To elaborate on your last point: beyond the benefits of appearing to have power, there’s a difference between being in a high-power world state (having billions of dollars) and being able to actually compute (and then follow) policies which exercise your power in order to achieve your goals (see Appendix B.4 of Optimal Policies Tend to Seek Power for a formal example of this in reinforcement learning).
To elaborate on your last point: beyond the benefits of appearing to have power, there’s a difference between being in a high-power world state (having billions of dollars) and being able to actually compute (and then follow) policies which exercise your power in order to achieve your goals (see Appendix B.4 of Optimal Policies Tend to Seek Power for a formal example of this in reinforcement learning).