It’s a completely fucking stupid idea in all ways if you throw any engineering rationale at it at all.
Negative on: Risk, cost, thermals, radiation, maintenance, latency, complexity, environmental.
The only win they have is making more outlandish shit to inflate value until it goes snap. One trick pony.
It needs a single use case to justify its existence. And I am pretty sure it has one. The question is will google reveal it? Probably not.
Even as some sort of failsafe compute for when some meteor hits the earth - already enough for a government to pay 10000x premium for it and it will bring in cash.
Obviously this is not intended to play in the same league as common enterprise datacenter.