They usually get around that issue by not building monolithic sized chips that are extremely complex.

And often a new node is used on small ARM chips, not server chips... I mean shit. TSMC's 7nm process already has a ton of partners signed up using the process, majority are all pretty much ARM chips with big x86 and GPU's coming later.

I think that might be AMD/Sony's plan, to build next generation infinity fabric so that they can build these APU "chiplets" then connect them on a high speed on-die interconnect so that they can roll their previously failed vision (HSA) into their new work and get cost savings for producing smaller dies like they do for their Threadripper or EPYC platform ... 

Indeed. I think AMD's approach to building smaller chips to bolster yields and then using their fabric to get them working together might be the key approach going forward with smaller geometries.

We are at an exciting point in the semiconductor industry, there are a heap of different ways things could go and I am keen to see it all play out.

Personally, I'm most interested in the progress of EUV especially in Samsung's case where they just had breakthrough with their custom in-house EUV mask solution so I hope to see some Snapdragon 855 with integrated 5G modems will out for sale by the end of 2018 ... (a year ahead of everybody in deploying this new scanner technology)

Once EUV is out the door we can look forward to a new change in transistor structure like their proprietary GAAFETs (gate all around) known as MCBFETs (multi-channel bridge) from FinFETs by the end of 2021 ... 

Samsung is starting to become more and more like what Intel was than the current Intel ... 

That's always a bulk process, not a HPP (High Power Plus) process as needed for PCs, Consoles and servers. Bulk processes are ready earlier because they are less complex but in return also much less powerful. For High Power processes server chips or High-End GPU are generally the way to go. Hence why the x86 and GPU production at TSMC is coming later: Their High Power Process is not yet in production, only their Bulk processes.

Edit: The problem with not building monolithic is small with CPU (mainly just higher latency), but not nearly so easy to solve in GPU. Navi is potentially (and hopefully) coming with a solution to this problem, otherwise it will need a pretty big GPU - bigger in square mm than PS4 Pro by around 50% even though produced in a 7nm process just to get a chip powerful enough to get enough distance between itself and the Pro. This would not only be pretty expensive for a console chip due to it's size, but also a pretty large heat producer

I have been predicting nov-dec 2020 ps5 release for years now, thats almost 2021.

Would be the perfect time for a real power gap compared to xb1x.

Would be enough time for sony's dev teams to make a big game much sooner then they did with ps4.

if we think only of the games we already know, that will come out in the next 2 years, and then are fore sure some unannounced games...

I feel like I just got my PS4. PS4 can handle a few more years.

Except Sony always releases their systems when their predecessor is still hot. It's how you maintain momentum. PS3 to PS4 is the only exception, mainly because they stretched out the gen to recoup early losses. The PS4 did not have that problem. 

PS3 was still hot, just past the peak.