Main challenge is the value of the nixpkgs repo is enormous, and that looks tightly tied to Nix the language and its implicit constructs. I think instead of an FDI one would have to provide a true competitor that is more attractive on multiple dimensions like:
- reworking Nix as more a modular architecture (see Tvix: https://tvl.fyi/blog/rewriting-nix)
- Tackling a package repo and Nixos equivalent with day 1 reproducibility (not just repeatability) like https://reproducible-builds.org/
- Better UX experience on the "entrypoints" of Nix like home-manager and dev-shell flakes - I think CUE has some nicer language features for this and does not require figuring out derivation generation, just referencing the existing nixpkgs set