Something ridiculous like 90% of arXiv papers get lost in the sheer volume of submissions - 90% of it could be wrong (that's a made up number, sue me).
Just like flat earth or anti-vaxer theories on YouTube, incorrectly published work could be damaging and dangerous.
Getting peer-reviewed & published legitimizes the research work. It has been verified by a panel of experts.
Good for those start ups. I can't wait to attack all their Deep Learning APIs [1].