I think to gain adoption any general purpose language needs a clear niche to gain a foothold. If you are choosing a tech stack then choosing a language that is not already widely adopted is always a risk. The tooling is usually inferior, there are not necessarily high-quality libraries in domains you care about and you pay a cost in training new employees in a new tech stack.

Java was an OO, garbage collected language that supported (sort of...) hot swapping code. Go was a reasonably performant, garbage collected language that compiled to static binaries and has great cross-platform support. Some grad student decided to write Spark in Scala so you had to use it for Spark jobs Rust is a systems programming language with zero-cost abstractions and no GC that allow you to write mostly memory-safe code.

What is the killer feature of OCaml that makes it worth the investment?

Not saying it needs to be planned in advance, but that there needs to be a very strong competitive advantage in some use case, not just checking the feature boxes. Continuing on the case of Python users moving to Go, they're probably not doing so for interactive use-cases or avec to scientific libraries...

Does checking the concurrency box make OCaml an awesome choice for some use case?

You mention Rust for example but that language is sort of built around a zero cost abstraction principle, which is very much not the case of OCaml, so I still think it's going to be viewed as risky for highly performance sensitive work.