Why is APM Server a separate component?

The APM Server is kept as a separate component for the following reasons:

It helps keep the agents as light as possible and since the APM Server is a stateless separate component, it can be scaled independently.
For real-user-monitoring (beta), data is being collected in agents running in browsers. APM Server prevents browsers from interacting directly with Elasticsearch (which poses a security risk), and controls the amount of data flowing into Elasticsearch.
In cases where Elasticsearch becomes unresponsive, APM Server can buffer data temporarily (configurable) without adding overhead to the agents.
APM Server serves as a middleware for source mapping for javascript in the browser.
The APM Server provides a JSON API for agents to use thereby improving compatibility across different versions of agents and the Elastic Stack.