Embedded systems have flash size constraints. Systemd depends on dbus, journald and udev, which are several megabytes in size. Busybox is a popular alternative here, which contains a bootable userspace (including init) in less space than Systemd’s init code alone. The cost difference is significant for mass production.
Servers and containers
The size overhead affecting embedded systems also affects areas where size is limited due to scalability concerns. No server needs udev or D-Bus, the necessary hotplug functionality can easily achieved by mounting devtmpfs. I want to remind that D-Bus stands for “Desktop Bus”.
Alpine Linux got much popularity for being a smaller alternative for Systemd-based Linux systems (especially in docker, hah!). FreeBSD also showed to be popular, but i avoid that project due to other reasons.
Alternative libc’s like Musl are not supported by Systemd. Poettering considers musl a non-useful libc. Despite that, embedded applicances (like Busybox or OpenWRT) and Alpine Linux (which all target the Musl libc) are in widespread use.
There haven’t been any production-ready ports to BSD yet.
Void Linux dropped Systemd in 2015 to improve Musl support.
High reliability platforms
Systemd still suffers from an amount of race-conditions, as the amount of issues on github indicate. Race conditions are a frequent cause for non-deterministic behavior of complex programs.
For medical or aviation appliances, using Systemd is a risk many companies dont want to take.
At last, there are People who dont want to trust their data to software whose maintainer considers
rm -rf /‘ing your systems not much of a problem.
Maintainers should take responsible for things they break, and not expect others to change their software to restore functionality.