Oh, that sort of prediction is just asking for trouble.
The Nvidia Pascal architecture seen in the 1080gtx and its siblings have just hit the market, and already there are plans to go further and faster with the Volta architecture.
14 and 16 nanometer products will doubtless be around for a while as there's still a bit more efficiency to be pulled from the technology, but the 10 nanometer process is already in train, as is the eventual switch to HBM memory and beyond (HMC?) from the current GDDR5x in the Nvidia 10 line.
Then there are the new efficiency's from
FinFet technology, and the new Api's like Vulkan that promise to allow better control of Gpu resources "Closer to the metal"
Beyond
that is the fact that simulators with their comparatively small user base simply don't receive the technical/financial and other resources to compete with AAA gaming titles, which means that we're several generations behind the curve technically in what we see in this genre as opposed to whats actually possible.
Judging from that, I think, is going to be an increasingly larger mistake as time goes on and we fall further and further behind.
Unless something like Aerofly can manage to make simming profitable again to anything but third parties. (and many of
them seem to be just hanging on)