I think in 3 years, there will be purely corporate distros, like Ubuntu, Centos (oh, no, that one is RIP, right), RHEL which will have became systemd/Linux instead of GNU/Linux.
Their desktops will be gnome, cinnamon, mate, KDE... but mostly GnomeD^W Gnome. Maybe Gnome will be merged into systemd and renamed in desktop-environment-d, aka DED.
There, Xorg will be a thing of the past, wayland everywhere, and probably lot of security failures despite the all safer components, because things will be a nightmare to setup correctly. Maybe this nightmare will be voluntarily done, to make corporations more worth of the money they want for the "support", but I think it'll be mostly because knowledge of how system work will be lost.
Things will be very big-data-AI centric, computers will guess a lot, and that'll work nicely... for normal usages.

I explicitly don't put Debian in there, because I believe (as shown by the fact in Debian 10 "buster", "init" can be either: systemd-sysv, sysvinit-core or runit-init) that while systemd will be the major and most widely used system framework, it will still be a GNU-able/Linux distro, with half-ass in corporate-side, and half-ass in hackers-side. That's something I now believe is Debian's trademark after all.

Then, there will be the anti-systemd distro, most of them keeping old sysVinit uglyness by default: Devuan and the others. Sure, they work to use more advanced init and supervision tools, but I don't think they'll be there in 3 years.
Those are the "old guard corporate distro".
Those will provide limited support for big DEs, probably good one for lone window managers.

Another side will be the hacker-distros. Here, I put mostly voidlinux, guix, k1ss, but I'm pretty certain there are others, and we might see more appearing.
Since they are hacker-oriented, they will have completely random support for Xorg or Wayland, not to mention flatpack & the like.
Basically, they'll have completely random support for everything outside the linux kernel, because, well, they *are* hacker-oriented, mostly targeting amateurs (in the meaning, doing stuff for fun).
By random, I mean, some will support a lot a solution, others another solution, and not that things will fail 1 time every 10 times.