I think the only reason why there was a temporary "language bottleneck" in the late 90s and early 2000s was that CPUs got too complex around that time to deal with the resulting code generation complexities in a "homebrew compiler". LLVM and later projects like Cranelift changed that, and now it's possible again for individuals and small teams to create new languages which have competitive performance. In my book that's great, competition and experimentation is always a good thing for progress (even when it means there's also many more hobby projects that won't move the needle).