I imagine it would be kind of hard to switch away from Intel in the workstation/cluster space.

Like you have to replace OneAPI, which sounds easy because it’s just one thing, but like do you really want to replace BLAS, LAPACK, MPI, ifort/icc… and then you still need to find a sparse matrix solver…

That is really more of a switch from CUDA to HIP; for most HPC applications, cpu speed isn't the question any more.