Which AWS compute service should be used for an API designed to handle highly variable request volumes, requiring asynchronous processing within seconds, to minimize costs?