This is a rather narrow perspective focused on (and giving examples from) one subfield (deep learning) at one point of time (year 2018). Have factors like (i) commoditization of software + hardware, (ii) the limited mathematics required, (iii) many open problems, and (iv) a lot of industry funding made a PhD unnecessary for doing research in deep learning in year 2018?: Yes. But does this mean that a PhD (with several advanced courses and a few years of struggle solving hard research problems) won't be useful for doing Computer Science research for the next 30 years? The answer is probably "No". If you want a long-term, intellectually satisfying research career, whether in academia or industry, a PhD is extremely useful.
I think an important factor is the salary disparity. Unless you live in the middle of nowhere, academic salaries cannot provide an upper middle-class lifestyle anymore. Academics are only human and it is natural to want the lifestyle all of their peers from grad school, college (or even their recently graduated student who works for FAANG) seem to enjoy. Unless academic salaries in CS increase significantly (and match say, business school salaries), this trend will continue.