It would not be cost effective for a NN-as-a-service using current GPU technology.
I have no idea about the hardware costs, especially GPU. Could anyone, please, provide some example costs for that? Like how much for XYZ GPU cluster per hour and example thing that XYZ GPU cluster can perform in an hour.
Correct me, if I'm getting it wrong or such example is hard to provide.
Either a) you can only address one customer at a time, meaning it won't scale or b) you charge a high price to make a profit, which will dissuade customers