if you talk about converting to audio the audio result of a vsti, it's simple : it lowers the CPU use. For example, I sometimes render my SSD drums because it uses too much CPU in my biggest projects. Same with orchestration when I think it is finalized. If I want to change something, I can reactivate my vsti, and rewrite it, and rerender it.
I'm not sure if I understand the question correctly, but MIDI itself has no kind of audio information, the information could trigger some kind of device that transforms it(the information) into audio signal by interperate it to different notes on for an example virtual synth and/or virtual drum module.
Another thing is that sometimes it's more comfortable to have an audio track with waves to look on while processing or mixing/editing, for convenience and workflow for example.
I actually keep one project with all the MIDI stuff as MIDI, and then I make a second project that uses wav-exports of the MIDI stuff. Freezing is a bit unreliable in Cubase, as it tends to crash it with certain VST-instruments with extensive automation on them.