“systemd is always better, no matter the situation” is absolutely dangerous for the entire FOSS ecosystem: both diversity and rationality are essential.

I agree with this, however the rest is more open to discussion.

Systemd might make sense for most people on desktop targets (...) “embedded” targets, simpler and smaller is always better.

A few years ago I was working on a bunch of “embedded” devices (4 x ARM @ 800 Mhz + 256MB of RAM) and whatever we the popular alternatives and the truth is that only with systemd we were able to boot and have a usable system (timers, full dual stack DHCP/SLAAC networking network time, secure DNS) without running out of resources for our daemons later on.

The issue with sysvinit and OpenRC etc. isn't that they aren't good, it is that they're simply init systems and nothing more. In order to have just the bare features above we would have to depend on tons of other small packages and daemons that would all eat up RAM and deal with all the integration pain because they weren't designed to work together. Are you aware of the pain and number of things you've to setup to just have dual stack networking? With systemd you cut a lot of those smaller daemons and end up a few that have a much smaller RAM footprint and are actually made to work with each other.

Systemd also providers very useful features like socket activated services in that can be leveraged to have the system wait for incoming connections and once it gets one launch a program. Without systemd it would've been one more constantly running daemon. It also provided us the ability to monitor if all required services were running, kill things going over the line, restart on specific conditions and even trigger alerts.

Yes, you can do all of the above without systemd but the amount of stuff it required didn't fit our 256MB target, nor the power budgets - we tried it, trust me. Besides all that without so many moving parts and by relying on systemd our solution was way more robust and easier to develop / debug.