But for supporting more esoteric targets you might be interested in the goals of this ultra-early-stage assembler. ("Planned targets: All of them.")
If you tell me that I can write better, safer code by using Zig, but I can also compile it into a .c artifact that anybody can use, now that is a tempting proposition!
Zig should be just as easy to integrate. Sure, it's one more thing to install, but it'll spit out .o files just like a C compiler would if you tell it to (which means you can shove it in your Makefile or what have you), and will spit out .h files for linking. You miss out on Zig's build system niceties that way, though (including the cross-compilation demonstrated in the article).
Regardless, being a "C replacement" kinda implies (if not outright explies) that it's replacing C; compiling to C kinda defeats that purpose. It'd still be useful, though, and is probably possible (might even be relatively trivial if LLVM and/or Clang provide some mechanism to generate C from LLVM IR or some other intermediate representation).
After you get a taste of a modern toolchain (with cross-compilation, dependency management, but withou not-quite-portable build files to endlessly fiddle with, without outdated compilers to work around), you will not want to have to compile a C file again.
Languages like Zig and Rust are easy to install. Mostly it's just a tarball, so it's less of an inconvenience than getting the right version of autotools.