironmentVariable('EVENTHUB_CAPACITY', '1'))
param cosmosDbRUs = int(readEnvironmentVariable('COSMOS_DB_RUS', '400'))
param logicAppsSkuCapacityUnits = int(readEnvironmentVariable('LOGIC_APPS_SKU_CAPACITY_UNITS', '1'))
param apicSku = readEnvironmentVariable('APIC_SKU', 'Free')
param keyVaultSkuName = readEnvironmentVariable('KEY_VAULT_SKU_NAME', 'standard')

NOTE: Above is the default parameters configurations. It needs to be adjusted based on the selections made based on final sizing.

© 2026 Azure AI Landing Zones