See, they could meet power demands in space, solar panels are much more efficient in space vs on the surface of the Earth. I don’t know that even modern panels are efficient enough to supply what is needed, but the numbers are going to be better than what we would need on earth.
But datacenters? In space? The whole idea is half baked at best. Data center equipment isn’t light; and heavy stuff doesn’t like to go up into orbit. Then you need to consider how much thrust you’re going to need to keep that stuff in orbit… The numbers just don’t work in my mind…
If we had a thruster system that didn’t require burning a skyscraper worth of fuel to get into orbit, then maybe? But we don’t, so …
I could maybe see it happening on the moon, because then you wouldn’t need to worry as much about your orbit, but then you have at least three big problems to solve, how the heck are you getting the equipment there, how are you powering it, and simple latency.
Getting it there will burn so much fuel that I’m not sure it’s a valuable thing to do at all.
For power, yes, solar will be pretty good on the moon, just like in orbit, but the moon rotates. One of the faces of the moon is always towards the earth, so when it’s between the earth and sun, that face is in darkness, and if you build on the other side, it will be in darkness when it’s on the far side, away from the sun. You would effectively need an array of solar that runs a loop around the whole surface so at least something is in the sun pretty much all the time, especially considering the moon rotates every 29ish days. I don’t know of any power storage system that’s robust enough to store the power requirements of a datacenter for half a month while the moon slowly orbits back into the sunlight.
The last thing is latency. Light is the fastest “moving” thing in the known universe. We have yet to observe anything that can propagate faster than light. Some things can match the speed, but nothing goes faster. The Moon is approximately 1.3 light-seconds away. Regardless of all other factors, it will take no less than 2.6 seconds, round trip. I don’t know of many applications for data center tech that is ok with that kind of delay. Super computers, maybe, but datacenters, not so much.
The whole thing is wrought with issues from the ground up. And I’m not even a scientist, and I can see the obvious problems here.
Meanwhile, we have 2/3rds of the planet covered in water, which is basically unused space by humans. It’s vast and plentiful, and as a bonus, has built in cooling. Microsoft was testing datacenter stuff at sea and AFAIK, it went pretty well. I believe they’ve discontinued it since it’s still not as practical as land-based datacenters, but the idea is solid at least. Space based stuff is even less practical. I don’t see why anyone would want to take on the cost of something like this when there are cheaper and more profitable alternatives.
You make some good points, and there’s also thermal issues.
The whole reason the datacentres use so much water is cooling all those processors.
Rejecting heat in space is super hard because you’re ultimately relying only on radiation (not enough matter for conduction/convection), no matter how many heavy/expensive/complex/maintenance-needing cooling systems you use.
But at least if it was able to be done, the heat would be 'outside the environment ’ as it were. The idea of using earth’s seas to cool datacentres on top of everything else does not sound ideal…
Water in a vacuum broke my $500 Dyson and I’m still pissed about it. I should have just got the $80 garbage from Target, and if it breaks I buy a new one. Which is so wasteful.
Launching anything into space is heinously expensive. And CO2 emissive.
With very generous math, you’d need a radiator like a mile across to cool a space data center, but practically? Larger.
Datacenter hardware is unreliable and goes obsolete quickly, and any kind of maintenance in space is basically cost prohibitive.
There are other smaller yet still crippling engineering challenges, like bit flips from radiation (which gets move severe as lithography shrinks; look up Nvidia’s research on this), assembling large structures in space reliably, cooling loops for such gigantic structures, and extremely difficult/expensive networking (with distinct issues in LEO or geosynchronous).
And most of all… Solar is dirt cheap on Earth, compared to that.
So is just sticking a pipe in the ground for a geothermal loop, or ambient radiative cooling. We literally have tons of mass to dissipate heat into for free, instead of having to radiate it thermally, yet that’s too expensive for ground data centers, apparently.
That’s the joke.
