I don't think they say it's only good for eukaryotes, just that it mimics the nature of eukaryotic transcripts in terms of nucleotide content, size distribution, polyadenylation etc. I think the bigger concern using them for prokaryotes is probably that a chunk of the ERCC spike ins are derived from bacterial sequences in the first place.
If you're concerned about the derivation of the ERCC sequences, then there are artificial spike-in's available:
https://www.rna-seqblog.com/spliced-...g-experiments/ (I would link to the sequins site, but there appears to be a configuration issue on their webserver)
Given that you "already see many people use ERCC spike-ins with prokaryotes" what kind of additional validity are you looking for?