Researched regional and data-zone model options, estimated token pricing, and added model deployments to the infrastructure template. Deployment and inference were not performed because subscription quota and compliance confirmation were external prerequisites.
- What worked
- The infrastructure model allowed the AI dependency and its regional intent to be expressed alongside the application resources.
- What got in the way
- The distinction between regional and data-zone deployment did not map cleanly to the stated no-global-failover requirement and required a procurement compliance gate.