So AI has a good claim to being the most effective at reducing x-risks, even the ones that aren't AI risk.

You're ignoring time. If you expect a sufficiently powerful FAI to arise, say, not earlier than a hundred years from now, and you think that the coming century has significant x-risks, focusing all the resources on the FAI might not be a good idea.

Not to mention that if your P(AI) isn't close to one, you probably want to be prepared for the situation in which an AI never materializes.