It’s like saying “air conditioning is difficult” and proposing “I know! Let’s live under the Antarctic ice sheet!” That’s not hyperbole. It might be more practical, actually, as getting mass there is waaaay cheaper…
I hate that the anti AI stuff is 90% idiotic planning permission and capitalism, 5% “The idiots making this put no effort into it” and 5% “I just don’t like it, yuck”.
Like, most of what AI bad is the cultism and corporate shit. Like literally shaving 2% off costs to drain a town’s water or something, or proselytizing scaling up transformers while ignoring the efficiency/scaling papers that keep coming out (because that would break the Tech Bro grift).
…At the same time, the absolute energy cost is ridiculously overstated compared to, say, global aluminum or steel production.
And then you have the ridiculous politicization. An example I often cite is a TV series that was ‘fan remastered’ and (as one component in a long chain) upscaled with an oldschool GAN that cost peanuts to train. Beloved years ago, but all of a sudden the fandom hates it because it has something to do with ‘AI’.
…At the same, you can’t ignore how irresponsibly its presented, where these companies are making pennies from spam/slop literally destroying everything. It’s quite reasonable to say “The idiots making this put no effort into it” or “I just don’t like it, yuck” when 99.99% of user-visible AI generation is slop/spam.
You’ve got most of it right, the part you didn’t show you picked up on/I didn’t make clear enough is the complaints that AI is taking all the electricity and water away from a town. Who the fuck gave the permission to have a data center be built that would impact the quality of life of the people living in an area? Why wasn’t the zillon dollar corporation responsible for this increase in power and water consumption?
And then because most people are using corporate ai with shitty , generic prompts, there’s never a chance to discuss using it as a technical art form or within it’s accurate placement in art history, which is something that would be intellectually stimulating to discuss.
The really hilarious thing is evaporative cooling (that takes so much water) is simple penny pinching over a closed loop system. That’s all.
…Yet Bezos and Musk are talking orbital datacenters?
Pick a lane?
See, they could meet power demands in space, solar panels are much more efficient in space vs on the surface of the Earth. I don’t know that even modern panels are efficient enough to supply what is needed, but the numbers are going to be better than what we would need on earth.
But datacenters? In space? The whole idea is half baked at best. Data center equipment isn’t light; and heavy stuff doesn’t like to go up into orbit. Then you need to consider how much thrust you’re going to need to keep that stuff in orbit… The numbers just don’t work in my mind…
If we had a thruster system that didn’t require burning a skyscraper worth of fuel to get into orbit, then maybe? But we don’t, so …
I could maybe see it happening on the moon, because then you wouldn’t need to worry as much about your orbit, but then you have at least three big problems to solve, how the heck are you getting the equipment there, how are you powering it, and simple latency.
Getting it there will burn so much fuel that I’m not sure it’s a valuable thing to do at all. For power, yes, solar will be pretty good on the moon, just like in orbit, but the moon rotates. One of the faces of the moon is always towards the earth, so when it’s between the earth and sun, that face is in darkness, and if you build on the other side, it will be in darkness when it’s on the far side, away from the sun. You would effectively need an array of solar that runs a loop around the whole surface so at least something is in the sun pretty much all the time, especially considering the moon rotates every 29ish days. I don’t know of any power storage system that’s robust enough to store the power requirements of a datacenter for half a month while the moon slowly orbits back into the sunlight.
The last thing is latency. Light is the fastest “moving” thing in the known universe. We have yet to observe anything that can propagate faster than light. Some things can match the speed, but nothing goes faster. The Moon is approximately 1.3 light-seconds away. Regardless of all other factors, it will take no less than 2.6 seconds, round trip. I don’t know of many applications for data center tech that is ok with that kind of delay. Super computers, maybe, but datacenters, not so much.
The whole thing is wrought with issues from the ground up. And I’m not even a scientist, and I can see the obvious problems here.
