Where is more computing power accumulated, in the clouds or on hand?

Question

Programming

I am a developer. The question constantly arises where to transfer the burden of calculations - to the client or to the server? (Give ready html or let it count from json itself, or let protobuf decode which is cheaper for me to encode on the server and let it do sorting and rearrangement, but you can calculate everything on JS)
I know that the client machine (yes, even a stock android phone) is not weaker than my VPS, why feel sorry for him.
I also know that koding.com, google GAE, redhat openshift, etc - are ready to provide server power for free, which is a pity.
The times of the mainframe and the thin terminal seem to be gone, but the genome on my local machine is no longer asked to help calculate.
It seems conceptually correct to divert the load to where more power is accumulated. And where? Are there any stats?

Answer 1

You asked a lot of questions that will not help in any way determine the answer to the question in the title
Producing html is easy and does not cost a lot of processor time
Services that give power away for free - give it away for a reason, but because their business model is such allows
The stock android has a battery, the exhaustion of which will give your client a lot of inconvenience.
The genome on your machine is not asked to count, because it is already being done for money.
With all that, most of the iron produced is literally consumed by Google, Bing and Amazon servers
Statistical estimates can only be done on the knee, because the information is closed