Azure ML Serverless Endpoint
Does ZopNight manage Azure ML Serverless Endpoint?
Azure ML serverless endpoints serve foundation and catalog models with pay-per-token billing and 0 provisioned instances, so an unused endpoint costs nothing. ZopNight inventories them through the AML enricher and attributes usage spend from Cost Management, because token-metered costs scale with adoption rather than with idle time.
Rules that fire on Azure ML Serverless Endpoint
No active rule family targets Azure ML Serverless Endpoint today. Rules that used to are retired, and retired rules publish no pages and fire no findings. Scheduling and permissions coverage are unaffected.
At a glance
| Field | Value |
|---|---|
| Scheduling notes | discovery and cost visibility only; there is no provisioned capacity to stop. |
Serverless endpoints serve foundation and catalog models with pay-per-token billing and no provisioned instances. There is no idle cost, but usage can grow quickly with adoption.
Tokens in, tokens out: the only meter
A serverless endpoint charges for what flows through it: tokens consumed and generated against a catalog model, with no instances provisioned and no hourly floor. The idle-waste class that dominates the rest of the ML surface simply does not exist here. An endpoint nobody calls costs nothing. What replaces it is a usage-shaped risk: token spend tracks adoption, and adoption compounds. The model that cost pennies in a prototype can become a five-figure line once three products integrate it.
Attribution for pay-per-use model serving
Discovered via the AML enricher for inventory, with usage spend attributed through Cost Management billing. Inventory matters more than it sounds: serverless endpoints are cheap to create, so they proliferate across workspaces, and the first step of governing token spend is a complete list of which endpoints exist and which model each fronts.
Why there is no schedule to apply
With no provisioned capacity to stop, scheduling is meaningless: a stopped pay-per-token endpoint would save exactly nothing. ZopNight scopes this type to discovery and cost visibility, and the cost conversation shifts from “when should this be off” to “which callers are growing and should they be.”
How token spend gets away from teams
The recurring patterns: retry logic amplifying every failure into multiplied token charges; verbose prompts and long contexts inflating per-call cost invisibly; and prototype endpoints quietly adopted by production code without anyone re-forecasting the volume. None are visible in infrastructure inventories. They surface in the spend attribution.
Serverless endpoints in the studio
Azure ML studio → Endpoints → Serverless endpoints lists each endpoint and its model; pair the list with attributed spend to see which of them are becoming products.