Not so sure about that. When you consider time spans.
Currently we can emulate the majority of early games consoles. So theoretically with time and Moors law any hardware will be emulate able in a few decades. With enough information.
The advantage of open source software. Is it can be used with the original binaries to reverse engineer the instruction set even if the original manufacturer wishes to hide it. So with will and effort even the most complex hardware will be able to be emulated on future much faster hardware.
The direct numerics of moors law may not be definite.
But the principal it defines is. In the future computers will have much more power then they do now.
The reason modern GPUs use things like shaders etc is to allow them to archive massive manipulation of data in more efficient ways specific for the task desired.
Honestly this is why I mention time scale as the main thing that will make this possible. How modern gpus or other specialised processers do the task is less important then what the game code is asking the gpu to achieve.
The idea that at a unknown future date. The CPU GPUs or what ever future tech we have. will never be able to run fast enough to read current cpu or gpu instruction sets. And generate the effect defined using future techniques is not viable as an argument. The only questions are how long and is anyone going to have the motivation to reverse engineer the large but finite instruction sets used by secretive hardware corps today.