The scope isn't revolutionary - I've used https://mailu.io/ for years for a few 10s of users, and I really like it. I've heard similar good stories from Mailcow users.
On a quick skim I couldn't tell what was new relative to these older compose-based solutions but (as co-author of similar solution 10-15 years ago) I'm interested to know!
What's revolutionary is that mox implements the entire stack itself as tightly coupled monolith.
With mailu and everything else you get a gargantuan stack of half a dozen massive, enterprise scale servers that are held together with duct tape and prayers. They can scale out to hundreds of thousands of users, but if anything breaks they expect there to be dedicated sysadmin staff to fix it, and gluing leaky abstractions on top doesn't really solve these fundamental architectural challenges.
I suppose it's mostly from a technical perspective: Single code-base, single binary, written in modern language (Go).
There are advantages to an integrated solution: The junk/nonjunk flags you set in your mail client are used to assess incoming email (based on full sender/domain/orgdomain/spf/dkim/ip/ipsubnets). I'm not sure how common/feasible that is with solutions that combine mail server components. I've used postfix & dovecot before mox, I admittedly didn't closely look into it but assumed that combo cannot do that.
With component-based solutions like Mailu, the junk/non-junk categorization you do in your mail client (via folders, not flags) indeed does inform the training of the spam filter used on incoming mail.
On a quick skim I couldn't tell what was new relative to these older compose-based solutions but (as co-author of similar solution 10-15 years ago) I'm interested to know!