We will eventually go full server rendering and just stream every frame to a thin client. We have good enough network now with gbe consumer bandwith and less then 1ms latency up to 50km. But we need more powerful, smaller and cheaper servers.
You just described every single non-SPA...
I can pay to render everything for my user, essentially streaming their webpage to them. Or for the potential cost of a slightly slower page load I can offset the cost of all that rendering to their xGHz multi core CPU that's likely sitting idle.
Not to mention the cacheability of the data.
Yes, latency for a single bit of payload to ever hit the client. The rest of the data then needs to be transferred as well. Transferring that amount of data constantly would be terrible with regards to battery life.
> The CPU and the GPU could be several miles apart.
How exactly? Would you send draw commands over the internet?