Afaik one of the downsides in computationally expensive tasks is because of immutable data. For certain problems there simply are more performant algorithms around for mutable data then you have for immutable data. Sometimes you can mimic those by using ets, which is an in memory database within the beam (so mutable), but I’m not sure what the determining factors are for when this works or not.

CPU intensive tasks in Elixir will have the problems @LostKobrakai mentioned, however if you will use NIF or any other interoperability approach offered by BEAM then this will not be much of the problem (your example with Elixir ↔ Rust is great example of that). So if you want that approach then go for it, just shift CPU-intensive work to the Rust (or any other language) and keep Elixir as a frontend for IO-intensive work.