MDASH

Operational guides for managing Microsoft AI Foundry models: selecting a model, limiting tokens and spend, and getting email alerts when a cost threshold is reached.

Contents

• 1. Select and deploy a model
• 2. Limit spend and get an email alert
• 3. CLI: get the resource ID and create the budget
• Optional: hard stop at the cost limit
• Quick decision guide

There are two separate levers: tokens (throughput, controlled per deployment in Foundry) and spend (dollars, controlled in Azure Cost Management). Budgets alert you, they do not hard-stop spend.

1. Select and deploy a model

In the Microsoft Foundry portal:

1. Open your project, go to Deployments (or Models + endpoints).
2. Deploy model > Deploy base model > pick the model (for example GPT-4o) > Confirm.
3. On the deploy dialog, set the Tokens per Minute (TPM) rate limit. This is your first throughput cap. TPM adjusts in increments of 1,000 and also sets a proportional requests-per-minute limit. Requests over the limit receive HTTP 429.

Change TPM later from Deployments > select the deployment > Edit, or under Management > Model quota.

TPM as a cost ceiling (guidance)

TPM is a throughput limit, not a price. But it does set the theoretical maximum tokens the deployment could consume if it ran flat out. Multiply those tokens by your model's price per 1M tokens to get a worst-case cost. Real usage is almost always far lower.

Tokens per minute (TPM)	Max tokens/hour	Max tokens/day (24h)	Max tokens/month (24/7)
1,000	60,000	1.44 M	43.2 M
10,000	600,000	14.4 M	432 M
50,000	3.0 M	72 M	2.16 B
100,000	6.0 M	144 M	4.32 B

Example: at 10,000 TPM, the ceiling is about 432 M tokens/month. If your model costs $5 per 1M tokens (illustrative blended rate), the theoretical maximum is about $2,160/month. Substitute your model's actual input/output prices from the Foundry pricing page. Because most workloads run well below the cap, treat this as an upper bound, not a forecast.

2. Limit spend and get an email alert

Foundry does not bill by dollars, so dollar limits live in Cost Management, scoped to the Foundry resource or its resource group:

1. Azure portal > Cost Management + Billing > Cost Management > Budgets > Add.
2. Set Scope to the resource group holding your Foundry resource, or filter to the specific AI Services resource under Cost dimension filters.
3. Set Reset period (Monthly), the Amount, and the expiration.
4. On the alerts step, add one or more % thresholds (for example 50, 90, 100) and the email recipients (up to 5). You can also add Forecasted alerts.
5. Create. Emails arrive within about an hour of the threshold being crossed.

Add azure-noreply@microsoft.com to safe senders so alerts do not go to junk. Reference: Tutorial: Create and manage budgets.

A Cost Management budget cannot target a single model deployment. The finest scope is the whole Foundry (Azure AI Services) resource. For per-model dollar control, isolate the model in its own resource.

3. CLI: get the resource ID and create the budget

Get the Foundry resource ID, then create a budget scoped to that resource with email alerts by deploying the included ARM template. This works the same in Bash and PowerShell.

Step 1: Sign in and select the subscription

az login
az account set --subscription "<subscription-name-or-id>"

Step 2: Get the AI Foundry resource ID

The Foundry resource is a Cognitive Services account. Replace the resource group and account name:

az cognitiveservices account show --resource-group "<resource-group>" --name "<foundry-account-name>" --query id -o tsv

If you only know the name, find it across the subscription:

az resource list --query "[?type=='Microsoft.CognitiveServices/accounts'].{name:name, id:id, rg:resourceGroup}" -o table

Step 3: Create the budget (ARM template)

Use the included budget.json ARM template. ARM supplies the api-version itself, so it works identically in Bash and PowerShell.

1. Edit budget.json: replace <RESOURCE_ID> and <EMAIL>, and adjust the amount and dates.
2. Deploy it to the resource group that contains the Foundry resource:

az deployment group create --resource-group "<resource-group>" --template-file budget.json

Notes:

• To budget the whole resource group instead of one resource, delete the filter block.
• Add a notification block with "thresholdType": "Forecasted" for an early warning before actual spend reaches the limit.

Verify the budget in the portal

1. Azure portal, open the resource group that contains the Foundry resource (for example rg-admin-0578).
2. In the left menu, select Cost Management, then Budgets.
3. The budget (for example FoundryBudget) appears with its amount, Monthly reset, and current spend versus budget.
4. Click the budget to see the ResourceId filter (scoped to the Foundry resource) and the alert thresholds (for example 90% and 100%) with the notification emails.

Alternative path: Cost Management + Billing > Cost Management > Budgets, then set the scope to the resource group. If it does not appear right away, refresh and confirm the Budgets blade scope matches the resource group you deployed to.

Optional: hard stop at the cost limit

A budget alert does not stop usage by itself. To enforce a cutoff, wire the budget to an Action Group that triggers a Logic App or Automation runbook which, at 100%, disables or deletes the model deployment. Pattern: Manage costs with budgets.

Quick decision guide

Goal	Control
Cap throughput per minute	Deployment TPM
Dollar email alert	Cost Management budget with email thresholds
Hard dollar cutoff	Budget + action group + Logic App to disable the deployment