A production agent in Microsoft Foundry incurs costs for models, tools, knowledge sources, search, hosting, data, networking and monitoring, depending on its architecture. The platform is not described by a single price per user.
Checked: 8 September 2026. Microsoft now calls Azure AI Foundry Microsoft Foundry. Prices depend on region, currency, model and contract. Use the official Foundry pricing page and the Azure pricing calculator for your subscription.
Which meters matter
| Component | Typical billing |
|---|---|
| Model inference | Input/output tokens or provisioned throughput |
| Agent Service | Models and tools used; hosted agent resources where applicable |
| File Search | Vector storage |
| Code Interpreter | Session |
| Web/Custom Search | Transactions |
| Foundry IQ / Azure AI Search | Selected search capacity and additional features |
| Application hosting | Compute plan or serverless consumption |
| Data and network | Storage, database, backup and transfer |
| Monitoring | Ingested and retained telemetry |
| Operations | Engineering, evaluation, security and support |
The Foundry Agent Service pricing page states explicitly that models, tools and knowledge connections can trigger separate costs and, in some cases, separate licences. The Foundry FAQ notes that model quotas apply to Agent Service usage.
A reliable formula
Monthly cost = agent turns × actual model and tool usage + provisioned capacity + hosting + data + network + monitoring + operating effort.
In a pilot, measure at least:
- Input and output tokens per successful turn
- Number and type of tool calls
- Search queries and index size
- Retries and failure paths
- Peak load and latency
- Log volume and retention
- Human effort for evaluation, approval and operations
Data residency and Purview
An Azure region alone does not prove that all of an agent's data stays there. Models, search services, connected knowledge sources, tools, logs and global features can have different availability and data-processing terms. Check every resource against Azure Products by Region, the product terms and your architecture.
Microsoft Purview can support governance and compliance capabilities for Microsoft Foundry, but scope and licensing requirements depend on the capability. The Purview overview for generative AI is the place to start.
Cost levers you can measure
- 1The right model for each task instead of one model for everything.
- 2Limited context and controlled response length.
- 3Tool calls only where the task requires them.
- 4Caching only for suitable, permission-checked answers.
- 5Search capacity sized after a load test.
- 6Budgets and alerts per resource group.
- 7Evaluation of quality and cost before every model or prompt change.
Conclusion
Foundry offers fine-grained architecture and consumption options. That is exactly why there is no credible universal monthly amount for 100, 500 or 1,000 users. Export real meter data and calculate infrastructure, licences and staff effort together.
Sources, retrieved 8 September 2026: Microsoft Foundry pricing · Foundry Agent Service pricing · Foundry Agent Service FAQ · Azure Products by Region · Microsoft Purview for generative AI

