It's primarily thermal cycle stress on the semiconductors. They really don't wear out in "in spec" steady state operation. This is particularly true with the ones that dissipate a lot of heat. A common failure mechanism is thermal stress (sort of like work hardening) of the interface materials between the silicon and the heat sinks, which is typically some sort of solder like alloy.
The big exception is large electrolytic capacitors which do slowly "cook to death" (i.e. the electrolyte leaks out).
Electrical transients at turn-on are only significant in bad designs, which aren't uncommon. This is opposed to electrical transients that occur at random (like lightning hitting your building). One is bad luck that can be minimized by turning the equipment off so it's not as exposed, the other is lack of knowledge or minimizing costs (often development time).
You will seldom see it in any specification, but good equipment designers are concerned with the number of power cycles that there equipment is expected to see. This often requires accelerated life testing of key components, like power semiconductors, since the manufacturers will not give you that data. Some device manufacturers are, in fact, much better than others in this regard. This is why large companies, that can afford it, will do "acceptance testing" for new suppliers. The data sheets are a great starting point, but if you want a very reliable product, you must do your own evaluation and testing.
