It will be useless
Facial recognition works with a sliding scale of false positives and false negatives. The lower the rate of one, the higher the rate of the other. This isn't a problem when you want to recognize one person "who is unlocking my phone?" or a small set of people "is this person authorized to access to this facility?" but when you have a database of millions of criminals/terrorists/etc being compared to the millions who enter/exit the border daily, you will have to accept an insanely high false negative rate in order to avoid massive delays if 1 in 10 people was listed a possible match.
Automated facial recognition simply doesn't work when trying to match millions of faces to millions of suspects. If you had a small set of suspects, like the FBI 10 most wanted plus the 10 most wanted world terrorists it would work fine. If you were only testing people who triggered some other alarm like the 'puffer test' thinking you have explosive residue on you then it would also work fine. It won't work the way they want to use it, or the way China wants to use it. Maybe someday the technology can be improved enough that it will, but there isn't anything even in the research stage that I'm aware that is anywhere near accurate enough to make this viable.
So other than a few billion in wasted taxpayer dollars, I'm not too concerned with this. They've had cameras at the borders since before 9/11, and probably have been saving that stuff for an unlimited time since 9/11, so all they're already collecting the information.