You've got the idea slightly wrong. It isn't that Apple will combine two M1 Max, but that they will fail to cut them apart.
Here is a picture of the 3 existing M1 variants. Notice that M1 is radically different than M1 Pro and M1 Max. This is because they appear to be different cuts of the same bit of wafer real estate:
<informative picture deleted due to policy restrictions: google architosh M1 Pro chop and click on the first link to see the image>
It seems quite likey that a 20-core M1 <whatever> will simply be Apple cutting the same wafer so that the circuitry that might make two M1 Max, will remain combined. THis assumes that there already exists a connecting fabric between them that Apple disables when the regions are split. The interesting thing to ponder:
is it possible that there are already existing connections between four such regions that Apple could enable/disable depending on how they cut up the same wafer so that one could get the rumored 40 core Mac Pro SoC out of the same wafer as. you get the M1 Pro and M1 Max?
Wouldn't it be interesting if connections between eight or even sixteen contiguous M1 Max regions already exist in the same wafer?
This means that a single wafer could be cut to allow groupings of already connected M1 Max regions in groups of 1, 2, 4, 8 or 16, giving you anywhere from 10 to 160 CPU cores in an already connected SoC.