I disagree on the conclusion. Intel is not dumb. As soon as the first Ryzen models came to market, and caught everyone (including them) by surprise, they probably started development of a new chip architecture.
There has been no fundamental breakthroughs in CPU architecture in over 25 years, just go look at a the Alpha 21264's block diagram and compare it to modern CPUs, all fundamentally the same apart from ISA-specific overhead and back-end organization.
There is nothing fundamentally wrong with Intel's current architecture, Intel's main problem is simply hitting the limits of what is possible on 14nm and a new architecture won't change that. Going to 10nm on the other hand should afford Intel a little more timing margin to make the scheduler a little deeper and wider, which is why Intel can add extra execution ports to Icelake to nudge its IPC and SMT performance up. AMD is likely doing the same with Zen 2, keep fundamentally the same architecture apart from splitting it between chiplets and make the execution back-end slightly wider to accommodate whatever extra ILP the upgraded scheduler may be able to squeeze out of code.
Unless there is something fundamentally wrong with a given CPU design, such as stupid latency in the pipeline for the sake of higher clocks that end up causing more harm than good (Netburst, driver/dozer/etc.), it does not make sense to start from scratch when a scheduler and back-end upgrade (basically what Intel has been doing since Nehalem/Core iX-nnn) can already get you most of the IPC benefits a new process can provide. Unless AMD hits a brick wall with its Zen core design, it will likely end up reusing the same foundation for the next 10+ years too.