PC moved away from SLI for a reason. It might actually work better on console with a specified API and layer to handle all the complications, but it's still unquestionably worse for all than having one dedicated GPU.
The AMD infinity Fabric solution - that has won AMD their first US supercomputing government contract using the latest Eypc chips - seems to have cracked the problem.
The input Sony has had on AMD's designs in the last 10years has clearly been massive for AMD. Sony's own multi-chip solution will likely be to stack two APUs interconnecting by some form of Infinity fabric but at very short interconnects.
Console size wise, coming from PS5 monolith, Sony will probably be doing a PS5 slim using a vapour chamber when PS5 sales lower their BOM enough to incorporate one in a £300 console at half the PS5 size, and the PS5 Pro will probably use a lower lithography - if 3nm is ready - to fit a PS4 Pro formfactor.
I never believed that Sony wanted to do a mid-gen refresh with the PS4 Pro, but they knew Xbox was going that way and couldn't risk it being a successful strategy undermining PS4. This info coming to light at this point looks like a pre-emptive warning to Xbox thinking about a XsY model already, that they already have far more in reserve on the hardware side for a 2023 release of a Pro that will maintain the PS5 advantages and do so in a scalable way.
At this point the hardware market for Xbox already looks done IMO. As a stacked APU at the PS5 chip level will be way beyond anything Xbox could even shop for from AMD/nvidia by 2023 even if they were prepared to double Sony's BOM.
Xbox can only really push early for a mid-gen refresh if the performance of the refresh leaves them with a more performant premium console than Sony. But at this stage going early or late won't help if a PS5 Pro will be a stacked APU - as the performance gap of Ps5 over XsX will only widen between PS5 Pro and XsY.