In Latin, there are words from Etruscan and unknown sources. In Proto-Germanic, pretty much all words are from Proto-Indo-European. Why is that? Are the Proto-Germanic peoples and language very similar to PIE?
It is not the case that pretty much all PGmc words are from PIE. Many well-known linguists have turned their attention to the problem of precisely why PGmc had such a large proportion of non-IE vocabulary. The classic estimate, as reviewed in Vennemann (2012; 'Was Proto-Germanic a creole language?') was one-third.
Vennemann and his student Mailhammer are well-known proponents of the view that large sections of PGmc vocabulary came from a Semitic superstrate, particularly in certain semantic fields (war/weaponry (Waffe, Krieg, Spieß, Schwert); law (stehlen, Dieb, schuld), etc). But one does not need to subscribe to Vennemann's particular views on Proto-Germanic language contact to recognize the large proportion of non-IE vocabulary.
I suppose the number of accepted PIE etymologies in Proto-Germanic may depend on the source you use. I would consider checking out the Leiden Proto-Germanic Etymological Dictionary by Lubotsky and Kroonen, the Leiden series dictionaries are probably the most up-to-date source on the matter (but of course not everyone agrees on the Leiden school's model of PIE). According to it, a sizable chunk (nowhere near one third though) of Proto-Germanic vocabulary is non-IE.
Germanic shares many words with the surrounding languages. Reconstruction from just two branches (iso-glosses) is thought legitimate by some (greedy, the more the better). Thus, many words that might just be old loans, or perhaps come from a common substratum, eitherway predating soundshift, or incidently of a form not subject to soundshift in either branch, or incidently looking like soundshifted but actually unrelated, will be nominally PIE, to some. This is reasonable to help looking for cognates in further branches, or to break down the descent based on isomorphisms (i.e. legitimizing a hypothesis about Italo-Celtic).
This contrasts with @legatrix fine answer, refering to Venneman, to which I'd add Kroonen's work into Pre-Germanic substrate. Indeed, any language seems to entail a substrate hypothesis, be it Pre-Greek, Indic or Hurrian and Armenian.
This would be most severe if a potential adstratum was IE itself, leading to false images of unbroken descent, that's otherwise subsumed under the moniker of dialect continuum. At that, the many doublets in reconstructed roots are notable on the one hand (examples left as exercise), and yet unbearably huge number of unknown etymologies on the other hand.
However, I've recently read a comment linking a paper that tries to prove the substrate content in Germanic far below the apparently common count of 30%. If I find it, I will post back.