Data Science is (already) redefined as glorified analytics/BI. That would be an easy transition for you. All of the analysts in my company literally got their titles changed to Data Scientists. There is a glut of talent in Applied or Research Science. Again, top talent is rare - but if you are self-learning you will be competing with new graduates who are more likely going to be better qualified and cheaper (you can get up there with tremendous amount of work but don't underestimate this) I feel doing ML implementation well is where the money is going to go next. Scaling ML implementation is super hard. If you have good sys design chops and can marry that will ML knowledge you will be in-demand.
Depends on current role and what skills need to be proven based on self-taught knowledge. If you are SWE and need to prove ML knowledge, you can do hack-week type projects in current job that showcases your ML skills.
Wrong. Demand for PhDs will stay strong to build innovative algorithms and compete on features. It’s the lower skillset infra/tooling that’s getting commoditized fast. Why build your own model deployment pipeline when you can pipe SQS into sagemaker and monitor it with cloudwatch