An IVR backend requires a separate persistent gRPC connection per call session, calls last 1–30 minutes, occur mostly in business hours with predictable spikes, and you want minimal cost and ops overhead. Where should you deploy the backend?

Cloud Run is the best choice due to its serverless nature and ability to handle persistent gRPC connections. It scales automatically from zero to meet demand, ensuring minimal cost during off-peak hours and handling predictable spikes without manual intervention. Its per-request billing model is cost-effective for this workload. Compute Engine would incur higher operational overhead for managing VMs and scaling. Google Kubernetes Engine (GKE) Standard mode also involves significant operational overhead for cluster management, which contradicts the "minimal ops overhead" requirement. While Cloud Functions are serverless, they are generally designed for short-lived, event-driven functions and are not ideal for maintaining persistent gRPC connections for up to 30 minutes.