[MattSteelblade]

Not even a little; doesn’t pass napkin math. It doesn’t solve any problems while adding a litany of new ones: massive radiators for heat rejection, radiation hardening, and enormous launch + repair costs (assuming repairs are even possible). The idea exists to separate investors from their money; the product is the funding round.

[codethief]

I haven't done the actual math and I might be a few orders of magnitude off but shouldn't electrical resistance drop quite significantly in space, too? (Of course there's the other issue that information processing is an inherently dissipative process because entropy etc.)

[cguess]

How would electrical resistance drop in space? If you're thinking "because it's cold" that's actually the biggest issue. The vacuum means you can't dispose of heat easily, so you need giant radiators, which are expensive, heavy, etc.

[fouc]

there's no repair involved. imagine a series of throwaway satellites on an orbit that essentially leaves them close enough together for effective mesh networking, and probably on an orbit that slowly takes them away from earth.

the compute is used for training, not inference. the redundancy and mesh networking means that if any of them die, it is no big deal.

and an orbit that takes them away from earth means they avoid cluttering up earth's orbital field.

[MattSteelblade]

It sounds like you're describing Google's proposal, which I believe is at least feasible (though likely uneconomic) unlike, say, Starcloud's. I don't think you are correct about the orbit, though; Google's proposal lists the satellites at 650 km, which would give them approximately 20 years in orbit without boosts. They list estimated life at 5 years given radiation concerns, so they almost certainly would purposely deorbit them earlier.