It is a definite problem with no easy answers. A partial solution is that a STEM PhD can provide a gateway to related fields. Going from a Math PhD to a job in computing or data science is a well-worn path. If you have the skill-set to get a PhD in differential geometry then you have the skill-set to e.g. rebrand yourself as a data scientist (if need be). Since a fundamental problem in data science is trying to find solutions to the [curse of dimensionality][1], having a deep understanding of geometry is actually relevant. A graduate-level knowledge of mathematics linked with an ability (but not necessarily expertise) in programming can go far in industry. 

  [1]: https://en.wikipedia.org/wiki/Curse_of_dimensionality