How can a PhD in Machine Learning and AI help you in a life out of Academia?

Working at your startup

One thing for sure is that, if you want to start a company (a startup), you should not be afraid of being overqualified, since you will be "the boss" and there is no one to classify you as such.

Working in the industry

While being overqualified sometimes (perhaps, even, often) applies to positions in the industry, a Ph.D. graduate is highly unlikely to be considered as such in ML/AI domain, where it is often a requirement for most positions.

Working for other people's startup

Even though most startups generally don't care about degrees, those, working in ML/AI-related domains, will be happy to consider and hire a good Ph.D. graduate for the obvious reason: due to relatively high complexity of the required ML/AI-related subject domain expertise.

P.S. Please note that my answer addresses only your concern of being overqualified, but does not touch a larger question of feasibility of your plan, factors to consider in decision making and so on. However, it implicitly suggests that this one of possible routes to achieve your goal (whether it's the best or the optimal one, depends on a lot of factors, including your personal circumstances).

Both fields are currently "hot" and have several practical applications. Thinking IBM research, MS research, tableau.com, just to name the few that I've applied to... there is absolutely no shortage of jobs. If you interface that with big data, you should be ok for a few decades, at least...

Full disclosure: My phd was in image processing (well, not quite, but not ML either) and I'm currently changing my research area exactly because of that demand...

I know it is not on your plans, but just to be complete, that is even more true for professors and TT positions... cra.org has at least 1 new opening/day in those fields...

So yes, if you wanna work ML Research, a good PhD, with publications (think CVPR+) is the way to go.