I found various sources indicating that the Latin word inferus (or infer) comes from a Proto-Indo-European form like *n̥dʰer, the source of English “under” and Sanskrit adhara, adhas. (The Sanskrit cognate(s) are mentioned in Lewis and Short and A Concise Etymological Dictionary of the English Language (Walter William Skeat, 1885); the reconstructed PIE form and connection to English “under” is mentioned in Wiktionary.)
Skeat further says that it is a comparative form, which kind of sounds plausible to me, although I don’t know if this is considered correct by modern scholars.
The thing that's puzzling to me is why Latin has “infer(us)” rather than “inder(us)”, since the regular word-medial reflexes of PIE voiced aspirates in Latin were voiced stops rather than voiceless fricatives. (For example, Wiktionary says PIE *n̥bʰrís developed to Latin “imber”.)
Was PIE *n̥dʰer (or if the “r” is part of a comparative suffix, *n̥dʰ- or whatever) a compound of the negative morpheme *n̥- and some other morpheme? It doesn’t seem obvious to me what the second morpheme would mean. Or was it not originally a compound, but later reinterpreted as one in Proto-Italic or something like that? I’m wondering if there’s any explanation of the phonological/morphological development that gave us “inf” in this word in Latin.
As far as I know, all other instances of “nf” in Latin can fairly easily be explained as the result of compounding: Aspects of the Phonology and Morphology of Classical Latin, by András Cser says:
The overwhelming majority of the nasal vowel + [f] cases are prefixed forms consisting of in- or con- plus an f-initial stem (conferre ‘collect’, infamis ‘disreputable’), but see also the simplex inferus [ĩːferus] ‘lower’ and its derivatives.