let's use a simple example

you have 4 LED that is fits in car headlight but only 4 fits because they're huge. it uses 80 watts total (20w each)

next year they release new LED's that are 20% more efficient and half the size, so 16w each. but since they're half the size, you can fit twice as many now. 8 led's * 16w = 128 watts total

this is what happens with cpu's. the size of the transistor shrinks, and so does power use of each transistor. but since they're smaller, the companies pack way more of them onto the processor. if you have enough, it can end up being more power. if they kept the transistor count the same, it would use less power