A machine learning startup needs to build and deploy custom machine learning models on GCP. They require a managed service that supports both training and inference. Which GCP service should they choose?

Google Cloud AI Platform is the correct choice because it is a fully managed service specifically designed for the entire machine learning lifecycle, encompassing building, training, and deploying custom models. This directly addresses the startup's need for both training and inference capabilities in a managed environment. Google Kubernetes Engine (GKE) is a container orchestration service, not a dedicated ML platform, and would require significant manual setup for ML workflows. Google Cloud AutoML provides pre-trained models or allows training with minimal code, but the startup explicitly needs to build and deploy custom models, which AI Platform excels at. Google Cloud BigQuery ML integrates ML directly into BigQuery for data analysis, not for general-purpose custom model development and deployment. Google Cloud Functions is a serverless compute platform, suitable for event-driven functions, but not optimized for the complex, long-running processes of ML model training and serving.