I don't even think 128 cores are in for Epyc Milan :- ). There just isn't any space on the chip with current node and architecture design. They either have to jump node to 5nm or features some 3D stacking.
There would be fantastic IPC gains of course, and slight frequency perhaps.
(Speculation time!) I believe two things happened at same time:
1) The Lenovo deal was failure so far. I think we can clearly deduce this from fact there are almost 50perc. discounts on the website... taking the usually over-priced corporate builds from 20k + into absurd 10k price.
So now we have plenty of 3rd gen Threadrippers in "pro" variants that need to go somewhere. If 4th gen (5xxx) TR chips arrived now, these chips would be dead weight. Corporations don't buy obsolete hardware even if it's single generation past.
2) With TSMC foundries being overcrowded to hell, there is little reason to cannibalize the little amount of Epyc (Milan) chips that is possible to manufacture to select clients by dilluting it with DIY Threadrippers and OEM integrators.
Perhaps if things go well, 5xxx Threadrippers might be still announced this year, but perhaps in Summer and later. Why even have Halo product when you are winning so much.
Oh c'mon, someone is still runing 4 core cpu here.
Sometimes around year ago, when I bought my 3990X I finally reached a point (together with our render farm) when rendering power is no longer limiting me...not even tiny bit (well we don't do animations though...) and I realized further how slow I am.
I am now rendering even previews in 8k... (and downscaling before showing of course, but I can set-up my post-production layouts in final mask dimensions, feature very clean AA, catch mistakes earlier and have at least some semblance of reason to procrastrinate)