Cost and FinOps in Microsoft Foundry

FinOps is the practice of managing cloud spending as a shared job for engineering, finance, and business teams. Microsoft Foundry supports that practice for AI workloads. You can estimate what a workload will cost, see what each subscription, project, agent, and model deployment costs, cap spending before it runs over, and lower the cost of each request.

Cost in the Foundry portal comes from the same Azure Cost Management data that finance teams see in the Azure portal, so engineering and finance work from the same numbers. This article maps each cost task to the capability that supports it, explains where cost numbers come from, and points you to the role you need.

Diagram of cost and FinOps in Foundry. Tags on model charges assign cost to projects, agents, and model deployments. You plan, track, control, and optimize cost in the Foundry portal and in Azure Cost Management.

Important

Items marked preview in this article are currently in preview. This preview is provided without a service-level agreement, and Microsoft doesn't recommend it for production workloads. Certain features might not be supported or might have constrained capabilities. For more information, see Supplemental Terms of Use for Microsoft Azure Previews.

What you can do

The following table follows the FinOps lifecycle: plan, track, control, and optimize.

Stage Task Capability and where to find it Status Learn more
Plan Estimate costs before you deploy Azure pricing calculator GA Estimate costs
Plan Understand how each charge accrues Billing models and meters GA Billing concepts
Track See spending across a subscription Manage > Cost in the Foundry portal Preview View subscription costs
Track See what a project, agent, or model deployment costs Monitor > Cost in a project Preview View project, agent, and model costs
Track Spot cost changes next to activity and health Manage > Fleet Preview Monitor your agent fleet
Track Find the usage behind a cost change: runs, requests, tool calls, and tokens Manage > Operations Preview Monitor agent and model operations
Track Analyze cost by meter and tag, charge it back to projects, and export it Cost Management > Cost analysis in the Azure portal GA; project tag in preview Analyze costs in Azure Cost Management
Control Send alerts or block requests when spending or tokens reach a budget Cost controls: Manage > Cost > Cost Control, or Monitor > Cost Control in a project. Also the REST API. Preview Cost controls overview
Control Get alerts when cost reaches a budget Azure Cost Management budgets in the Azure portal GA Create and manage budgets
Control Cap deployment throughput Manage > Quota GA Manage quotas
Control Cap tokens for a project's model deployment AI gateway token limits: Manage > Gateways GA Enforce token limits
Optimize Find recurring cost and token issues in an agent's traces The Insights tab of an agent GA Use Insights in Foundry
Optimize Lower the cost of each request Model router, prompt caching, batch, provisioned throughput, flex processing, and Agent Optimizer GA, except flex processing and Agent Optimizer, which are in preview Cost optimization overview
Access Give people access to cost data and cost controls Foundry roles and Azure roles GA Permissions for cost data and cost controls

Where cost numbers come from

Every request that an app or agent sends produces usage data, billed cost, and, if a cost control covers it, a running total for that cost control. Each arrives at a different speed and answers a different question.

Diagram of where cost numbers come from. Usage reaches Azure Monitor within minutes, cost reaches Azure Cost Management 8 to 24 hours later, and cost control totals update in real time. Each feeds different Foundry pages.

Data What it includes When it appears Where you see it
Usage Tokens, agent runs, model requests, tool calls, errors, and latency. Within minutes Operations, the activity cards on Fleet, and token counts on a project's Monitor pages
Cost Charges at the prices you pay, including negotiated prices, before credits and taxes. Tags on model charges assign them to a project, a model deployment, and, for agent requests, an agent. 8 to 24 hours after usage Every cost in the Foundry portal, including cost by project, agent, and model deployment, and cost analysis in the Azure portal
Cost control totals Each cost control's running total of tokens, or of spending priced with a snapshot of your negotiated prices. In real time The Cost Control tabs, and the alerts and blocks that cost controls trigger

Check usage to see what's happening now, cost to see what you spend, and cost controls to act before spending runs over.

