Something puzzling is that I keep reading that the sub was made from a carbon fiber tube with titanium end caps.
"Carbon fiber" means a plastic resin with carbon fiber embedded. For weight that material is extremely strong under tension, such as when pressurizing the inside of an airplane hull or under the lift force on a helicopter rotor blade. But compression? Why would a carbon fiber composite be any stronger under compression than is the plastic resin in which the carbon fiber is embedded?
Imagine trying to deform a cable by pulling on it; That would take tremendous strength. Now imagine trying to deform a cable by pushing on it; that would require minuscule force. Are carbon fibers in the submersible hull pushed together by compression of the hull, not pulled by tension?
Well, both, it depends on the direction of force and the orientation of the fiber. A radial compressive point load on a cylinder, under which the circumferential fibers are under compression, is also an axial load, under which the longitudinal fibers are under tension. However, depending on the combinations of fiber orientations, a carbon fiber cylinder could demonstrate compressive deformations along axes in which no fibers are under tension. For example, with only those two combinations of fiber orientations, a carbon fiber cylinder would buckle under force applied inward, perpendicular to a line from its base to top. Now, a carbon fiber cylinder should also have cross-oriented fibers to resist that particular force under tension. But was load analysis performed with the understanding that ONLY the cross-oriented fibers would resist longitudinal buckling under compression?
Maybe what happened is that the sub designer thought to himself, "Carbon fiber is very strong, I'll use that" and it was rated for some load that exceed pressure at depth. But there was no mechanical analysis or testing of compressive deformation that would have caught the difference between evaluating compression and tension.
On the other hand, the failure could be anything, the entire effort sounds like ad-hoc, off-the-shelf, lets-try-it-and-see-if-it-works methodology. Which is all fine and awesome because it's cheap and fast and effective. Build early, test often. You just do not put people in it.