According to this news article, AIBs are suggesting up to 1200w, not 1000w.
https://www.tomshardware.com/news/1200w-power-requirement-rtx-4090
Now, you have to know that they exaggerate to cover their own butts and because the power use from someone's PC can't be predicted.
Nvidia's own recommendation is an exaggeration based on a system with a 5900X in it - quite an efficient cpu in its own right.
If the AIBs are recommending up to 1200w on some of their 4090s, power management involving manual overclocks+higher power limits must be more ridiculous than Ampere was.
You will have to wait for reviews to see what it's like. Gamers' Nexus will definitely cover it, and TechPowerUp likely will as well.
The Founder's Editions stand to offer the best performance per currency, or cost per frame - yes, I know, 'value and 4090, PFFT' - for each gpu tier by far(that they exist in), and that value even includes one's psu options.
[2 generations ago, ending with Turing, that was not the case; Founder's were about 100 quid over MSRP.]
So zszabo may be right about being able to work with 850w units - at least with Founder's 4090, but if you're gunning for 4090 DarkOmegaEdgelordAbsoluteChaosPower model... well, you may have to step it up a notch.
The boost algorithm already does its own overclocking, and look at those psu recommendations compared to Founder's... up to a 350w gap?! These cards become woefully inefficient from just 'a little push'.
Really though, I feel it's gotten to the point that manual overclocks and higher board power limits than Founder's needs to come to a close, or become exclusive to an AIB's highest product tier; the folks getting those should already know what they're getting into.
Though that does mean AIBs would mostly be selling on looks, cooling, and sound profile. Continuing to sell on performance comes with significant costs.