Diversity in the ecosystem, possible reduction in memory footprint, ability to discover and fix issues in upstream projects that musl-based distros don't currently package, motivation to support CUDA (and other) runtimes that explicitly target glibc, unlock possible sources of funding to pay developers to work on these areas.

The ability to audit a smaller codebase could be compelling for some sensitive environments.

There is not an urgent or immediate problem with glibc in the HPC space, per se.