So you think China is also idiot [0], or Google [1], [2], or Blue Origin [3]?
Do you have aerospace engineering background? what are your arguments?
I don't know how all of these turn out to be, but when you keep repeat the same arguments, without anything to back it up, you should have some reflection.
For china its a future investment. Its R&D, an image topic and its the job of a state anyway to do fundamental R&D.
Google has so much money, I don't know why they are doing this, perhaps because they can, because the people in their lab would like to shoot something up, but thats not the core selling point of Google.
Blue Origin, you are referencing, is doing Terawave its a Starlink competitor. Which is not a AI DC in Space.
I suspect it would be maintained at the same quantization that it's delivered: pods. Pod has some unresolvable issue (things would have to go real bad), then eject it.
What maintenance do you have in mind, specifically, in a modular system like this would necessarily be?
If the argument is that off the shelf systems, designed within a context where humans are a Jira ticket away, won't work well in space, then I can agree with that. But, I don't think anyone is attempting that. Datacenter in space doesn't necessarily mean "standard rack computer in space". Keep in mind that redundant and self-correcting systems already exist in modern GPUs, with active rerouting around dead cores, detected with a BIST, at boot! These kinds of things are standard in financial systems.
And you are right that is possible. But the question is not, 'can you build a data-center in space', but rather is it cheaper to build it in space?
A data-center no matter where has 100'000s of connections, each connection is a failure point. And things that go on rockets that shake to an insane degree need to have lots of special attention paid to each such connection.
Heating again is 10-100x more difficult in space, again, it could be solved, but again, is it cheaper.
Same for transport, transporting stuff into space is 10-100x more expensive.
Same for lifecycle management, 10-100x more expensive in space. You need fuel and move around avoiding collisions and so on.
Same for mechanical complexity overall, a standard server rack on earth is complex (see Oxide) a 'space server rack' is likely 10x100x more complex.
So basically you have to re-engineer literally everything from the ground up for a new environment.
So the question is what is the argument for why you should bare this cost. The only argument seem to be constant sunlight, and maybe regulatory. Regulatory is not looking like it will be much easier in space, launch alone is a huge regulatory burden.
So really we are comparing setting up enough batteries to survive for a night or getting another constant source of power with all the cost mentioned above. I know what side I would be on.
> active rerouting around dead cores
That still means you have a dead core in your system. All you are saying is 'lets pay even more to launch extra capacity in case part of it fails'. And that goes for every part from the smallest core all the way up to the largest subsystems on the sat.
Or you can just order a replacement when you actually need it, like when you have a data-center on earth.
It'll be neat watching someone try, to see what kind of innovations come out of it. I suspect the same learnings would apply to more remote terrestrial datacenter (icy regions, ocean, etc, where people don't like to be).
Google has so much money, I don't know why they are doing this, perhaps because they can, because the people in their lab would like to shoot something up, but thats not the core selling point of Google.
Blue Origin, you are referencing, is doing Terawave its a Starlink competitor. Which is not a AI DC in Space.