Question

Your team is building a customer-support chatbot in Azure AI Foundry. Traffic is unpredictable and can drop to zero overnight, then spike during business hours. Management wants to avoid paying for idle infrastructure and prefers to be billed only for the tokens actually processed. Which deployment approach should you choose for the foundation model?

Accepted Answer

Deploy the model to a serverless API endpoint (pay-as-you-go, token-based billing).

Answer

Deploy the model to managed compute using a dedicated GPU cluster sized for peak load.

Answer

Purchase provisioned throughput units (PTUs) with a monthly reserved capacity commitment.

Answer

Deploy the model to a Kubernetes-attached compute target with cluster autoscaling.