The idea, I think is that these servers are your cloud infrastructure, rather than being used for any local purpose.
Imagine a future where computer technology is a bit more stable than it is now, so a server has a 10 year or so useful life before becoming obsolete. Also you have fibre to every apartment building or office block.
Now, you want to convert some electricity into heat for whatever reason. So you buy/rent a "brick" of servers of suitable size, probably an all-solid state affair with no moving parts at all, plug it into the power and the internet and arrange to move heat out of it for whatever purpose you have. As far as you're concerned that's it, and this is cheaper for you than just buying or renting a conventional electric heating element (ie you get paid, or subsidised power for doing it).
As far as the user of the computation is concerned, they buy computation and related services from Amazon or someone, just as they do now.
The middle-man is running a complex management layer that migrates VM instances and data around the millions of "bricks" that they manage, and allocates each as much work to do as the demand for its heat output requires. Balancing the compute demand against the heat demand requires partly scale, partly non-urgent background jobs, partly blanacing load between time zones and hemispheres and partly a few conventional data centres that can fill in any gap.