For reference, a GB300 server is now at about 8.5kw for a single server. A fully populated NVL8 server is about 15kw. Looking online, looks like the ISS is about up to 90kw, so I guess I'm off and the actual number is something like 6 to 10 of these servers per ISS scale facility.

I would still argue this is a crazy overhead for what they would now consider meager capacity, with, luckily for nVidia, a pretty hard deadline where the very expensive equipment burns up without potential for extended lifetime use.

I know who Scott Manley is. I'm subbed to his channel and saw that video when it came out.

That being said, one of the top comments in that video is someone who, claims to be, a spacecraft thermal engineer. And they bring up a few good points. But the one I'm most interested in is the loss of efficiency in heat dissipation the further the heat is pushed along the array. Which means you can't treat the entire surface area of the dissipation array at equal performance, so you need an even bigger array.

And btw, a 20kw satellite is peanuts for AI workloads. Which is the reason they're suggesting putting up a million of these. And that right there is, IMO, the biggest issue. We're already at 14k satellites (most of those are Star Link). And 100k satellites is the current figure we expect collisions between satellites to start becoming unavoidable, with the possibility of an out of control cascade of collision becoming a major concern from there upwards.

I think Kyle Hill did a better job at being objective on the problem:

https://www.youtube.com/live/4mx9Rp-SMNk

A 20KW data center is utterly feeble, barely worth the name, by ground-based data center standards, which easily go into tens of megawatts.