From my perspective (I've taken cursory dives into micro in the past), micro is analogous to the libraries developed by every company's infra team to decorate teams' APIs (opentracing, metrics, logging, etc), making them consistently useful with minimal impact on developer velocity.

If this analogy is accurate, it means can deploy these however you want. They're kind of like force multipliers on the effectiveness of distributed computing tools (RPCs, logging, etc).

I welcome refinements/corrections of this analogy!