Nope. Constant power just means that it is not allowed to exceed a certain threshold. Power draw is a result of transistor flips or work being done. If there is no work being done then there is no power draw.
It’s kind of both. Not being able to exceed a certain power threshold applies to fixed clocks, too.
We don’t know the specifics but it should be that when the system is under load it will adjust frequency to keep power constant.
As it has been described by Cerny that doesn’t mean occasionally scaling up to maximum clocks when there’s not much happening, but it being designed to sit at them under expected game loads and dropping them while keeping power constant when the power draw is unusually high.
Examples given for high power draw scenes include a low triangle map screen, or extensive use of power hungry instructions that they otherwise couldn’t anticipate if designing a fixed-clock system.
Variable clocks with a fixed power target that is independent of any die or ambient temperature is entirely about having a deterministic cooling target and being able to run game code at clock speeds impossible to run at with a fixed clock that has to anticipate edge cases like uncapped map screens or unusually high power hungry instruction usage, like the 128-bit native instructions Cerny gives as an example.
It’s how they’re getting so much out of many less CUs. It’s part of what was being whispered as “brute force vs efficiency” before any specifications were made public.