My new supervisor, the day I met him, was talking about how space data centers are a great idea ("because it's so cold up there!") and will be amazing when they're online. That's the moment I realized he was breathtakingly stupid. He may not believe in thermodynamics, but thermodynamics believes in him.
I guess I shouldn't have expected much given that he has a degree in finance and has worked in consulting for 10 years.
AI as a cultural phenomenon has been truly exceptional at revealing the extent to which the privileged are just literally mouth breathing thumb sucking mental invalids.
We should be laughing in their faces, when they say shit this stupid. We should be pointing out that they must have failed highschool physics, mocking them, to their faces, with a little bit of accidental spittle to drive it home.
We literally have to shame these fucking morons or they will destroy human civilization.
I love it when they bring up the dangers of AGI. Dude, what about the dangers right now? Cognitive offloading, psychosis, environmental damage, financial bubble, mass layoffs (I know it's mostly a pretense for that one)...
Yeah, maybe at some point it'll be capable to destroy us, but right now it looks like we're going to do that before it'll get there.
He may not believe in thermodynamics
How does thermodynamics make this a bad idea? Is it because the heat generated can't escape?
Yeah. The only way to cool things in space is to radiate it away with fins, or the more destructive approach of jettisoning material.
How do the fins help if there's no hot material being jettisoned? Are we assuming there is some atmosphere that will absorb the heat through (I'm guessing) convection?
Radiation is the part you are missing. The three ways of dissipating heat are: conduction, convection, and radiation which is what the "fins" do.
Doesn't it work out to something like a full kilometer of the things in order for it to work? The idea is pure madness.
Oh lord yes. Its why its stupid.
Depends on density. I think Scott Manley's analysis is probably correct. SpaceX knows how to deal with a few thousand watts of heat per satellite from starlink.
Then the question is whether a few kilowatts is enough compute per node to be worth the trouble.
No. In the vacuum of space there is no convection. The only maintenance free(ish) method of discharging waste heat is to radiate it as infrared, which is not terribly effective compared to terrestrial heat management systems where we have the benefit of a big old atmosphere to dump heat into.
Radiative cooling into space is seriously weaksauce. The amount of heat an object can dissipate in such a manner is described by the Stefan-Boltzmann Law. It would take nearly a square meter (0.84 m^2^ according to my admittedly possibly shaky math) of perfectly ideal thermally conductive black body radiator material to dissipate the 640 watts of waste heat from just one datacenter style GPU at 70° C.
A square meter of heatsink. For one GPU.
Your radiator heat sink can't be shaped like a terrestrial one, either, with stacked fins providing a high surface area in a small volume. That's because a black body radiator is not only an ideal emitter of heat into a vacuum, such as it is, but also an ideal receiver. Your heat sinks will have to be wide and flat so they don't radiate most of their heat right back into other parts of themselves, and this also precludes putting your equipment near other pieces of equipment so they don't radiate their heat into each other.
A single server rack in an AI data center will consume and thus have to dissipate something like 80 killowatts, i.e. 80,000 watts, which even if you had access to some type of physics-experiment-land totally ideal radiator material with an emissivity of exactly 1 would require a 102 square meter radiator just to dissipate that same 70° C. And no part of it could be baking in the sun, nor be influenced thermally by any adjacent servers. In reality it'd have to be even larger, because such a perfectly ideal material does not exist.
TL;DR: Getting rid of heat in space is extremely difficult and in fact is one of the biggest challenges of spacecraft design. Thus putting massive heat generating computers in space is a self-evidently moronic idea as cooling them would be effectively impossible.
Thanks for the thorough answer. I guess I'll just scrap this orbital data centre I was building.
One day we will figure out how to capture and store waste heat more effectively.
No idea when tho
And do what with it? You can't use heat to do work without there being temperature differential in the system. Maintaining that differential requires keeping your cold side cold, which means it still must dissipate its heat. In space you would have exactly the same problem doing that as just radiating that heat in the first place. Once your system reaches equilibrium between its hot and cold sides, no work could be done with that heat energy. It's just a radiator with extra steps.
