A media company wants to deploy a machine learning (ML) model that uses Amazon SageMaker to recommend new articles to the company's readers. The company's readers are primarily located in a single…

While SageMaker real-time inference with auto scaling provides low latency and scales with traffic, it does not scale down to zero instances during periods of very little traffic, resulting in higher costs compared to a serverless solution.

Amazon SageMaker Serverless Inference is highly cost-effective for workloads with intermittent or spiky traffic, as it automatically scales resources and scales down to zero when idle. By adding provisioned concurrency, the solution ensures that requests during predictable peak traffic hours experience minimal latency by eliminating cold starts, directly addressing the need to minimize delivery time.

SageMaker Asynchronous Inference is designed for large payloads or long-running inferences where immediate responses are not critical, which contradicts the requirement to minimize the time required to deliver recommendations (low latency).

A SageMaker batch transform task is used for offline processing of large datasets and is unsuitable for providing real-time recommendations that require minimal delivery time.