Understanding Foundry Costs
There is no single "Foundry" line item on your Azure bill. Instead, charges appear for individual resources:
Cost Components
| Resource | Billing Model | Typical Cost Range |
|---|
| Model Inference | Per 1K tokens (input/output) | $0.15–$60 per 1M tokens |
| Fine-Tuning | Per training hour + hosting | $3–$100/hour |
| Azure AI Search | Per unit per hour | $0.10–$10/hour per unit |
| Storage | Per GB/month | $0.02/GB |
| Managed Compute | Per VM hour | $1–$40/hour |
Cost Management Best Practices
- Set budget alerts in Azure Cost Management to catch runaway costs early
- Use tags on deployments to track costs per team/project
- Start with Standard tier — only upgrade to PTU when you have steady demand
- Use batch deployments for async workloads (50% cheaper)
- Monitor token usage via Application Insights dashboards
💡 Key Insight: Use the Azure Pricing Calculator to estimate costs. Search for each service individually (Azure OpenAI, AI Search, etc.) since there's no single "Foundry" calculator entry.