The market of people who are genuinely passionate about complex subjects in math and science is saturated relative to available opportunities. It makes more sense for an intelligent person to take the lower overhead and more achievable approach to becoming a value creator (e.g. full stack engineer with a strong focus on product development) than waste time competing against the countless PhDs vacating academia.
I use a similar argument for avoiding the ML/Deep Learning hype train. At a large corp, that job should be left to people who've spent a lot of time mastering the subject. And if you're using ML heavily in an early stage company and don't have a PhD, you may very well be out of your depth competitively or wasting your time optimizing prematurely.
But even ignoring all of that: anyone who's either spent time on or interacted with a data science team understands how difficult it is to create value with ML as well as how intangible the value that's created can often be. I worked at a fairly well known company that told clients we have a data science team and could use ML, knowing full well that the team rarely if at all manages to generate meaningful insights, because dropping buzzwords is an essential branding tool.
Here's a better approach and the crux of why higher math is often superfluous: the best way to create value is to specialize in problem-solving first principles and remain amenable to either adopting new skills ad hoc or hiring to fill any skill deficiencies.
The caveat: if you're passionate about STEM and that's a higher priority than 'creating value' in a deterministic and practical way (and maybe it is and that's perfectly fine and even reasonable), then by all means indulge in it. But it's important to align your expectations about what you want to do with yourself with the way in which you spend your time. A lot of pain arises in misconceptions around the question of what we want and the reality of what we're doing.