This should be the big highlight of your comment. A lot of people out there don't seem to realize that the mass adoption of AI that is being claimed to happen still can go up in smoke if
1) the tech is proven to be unscalable,
2) the tech has a huge expense (money, personal info, etc),
or
3) the tech is seen as tacky and gains the associated brand image
Right now, AI requires a lot of FPGA power to do, so 2 might be the most likely scenario in the short term. Long term, unless non-technical users can be convinced to let go of their privacy enough for highly public contextual computing to be performed, then 3 will occur.
As for the OP, I'd recommend Full Stack since you get exposure to almost everything that you'd encounter in 90% of dev environments out there.