It is a stable axis of rotation, about which mass is most evenly distributed and there is not enough imbalance to cause the ball to choose another axis.  In a symmetric ball like an Inferno, there is really only one PSA, at the pin.  That's why pin-axis balls don't flare (much), they are already spinning around the PSA.  The ball is also dynamically stable spinning 90* from the pin (on the equator, if the pin is at the north pole), but there's no particular spot on the equator that is more stable than any other spot.  Since there is no preferred axis 90* from the pin, there's no secondary influence on reaction once the pin is in the track.

For asymmetric balls with a significant PSA (most modern mass-bias balls, but in general, balls with an intermediate diff greater than 0.010" or spin times below about 8-10sec), there is another PSA 90* from the pin.  The ball is no longer stable simply with the pin in the track like for a symmetric ball, the mass bias (or PSA or whatever you want to call it) must also be either in the track (and thus the ball is spinning 90* from both) or on the current spin axis.  I believe that in either case, the ball is dynamically stable.  Since the ball is still unstable even with the pin in the track (or on the spinning axis), there is still track flare associated with the ball's effort to put the PSA in a stable position.  This is a secondary influence, and it's why the CG doesn't have as significant a role in the layout of MB balls as it does in the layout of a symmetric ball.

SH