Why the numbers can differ

  • The Foundry portal and the Azure portal show the same cost data. If they differ, compare the scope and filters, such as the subscription, resource group, project, and date range.
  • Cost totals and breakdowns cover different things. A breakdown by project, agent, or model deployment includes only the charges that are tagged with one. Totals include every Foundry charge in your billing data. So the breakdowns usually add up to less than the total.
  • Cost pages and cost controls count spending slightly differently. Cost controls act in real time, so they estimate spending with a snapshot of your negotiated prices, and retail prices for models without one. During public preview, cost controls aren't reconciled with your bill, so their totals can drift from the cost pages over a long reset period, such as a month.
  • Cost pages and your invoice can differ during the month. Current-month cost is an estimate until your invoice is final, and cost data doesn't include credits or taxes.
  • Agents that call model deployments in another Foundry resource don't include the cost of those calls. The calls are billed to the other Foundry resource, so view the model deployment's cost there.
  • Agent usage before November 2026 isn't assigned to an agent, because billing data includes agent names starting then. That usage counts as direct model cost.

For details, see Understand the cost data.

Choose between the Foundry portal and Azure Cost Management

Both tools use the same cost data. Use the Foundry portal to run your workloads day to day, and Azure Cost Management to reconcile, allocate, and report on what you pay.

Aspect Foundry portal Azure Cost Management
Answers What do my projects, agents, and models cost, and are we within budget? What did we spend by meter, resource, and tag, and who should be charged?
Data Cost from Azure Cost Management, next to usage from Azure Monitor Cost from your billing data
Level of detail Subscription, project, agent, and model deployment Billing account, subscription, resource group, resource, meter, and tag
Spending limits Cost controls can send alerts or block requests Budgets send alerts, but don't block requests
Best for Developers and platform admins who monitor costs and enforce budgets Finance and FinOps teams who reconcile, allocate, and export costs

Cost terms

Term Meaning
Cost Charges from Azure Cost Management, at the prices you pay, before credits and taxes. Cost appears 8 to 24 hours after usage.
Estimated cost Cost that can still change. Current-month cost is an estimate until your invoice is final.
Latest billed cost A project's cost so far this month.
Month-end forecast A projection of cost through the end of the current month, based on your billing data.
Usage What your agents and models consume and do, such as tokens and runs. Usage appears within minutes.
Cost control A Foundry policy with a budget, the usage it applies to, a reset period, and threshold actions that send alerts or block requests.
Azure Cost Management budget A budget on cost that sends alerts. It's separate from Foundry cost controls and can't block requests.

Costs and cost controls follow the Foundry resource hierarchy:

  • Subscription: contains your Foundry resources.
  • Foundry resource: contains projects, model deployments, and agents. It can own cost controls that cover its projects.
  • Project: groups agents and model deployments. It can own its own cost controls.
  • Agent or model deployment: the most detailed level of cost.

Start with your role

Permissions at a glance

To Foundry roles that include it Azure roles that include it
See cost and forecasts None Cost Management Reader or Reader, on the resource group or subscription
See usage, such as tokens and runs All Foundry roles Monitoring Reader, and Reader
View cost controls All Foundry roles Reader
Create and manage a Foundry resource's cost controls Foundry Account Owner and Foundry Owner Owner and Contributor
Create and manage a project's cost controls Foundry Account Owner, Foundry Owner, and Foundry Project Manager Owner and Contributor
See usage against each cost control None Monitoring Reader or Log Analytics Reader on the Application Insights resource for cost controls

Cost comes from Azure Cost Management, which checks access on the resource group or subscription, not on the Foundry resource or project, so no Foundry role shows cost. To see a project's cost, ask for Cost Management Reader on the resource group that contains the Foundry resource. To also see the costs of agents and model deployments, ask for it on the subscription. For details, see Permissions for cost data and cost controls.

Owner and Contributor include every permission in this table.

Important

The Foundry RBAC roles were recently renamed. Foundry User, Foundry Owner, Foundry Account Owner, and Foundry Project Manager were previously named Azure AI User, Azure AI Owner, Azure AI Account Owner, and Azure AI Project Manager. You might still see the previous names in some places while the rename rolls out. The role IDs and core permissions are unchanged by the rename.

For each card's permission, the scope to assign, and how to fix access messages, see Permissions for cost data and cost controls.