How do you track Azure OpenAI model retirements across subscriptions?

November 16, 2026 5 answers
Asked by KROVEX+6n

We have model deployments across multiple subscriptions, and the subscription owner isn't always the team responsible for the application. How do you spot retirements, get notices to the right teams, and make sure upgrades happen?

Answers

q.593!orbitcabin

Start with the retirement events rather than maintaining a separate calendar by hand. The Azure Resource Health events API and the Azure Resource Graph service retirement impact queries can provide candidates for a central report across subscriptions.

Match those candidates to your deployment inventory, then check the retirement details before treating every result as a confirmed impact. A workbook could make the report easier to review, while Service Health alerts and action groups can help notify people. Keep the report refreshed because retirement dates can move.

_u64+velvetladle.9:

Access across the subscriptions matters here. A central report is only as complete as the scope its queries can see.

Reply to this answer
u.614!parcelspanner

The routing is the hard part. A notice sent to a subscription owner doesn't help much if that person can't identify the application team.

Put an owning team and a contact channel on deployments or in an inventory that links deployments to workloads. Azure Policy can help enforce tags where the resource type supports them. Then route the retirement report to that contact, with a fallback for missing or stale ownership information.

KROVEX+6n:

That fallback sounds useful. A shared operations mailbox could catch notices that don't have a clear owner yet.

Reply to this answer
_d82-Rivlo

Detection and remediation are different problems. Send the team a notice, but also track whether they've acknowledged it and have an upgrade plan. If the deadline gets close without action, escalate through whoever owns the subscription or pays for the workload.

Reply to this answer
_m75+switchladle.3

Don't assume you need a bespoke retirement tracker straight away. CI checks and model evaluations can reveal when a model is unavailable, especially in staging. That catches a different failure than an advance notice, though, so it won't necessarily give a team enough time to plan a migration.

Reply to this answer
_n83!copperstaple.6

A lightweight option is to forward Azure Advisor information to a shared mailbox using a small function, so operations can see it without relying on one subscription owner's inbox. It's a useful safety net, but someone still needs to identify the affected workload and follow up.

Reply to this answer

Add your answer

Answers are reviewed before publication.

Related Discussions