This orchestra plays out of tune – microservice reality

This rings very true to me.

A progression from monolith to modularlized monolith to modularlise monolith with trivially sharded db - mostly gets you to a pretty serious scale without major re-writes and without original sin (premature optimisation). (This assumes you're not trying to re-train RNNs for every app user multiple times per session, or mining bitcoin blocks, or any other super-backend-heavy website or app...)

So long as you have a plan (and this is mostly a business plan for how you're going to budget for this) about how you're going to notice sufficiently in advance of running into the hard upper limits of a horizontally scaled modularized monolith (ie: where the hosting costs of incremental scaling hockey sticks towards vertical).

If your app and business model doesn't have an obvious mechanism to pay for a large enough engineering team to re architect and re write from scratch well before you get to that hosting cost inflection, that's a good reason to question the sustainability of your plans. (And I'd settle for a "We just need to land another 250 million in VC funds by xmas!", so long as you've got a realistic plan for how you're going to use and raise that capital...)