Because getting GWs of power into one building and the cooling it is expensive. Why not let other people buy the computer, provide the power, and it won't need expensive cooling.
Question is: are there enough high latency distributed workloads to sell?
And resilient enough to handle nodes having an accident, catching fire in a flood, parking underground with no signal for days, pausing to be used as a car...
This is an unreliable spot instance at best, with none of the features one can normally attach to an instance (like storage, managed databases, ...). How fast can its Internet be? Will owners need to pay for Starlink too? (What about when parked indoors?)
It would have to be cheaper than all regular hosting options by a long shot for anyone to consider this. A very niche, low-paying market, in other words.
Doesn't distributed computing generally cope with nodes going offline?
The point is it's just a way to use free compute that's sitting around, if you want to sign up to use it you'd obviously make sure it was in wifi range
Question is: are there enough high latency distributed workloads to sell?