Better to be a fool and right, than to be an expert who's delusional.
Microsoft have said they will provision 3X the compute in the cloud for every Xbox One they make. That's quite a lot of servers when you think the XB1 is expected to sell over 10 million a year, but they have got statistics on what % of people play at the same time and they don't have to provision for 100%.
So let's say fine, they have 3X the compute power in the cloud, and by this I mean 3X the CPU compute only, because I haven't heard and don't expect them to have 3X the GPU compute as well. So if we measure in FLOPS it's probably more like a only 25% extra via the Cloud than local, which is still less FLOPS than PS4 has available locally.
Anyway that's just statistics and lies. Fine, let's say that extra CPU time is available, so it's like having tripple the CPU power of PS4 at your disposal - but it's not the same tomatoes. The problem is you have on average 8mbits download, 0.5mbits upload, so you can only download 1MB/s OR upload 0.06MB/s to these servers, and they should work for the lowest common denominator. The XB1 has between 65,000 ~ 1,040,000 times more bandwidth locally, and the PS4 has almost 3 times more than that. How is Micosoft going to put their cloud CPU power to use to make the graphics look better when very little can fit through this pipe?
There are going to be some uses, but spotty bandwidth, latency, and the need to support running offline will limit this severely compared to having extra compute power locally.
Last Year Alone MS spent over a billion dollars on their datacenter to provide the X1 and office 365 support. I think MS is putting their money where their mouth is.
I am not sure if your calculations really work that way. The reason why is because we really do not know how the cloud compute can or will be used until we see development along those lines. Graphics are an interesting subject because pieces and parts do not have to be render in the same conventional since they are done today, just the final image. I have heard a few different unconventional statements from developers on this part whick will be very interesting if used. I might throw up a post on the subject to get a discussion going.
Also your solution really only goes down one path and thats bandwidth but there are solutions to get around such problems like having a hosted instance you sync to in the cloud where you are not sending a bunch of data but telling the host what should be processed, have the cloud process those parts and stream the results. As I mentioned in other post, you let the cloud process the environment and let the local system process the immediate stuff the user is interacting with to reduce bandwidth or issues with cloud streaming. I am sure there are a few stuff that can be streamed in advance or stuff that can be stream on demand that do not take a lot of bandwidth.
I believe there is to much conentration on the graphics part of a game without understand all the other parts that goes into a game. Also people are limiting the scope of what you can do to what you think is possible. A lot of people do not create games and I believe you are limiting your understanding based on limited info on the subject. Even I do not create games but I do develope software that interact with cloud based services and some of the solutions I mentioned is how you get around some of the issues mentioned. I am sure game developers will think up even more ways to geet around those issues as well as unconventional methods that solve those problems.
I see people dismissing graphics being rendered over the cloud but what about some of the initiatives by Nvidia with their Grid and Intel. Graphics can definitely be done in cloud space and we may actually see such development down the line