As was mentioned in the chat, you're almost certainly going to end up with various "classes" of AI, ala
Schlock Mercenary and
Troy Rising. Those classes will be a representation of how "smart" a particular AI is; that is, how much it can process simultaneously, how good it's ideas and suggestions tend to be, how well it can predict things, and so on.
Very obviously this will be directly related to hardware as well, though not necessarily only dictated by such. It's entirely possible that tossing a "Class 2" AI into a "Class 6" AI chassis will only produce a "Class 3" AI, simply because the original AI doesn't have the software capabilities to utilize all of the hardware capabilities that have been opened up to it. Of course, given sufficiently-advanced AI, it can always bootstrap itself up to eventually become a "Class 6", but in an immediate sense it's restricted by it's software.
Other thoughts to consider (in no particular order):
- AI rights -- Is there a legal difference between a biological sophont and an electronic sophont? If uploading is available, how does that change things? Is a warship AI recruited like any other individual, or are they considered military property with no rights of their own? If they're property, how do people reconcile traditional stances on press-gangs or being drafted with the thought of forcing a fully-aware, thinking individual to fly into a warzone without having any say in the matter? Do they accrue leave time (cue the "AI bar", where all your mechs go to relax and get tuned up while on shore leave)?
- Cyberwarfare -- An AI is only as secure as the machinery it's running on. The computer core of a ship is likely to be one of the most hardened and secure locations it has. Viruses, trojans, and other nastiness is going to be a prime tool for saboteurs, because if you can take out the AI controlling the ship, you've crippled your opponent's combat capabilities just as effectively as bombing their missile tubes.