This is based on my understanding of how memory channels work after doing some quick googling. Based on this article that talks about the compromises of 2DPC at high channel counts, and this article that gives a basic overview of how CPUs actually use that memory, I see what AMD has done as an upgrade, actually, and not a downgrade.
Let’s go back to AMD’s older Threadripper platforms. Having used X399 myself, I know for a fact that it is quad-channel and supports up to DDR4-2933 speeds officially. You can fill up all 8 slots for extra memory, but your bandwidth is going to suffer a bit. Remember that this is 4 channels of memory spread across 8 DIMMs. So you can pick either maximum capacity or maximum speed… although for lower slot counts, you can generally have your cake and eat it too (see the first article).
If you want capacity and speed, then ideally you would want 1DPC. But then you run into the project management triangle, and you get sky-high costs for developing a memory controller that can handle 16 channels per CPU (for each of the 16 slots), for example, instead of 8 channels. Admittedly, AMD’s Epyc 9005 CPUs have 12 memory channels and support 2DPC, but you are definitely not running the fastest memory possible if you wish to have 6 TB of RAM in there across 24 slots. Back to Threadripper, AMD giving us 1DPC is actually an improvement, given that they didn’t increase the number of available slots; WRX90 still has 8 slots, while TRX50 has been downgraded to 4. Even if we set aside the fact that DDR5 is faster than DDR4, having 8 channels of memory for 8 slots instead of 4 channels of memory for 8 slots means you can have both high capacity and higher bandwidth (memory speeds = bandwidth), all else being equal.
You can think about this similarly to how process node shrinks improve CPU performance: for the same power, you can get an uplift in performance; or, you can use less power for the same level of performance. Just swap power for number of memory slots, and node shrinks for an increase in number of channels (performance having the same semantics but bandwidth instead of IPC if you want to be technical about it).
Looking at Ryzen, which does have 2DPC, most boards only have 4 slots (and some only 2!). The impact of running 2DPC is basically negligible, and the cost of having proper quad-channel memory would be high for basically nothing (or actually a waste of time for those mini-ITX boards with 2 slots). On the flipside, I’m sure that because 8-channel memory is common on server boards now, this has brought down the costs enough where it could be profitable to develop it for a desktop/workstation platform and ship a nonzero volume at retail to a certain slice of the market. Hi, it’s me, I’m part of that slice.
@H-i-v-e has already said what I would about marketing and product segmentation so I won’t repeat it here.
TL;DR: For having the same number of memory slots, having 1DPC allows you to have both high speed and high capacity, though at a high cost. The more memory slots you have, the larger the performance hit from running 2DPC. Therefore, WRX80, TRX50, and WRX90 having 1DPC is actually an upgrade compared to older Threadripper platforms.
Hope this helps clear up why AMD went with 1DPC on later (and current) Threadripper platforms.