My impression, which seems to be supported by this article? Is that when you're small, you can actually get up & running with an MVP with a non-serverless monol

My impression, which seems to be supported by this article? Is that when you're small, you can actually get up & running with an MVP with a non-serverless monolithic approach and scale for a bit before you hit a wall. And only at that point: 1) Scaling becomes much more complex and 2) Monolithic infrastructure prevents you from easily implementing best-of-breed solutions within segments of your functionality to optimize various functionality. And, from a cloud services $$ POV, it's cheaper and perhaps require less dev time, though I'm guessing that will depend on the project This seems reasonable? At least at the early & early-mid stages. If you make it that far and see things like scaling issue in your future, it seems like you should also be at a stage of growth where you'll be getting reasonable funding offers and can invest the resources into migrating away from monolithic. Disclaimer: My opinion here is formed mostly from following a not-completely-dissimilar process even when working with things more on a monolithic side of things: I'll use a high-powered workstation to spin up a vm's on the same host, and then if I need to I can migrate individual vm's to their own better-resourced instances on other hardware to scale things. I did this some years ago with a Hadoop cluster and the process worked out nicely. Although as it turned out, that example didn't last long: Hadoop was overkill because I overestimated the bigness of my data, which turned out to only be on the bigger side of small. Or smaller side of medium. When I had a rethink on it, I wrote some python code against a the primary sql-based data source & used Keras to do what I needed instead: Iterating each pieceon a nicely-spec'ed workstation took an hour or so, and a full end-to-end run maybe 3-4 hours. But this is kind of my point: Starting out, it's easy to thing "Oh I need to plan for ever possible eventuality & level of scale." No, you don't. And a final caveat to this: YMMV since circumstances differ from project to project. But these are things to consider before you automatically go for slicing each piece of functionality into grains of sand with their own microservice.