I agree, but no "capacity problems" does not equal infinite supply. If you can sell every 3800/3900 you can make within a given time frame (ie the launch window) why would you choose to use that capacity to make lower margin parts on the same node instead unless the yields were bad enough that you couldn't turn all the chips into the higher end products? Simple answer is that you wouldn't from a purely financial standpoint... once the launch surge is over and demand normalises, that is when you would then choose to flesh out the product stack further with less profitable parts but a broader/different market appeal.
It doesn't make sense to tool up expensive production capacity for the demand anomaly that is a launch window, so supply will always be constrained during such times but that doesn't equal capacity problems.
Oh I entirely agree, production capacity does indeed need priority for the most viable products first. The real picture though I think is more granular than most people are claiming. For instance, if Zen 3 chiplets are made using N7P and the consoles are made with N7+ then they're not going to interfere with each other so there's little risk of meeting console demand impacting Ryzen demand, which is one thing people are suggesting; consoles will stunt Ryzen release.
Also, and specifically talking about Ryzen here, it's not a case that AMD choose to make chiplets with a specified core count. The chiplet is always 8 cores and then get binned to determine which CPUs get made. It's entirely possible that once EPYC has been fully allocated (because server always comes first), the leftovers for Ryzen might see a massive bias towards 6 core chiplets. In such a case AMD would be able to make more 6 and 12 core SKUs than 8 core SKUs, so the launch day focus would beon the 5600X and 5900X, with 5700X as a low volume, token release. Conversely, the bulk of the binning may see 8 cores, so the 5700X will be front and centre, maybe with the 5900X taking a back seat in favour of the 5600X.
It'll all come down to the binning.