An Azure service that provides access to OpenAI’s GPT-3 models with enterprise capabilities.
Hello @YUJIEYANGBeyondsoft-7976
Thank you for reaching out to Microsoft Q&A.
Migrating an AI workload from an EA subscription to a CSP subscription is possible in some scenarios, but it should not be assumed that all Azure AI resources and their dependencies can be moved seamlessly without any changes.
Azure resource moves are supported on a resource-type basis. For example, Azure Cognitive Services accounts support subscription moves, and Azure AI Search supports subscription moves, although some services may require additional steps. Microsoft also notes that when resources are moved between subscriptions, the resource ID changes, and dependencies, role assignments, policies, and other configuration may need to be updated or reapplied. Both subscriptions must also be associated with the same Microsoft Entra tenant for a standard cross-subscription move.
For an environment containing Azure OpenAI, Azure AI Search, Azure AI Speech, Microsoft Foundry projects, Agents and Connectors, we recommend validating each resource and its dependencies individually. In particular, Foundry projects, agents, connections/connectors, managed identities, RBAC, networking/private endpoints, model deployments, and application configuration should be reviewed before the migration.
There is also an important CSP-specific consideration for Foundry Models. Microsoft currently documents that CSP subscriptions cannot purchase third-party model offerings through the Foundry partner/community model experience. Therefore, if the existing EA environment uses third-party/partner models that require Azure Marketplace, those deployments may not be reproducible in the target CSP subscription. Microsoft recommends using models offered as first-party Azure consumption services in that scenario.
Recommended migration approach
- Inventory all resources and dependencies in the EA subscription.
Verify whether each resource type supports a cross-subscription move.
Identify all Foundry model deployments and determine whether they are Microsoft/Azure-sold models or third-party/partner models.
For resources that support a direct move, perform the subscription migration and validate dependent configurations.
For resources or deployments that cannot be moved, recreate them in the CSP subscription and migrate the required configuration/data.
Revalidate RBAC, managed identities, private endpoints, networking, application settings, resource IDs, endpoints and deployment names.
Test the complete application and agent workflow in the CSP subscription before retiring the EA resources.
Users can continue using the workload after the migration provided that the required resources, deployments, permissions and dependencies are successfully recreated or moved. However, we cannot guarantee a completely transparent migration with no configuration changes.
if the applications reference Azure resource IDs directly, those references must be updated because the subscription ID forms part of the resource ID and therefore changes after a subscription move.
Please refer this Microsoft documentation:
Move Azure resources to a new resource group or subscription
Azure resource types that support move operations
- Microsoft Foundry Models from partners and community
- Model inference SDK migration checklist & troubleshooting (endpoint/scopes/model-not-found): https://learn.microsoft.com/azure/foundry/how-to/model-inference-to-openai-migration
I Hope this helps. Do let me know if you have any further queries.
If this answers your query, please do click Accept Answer and Yes for was this answer helpful.
Thank you!