Azure OpenAI Account
Does ZopNight manage Azure OpenAI Account?
Azure OpenAI accounts carry no charge of their own. The money sits in what runs beneath them: PTU deployments billing hourly for reserved capacity, per-token Standard deployments, and vector stores accruing daily storage. ZopNight expands each account through the data plane into deployments, fine-tuning jobs, batch jobs, and vector stores.
Rules that fire on Azure OpenAI Account
At a glance
| Field | Value |
|---|---|
| Scheduling notes | discovery and cost visibility only. |
Azure OpenAI accounts host GPT, embedding, and image model deployments under Azure’s enterprise controls. Provisioned-throughput (PTU) deployments inside them carry large fixed hourly costs.
Where the spend actually sits in an OpenAI account
The account is a container, and containers are cheap. What bills is the resources it holds and the tokens that flow through them. Standard deployments meter per token consumed. Provisioned-throughput deployments reserve PTU capacity and bill for every hour of that reservation, traffic or not, which makes them the single largest line most teams see. Around the deployments sit smaller meters that add up: vector stores charging daily for stored embeddings, fine-tuned models carrying hourly hosting fees, and batch jobs settling per-token charges at completion. Judging an account by its top-level resource entry therefore tells you almost nothing; the cost story lives one level down.
Expanding the account through the data plane
An ARM inventory stops at the account, so ZopNight goes further: the Azure OpenAI enricher authenticates against the Cognitive Services data plane and enumerates what lives inside: deployments, fine-tuning jobs, batch jobs, and vector stores each become their own tracked resource beneath the account. Spend is attributed from Cost Management, and recommendation rules flag idle PTU capacity, the highest-value finding this type produces. Access for the enumeration uses the Cognitive Services OpenAI User role, a read-level data-plane role, not a subscription-wide one. The account itself is not schedulable; there is nothing to stop at this level, so ZopNight’s leverage is visibility and the recommendations that come from it.
Two account-level leaks worth a monthly look
First, PTU reservations sized for a launch spike and never revisited: the reservation bills around the clock while evening and weekend utilization sits near zero. Second, duplicate accounts per team or per experiment, each accumulating deployments and vector stores that nobody reconciles, so the same capability is paid for several times across a subscription.
Reaching the account in the portal
Azure portal → Azure OpenAI lists every account; opening one and choosing Go to Azure AI Foundry portal reveals the deployments, jobs, and stores that actually generate the bill.