Foundry estate and lifecycle operations

Chapter 1 of 6

Session scope#

What we will do#

Objective. Answer "what AI do we actually run, and where" from one command instead of a portal tour.

Deploy a shared Azure Workbook for live review, then run one Azure Resource Graph query that returns every Azure AI account across the approved management groups with its kind, SKU, region, network setting, and tags. The report adds model deployments, checks their live lifecycle and retirement data, then lists account, model, Service Health, and Advisor findings for the named owners.

Why it matters#

Problem. Azure does not provide one governed, tenant-wide inventory of AI accounts, model deployments, and retirement signals. An account created outside the approved path can stay invisible until it appears on an invoice or in an incident.

Solution. One report reads the estate, matches deployed models with the live catalog, and gives the lifecycle owner the signals that need a keep, replace, or retire decision.

Boundaries#

This inventory applies the platform-tagging, model-lifecycle, and observability controls across the approved estate.

Azure Resource Manager stays authoritative for the estate and the shared workbook. The repository holds the scope decision, query definitions, workbook definition, and deployment template. The report is written to a path the operator chooses outside this repository, or printed to the console.

Resource Graph exposes the account, Service Health, and Advisor records. It does not expose model deployments, so the report reads them per account. The report reads the regional Models API to check lifecycle status and retirement dates. Billed cost stays with Microsoft Cost Management, and the observability team owns runtime telemetry.