We are arguing architecture and you are saying something that has little to do, or something nothing to do with it.
Of course in a general sense you have it right, but does it excuse Intel/AMD's lack of ability? I don't think so. If you really like the PC ecosystem represented by x86, then you should want the same - for the main vendors to do better. Mind you, this complacency is what created ARM's resurgence in the first place. Intel wouldn't do it, so Jobs shifted to ARM.
Should you accept mediocrity because that's familiar to you?
2027 for a 10% gain? Is this a race to see who's worse?
Ok, lets look at that for a minute.
Lets say AMD releases Zen 6 in 2027. On desktop and possibly laptop I expect this would be N3P/N3X, but on server it would most likely be N2 or N2X.
But unlike the past where transistor budget went up by 50% or more with each node, we are likely looking at something closer to 15% or 20% over AMD's current N4P (someone correct my numbers if I am off here. I know there's a great table on this somewhere on this site ).
So first lets take desktop. I am not buying only 10%. The additional transistor count bundled with a better IOD and faster main memory support should easily eclipse 10%, but I would expect <20%.
For laptop, if they stick with N3X (which for cost reasons I suspect they will), they are likely in the same boat as Desktop.
In server, it is a completely different story. Moving to a 32 core CCD and N2X and higher memory bandwidth with lower latency IOD could easily produce double the performance of today's Turin for the next EPYC processors in 2027.
If Intel insists on its strategy of paying for billions of dollars of new lithography equipment in order to have a node change every year (or even more often), I think it likely that they will find their performance improvements dwindle as exponentially as the node change benefits do each node. This does not seem a sustainable business model to me.
I think it more likely that we are stuck with a certain sized chip for a much longer period of time as in the past. What you do with that die area will become more varied. As an example, specialized parts will be developed with over-sized AI processors in them. A specialized laptop gaming part might have an unusually large number of CU's in its GPU within an APU, but only a couple P cores. A specific desktop part might be created for workstations (AMD already does this) where in the case of Intel, there are just a ton of Skymont class cores in it with a single P core.
I just don't see how it is possible to expect 50% gains in performance at the same price and die size within very similar process nodes.
Oh, and I seriously don't see ARM taking over for x86 in workstation or DC markets. Laptops? Maybe, but not for another 5 years at least IMO. There is just too many things that would need to change to make this true.