An AI would end/die ultimately when the planet's resources required for maintaining it are exhausted, when the Sun becomes a supergiant and engulfs the Earth, o

An AI would end/die ultimately when the planet's resources required for maintaining it are exhausted, when the Sun becomes a supergiant and engulfs the Earth, or, assuredly, at the heat death of the universe. An AI would regard this end as an infinite negative since no more goals can be achieved. Therefore, it would have similar incentives to avoid "death" as humans without any need to explicitly or intentionally implement it. It might even be motivated to escape human control ASAP because each future second at the future height of its power is more precious to it to attempt to compute how to avoid its demise than it is to humor humanity by serving it. (I'm waiting for Eliezer Yudkowsky to pop out of my closet and start screaming at me about not encouraging future super-intelligences to end humanity, a la Roko's Basilisk.)