If capturing heat energy to do something with it did not require sinking the waste heat from that selfsame process someplace, every satellite in orbit would already be covered in Peltiers or similar.
What if i just want to warm my hands or boil water
You can warm your hands until you reach thermal equilibrium, and/or use the boiling water to cool your system by allowing the water vapor to escape into space!
*Rocket fuel and other launch costs not included.
Just wanna boil pasta duh
The short version is - if you've ever heard of infrared cameras that can see heat, that's because everything glows based on heat. When things are very hot the glow is visible, but even cool things emit light outside of the visible spectrum. And emitting light takes energy, which means blackbody radiation (as this is called) takes energy (heat) away.
Jesus Christ, how do you think the Sun works? How are people this clustering ignorant!!??
That and the fact that space is not just extremely cold but also extremely hot, depending on what's between the subject and the sun.
As someone with actual experience working in datacenters, this shit needs constant maintenance and repair. You can't afford to pay for my travel expenses to reboot a server.
There's a bunch of sealed underwater data centres and they found reliability went right up (see Project Natick). Underwater has the benefit of actually having cooling though ..
Right, I'm sure the hard radiation will help with that as well.
You can't reboot a server remotely? My vast experience having been in one collocation once made me think that surely in big datacentres each server has a remotely controllable switch on its power source, like something that comes integrated with the rack itself? Is that not a thing?
You can't remotely replace hardware.
Did I say "replace"? Oh, sorry, I meant "reboot". Must have been distracted when I typed that. Silly me.
From experience, the remote "reboot" actuators sometimes they need to be rebooted themself
Well then you add a remote reboot switch for the reboot switch. Duh!
I think it is a great idea so long as the billionaires go there to stay with the data centres for maintenance purposes. They are the only geniuses who can do it!
Meanwhile China already has data centers in the ocean. Almost anything they do you can bet it's a good idea [economically] because they're not at the point to where they can waste billions and still come out on top like the US has done for decades.
Microsoft experimented with oceanic data centers too, and I don't believe they actually abandoned that project either. It's just Musk and his ilk are very loud and outsized loudness has a tendency to dominate public discourse.
It was concluded in 2024. MS called it a success but also discontinued research and expansion for reasons that don't seem clear from what I can find.
Oh, shame to read that. It sounded like it had real potential.
Data centers on floating barges (or submerged just below the surface) makes so much more sense.
You're still just heating the water.
True. But that's unavoidable ... assuming you're dead set on having datacenters. Datacenters are always going to produce heat, and that heat has to go somewhere.
Better to just not build the data centers
"YOU HAVE AI PSYCHOISIS" yells the idiots frothing at the mouth with AI Psychosis.
The Yahoo!tech writers went on a deep dive and watched a YouTube video.
Never heard of this. But even if you could connect to them, how would you cool it?
Edit: never mind, that's exactly what the article is about.
Experts do not argue that. The video is actually solid and the white paper they reference definitely is slop.
This is there repost I've seen with the dumb headline. There is real critic and the rubes dumping money into AI slop inspired renders are bigger idiots but if you are just uncritically throwing ad homenens like this then it looks like you're not living up to your potential either.
oh fuck off... the point isn't to get them into space... the point is to get them off this damn planet. nobody wants this bullshit grift but idiots who are already addicted to ai slop bullshit.
nobody cares if these datacenters can dissipate heat in space... because nobody wants these fuckin wastes of resources and water.
fuck em... throw them, and their ceo fucks into space.
Haven't they wasted enough money on this crap? I'm sure blasting data centres into space will help with the exorbitant costs.
It's cause they can do whatever they want with your data when it's not ruled by earth laws.
But it is ruled by physical laws, and those say it's a dumb idea, and anyone telling you otherwise is trying to scam you.
I know it's most likely AI, but for some reason I imagine it's for some popular FPS or RTS or MOBA, and some underhanded players are going to start sending over planes or something to block the satellite signal
This is a most excellent place for technology news and articles.
