Yeah Cascade lake appears to be on 14nm ++ aswell. So next year you would still have Skylake refreshes on 2066 platform. The only 10nm die shrink of Skylake I can think of is Cannon lake but iirc this has now been relegated to low powered cpus or mobile chips so I don't think cannon lake would be coming to 2066.
Iirc from intel's manufacturing presentation they showed that 14nm++ offers superior performance to theirs 10nm and 10nm+ ( I don't know how or what benchmark they used in making this comparison). So I suspect that using Skylake refreshes on 14nm++ for HEDT in 2018 makes the most sense performance wise.
This would then lead straight to Icelake/Icelake-X possibly on a new socket (2076??) in 2020. Icelake would mosy likely be built on 10nm+ but because its a new architecture; its likely that Intel would tweak to make its performance better than 14nm++ .
What I have stated above is basically what the trends in discussions have been around the web regarding intel's direction in the future.
Ofcourse a lot of things can change between now and next 2-3 years so anything is a possibility regarding intel cpus.
The big worry there though is what's their core-limit, at reasonable cost and power, if they stick with 14nm++ into 2018/2019?
I know Intel's 14nm is better than TSMC/GloFo, but I'm pretty sure TSMC/GloFo 7nm (which are both different from each other) are supposed to be better than Intel's 14nm. GloFo's is also looking to be the best of the 7nm choices, pre-EUV at least (probably Samsung's post-EUV), partly since it was designed by IBM.
So Intel will be putting 14nm Skylake refresh-23 against 7nm Zen2. Sounds to me like the gap in price/performance could be even larger then than with Threadripper vs Skylake-X now.
If AMD are really going to charge $849 for the 16c/32t model, and they try to keep their price points, we'll see 24c/48t Zen2 for the same money in late 2018/early 2019.