There are not any moving parts (As far as I know), no filaments to heat up…
yet it requires heat sinks, fans and stuff like that.
Anybody? 
Resistance. There’s a lot of it.
Hmm… Yeah. I shoulda thought of that!
Thanks
They’re called semiconductors for a reason.
How exactly does your toaster generate heat?
You’re thinking of an artificially intelligent toaster, not a regular one.
Note that turning a transistor on or off creates heat as well. This is the main reason chips get hotter as you run them faster. There are more on/off cycles per second and so more heat generated. The electrical resistance factor is more of secondary effect. Unneeded current needs to go somewhere once a switch occurs, so it ends up run thru higher resistive paths and out as heat. If a computer chip were somehow put into a given state and not switched, the resulting heat produced would be tiny.
There are a great many moving parts. They’re just all really, really small.
A more accurate answer is to say that the heat is generated whenever a semiconductor connects itself to ground.
Whenever memory is “zeroed” out this happened. This waste heat will always be generated whenever you destroy information. Making the transistors smaller means there is less charge wasted every time you do this, but even if you make transistors as small as physics will allow, there is still some.
Just by walkin’ down the street, baby.
Mmmmm…
CMOS transistors generate most of their heat when they are switching states, but these days a significant portion of the heat generated comes from leakage, and has nothing to do with switching at all.
So why does switching to smaller processes improve efficiency? Resistance should go up.
Smaller transistors means it takes less charge to make a state a 1 instead of a 0. That means less current is required to change state. Going to lower voltages also reduces the charge needed for a 1, reducing the amount of current. On the other hand, faster clock speed means states change more often, increasing the current.
Maybe if you tried reasoning with it.
I think it’s the lossy imaginary part of complex dielectric behavior in parts that are changing state.
??? NO.
Suppose I use ice cubes to represent 1, and melt them to be 0.
You’re sort of correct but only because you’re “cheating”. You’ve just arbitrarily chosen a low energy state to represent the 1 and the high energy stated to be the 0. For your ice cube system to be useable as a system for recording volatile memory, it’s going to have to change state from high to low and vice versa.
Very significant. Back 20 years ago ASICs I worked on had quiescent currents in the microamp range. Now they are in the couple of amp range, and faster and bigger processors are even worse.
And it gets worse as you increase voltages for speed.
Well! That clears things up. 
Thanks all, for your responses. 