I find it modestly interesting that he is comparing power density (of tubes) to transistors per unit area (moors postulate, its not a law)... I think it could be worth noting the Dennard scaling here, in that the power density of transistors (chip wide) is remaining nearly constant due to us lowering the Vdd of the transistor on every generation of reducing the channel width.

So, the tubes got much more powerful, the transistors are getting more numerous, but at the same time using less and less power each.