4.2.2. Licensing and Cost Models

💡 First Principle: AI costs follow two patterns: subscription (fixed monthly per user) or consumption (variable per use). CFOs who need budget predictability want subscriptions; teams experimenting with variable workloads prefer consumption. Mismatching the model to the need either wastes money (paying for unused subscriptions) or creates budget surprises (consumption costs spiking). Match the cost pattern to the use pattern.

Microsoft 365 Copilot licensing:

• Model: Per-user, per-month subscription
• Includes: Access to Copilot in all M365 apps, Microsoft Graph grounding
• Budget impact: Predictable monthly cost based on user count
• Consideration: Not all users may need Copilot—license strategically

Azure AI Services pricing:

• Model: Consumption-based (pay per use)
• Examples: Per API call, per processed image, per transcribed minute
• Budget impact: Variable cost based on usage
• Consideration: Can scale dramatically with volume—monitor and set limits

Microsoft Foundry pricing:

• Model: Combination of compute, model usage, and services
• Components: Model inference (per token), training compute, hosting
• Budget impact: Can be significant for custom model development
• Consideration: Development costs plus ongoing operational costs

⚠️ Exam Trap: Questions about budget predictability often have per-user licensing as the answer (predictable) versus consumption-based pricing (variable). Match the organization's need for predictability to the appropriate model.

Reflection Question: A CFO wants to invest in AI but is concerned about unpredictable costs. They're considering both Microsoft 365 Copilot and building a custom solution in Azure. How would you advise them on cost predictability?