Predictability and determinism: In some environments, GC pauses are problematic.
Modern GCs in Go and Java running on fast enough CPUs can generally keep pause times down below 1 ms, but on a multi-GHz CPU that's multiple million clock cycles - a virtual eternity in some contexts.
Memory usage: GCs waste memory in exchange for performance. Go's GC doesn't compact (last I checked), so fragments the heap.
Ability to have no runtime: in some environments, the need for a runtime is problematic. That's a big part of why the Linux kernel supports Rust but not Go.
All of these issues tend to be more significant in e.g. hard real time contexts, in embedded environments with less powerful processors, etc.
Also btw "borrow checker fighting" is not much of an issue for experienced Rust developers. You quickly get used to how to write code to avoid borrow checker issues. And in general, writing code that way tends to be a good thing anyway.