Meanwhile, we have 2/3rds of the planet covered in water, which is basically unused space by humans. It’s vast and plentiful, and as a bonus, has built in cooling. Microsoft was testing datacenter stuff at sea and AFAIK, it went pretty well. I believe they’ve discontinued it since it’s still not as practical as land-based datacenters, but the idea is solid at least. Space based stuff is even less practical. I don’t see why anyone would want to take on the cost of something like this when there are cheaper and more profitable alternatives.
Orbital space station + pulley system? Sounds nuts, but…
You make some good points, and there’s also thermal issues.
The whole reason the datacentres use so much water is cooling all those processors.
Rejecting heat in space is super hard because you’re ultimately relying only on radiation (not enough matter for conduction/convection), no matter how many heavy/expensive/complex/maintenance-needing cooling systems you use.
But at least if it was able to be done, the heat would be 'outside the environment ’ as it were. The idea of using earth’s seas to cool datacentres on top of everything else does not sound ideal…
All that evaporating water is gonna trickle down eventually. Just like the money.
I don’t know if I can take them trickling on me much more than they already are.
What’s going to be in the water when it gets down there?
How many golden showers does Trump own?
how’s an orbital datacenter going to cool itself?
Because it’s cold in space, of course /s
To answer your question, it’s very hard at that scale/temperature: https://projectrho.com/public_html/rocket/heatrad.php
Imagine how good water could evaporate in space!
Technically it would be boiling or sublimating
Water in a vacuum broke my $500 Dyson and I’m still pissed about it. I should have just got the $80 garbage from Target, and if it breaks I buy a new one. Which is so wasteful.
Not really, no. It saves a shitload of electricity, which with current technology means not spewing as much CO2 in order to generate that electricity.
See this comment for math and specifics: https://lemmy.world/post/38090104/20233592
But the TL;DR version:
Launching anything into space is heinously expensive. And CO2 emissive.
With very generous math, you’d need a radiator like a mile across to cool a space data center, but practically? Larger.
Datacenter hardware is unreliable and goes obsolete quickly, and any kind of maintenance in space is basically cost prohibitive.
There are other smaller yet still crippling engineering challenges, like bit flips from radiation (which gets move severe as lithography shrinks; look up Nvidia’s research on this), assembling large structures in space reliably, cooling loops for such gigantic structures, and extremely difficult/expensive networking (with distinct issues in LEO or geosynchronous).
And most of all… Solar is dirt cheap on Earth, compared to that.
So is just sticking a pipe in the ground for a geothermal loop, or ambient radiative cooling. We literally have tons of mass to dissipate heat into for free, instead of having to radiate it thermally, yet that’s too expensive for ground data centers, apparently.
That’s the joke.
It’s like saying “air conditioning is difficult” and proposing “I know! Let’s live under the Antarctic ice sheet!” That’s not hyperbole. It might be more practical, actually, as getting mass there is waaaay cheaper…
I hate that the anti AI stuff is 90% idiotic planning permission and capitalism, 5% “The idiots making this put no effort into it” and 5% “I just don’t like it, yuck”.
Not sure I understand you but I think I get it?
Like, most of what AI bad is the cultism and corporate shit. Like literally shaving 2% off costs to drain a town’s water or something, or proselytizing scaling up transformers while ignoring the efficiency/scaling papers that keep coming out (because that would break the Tech Bro grift).
…At the same time, the absolute energy cost is ridiculously overstated compared to, say, global aluminum or steel production.
And then you have the ridiculous politicization. An example I often cite is a TV series that was ‘fan remastered’ and (as one component in a long chain) upscaled with an oldschool GAN that cost peanuts to train. Beloved years ago, but all of a sudden the fandom hates it because it has something to do with ‘AI’.
…At the same, you can’t ignore how irresponsibly its presented, where these companies are making pennies from spam/slop literally destroying everything. It’s quite reasonable to say “The idiots making this put no effort into it” or “I just don’t like it, yuck” when 99.99% of user-visible AI generation is slop/spam.
You’ve got most of it right, the part you didn’t show you picked up on/I didn’t make clear enough is the complaints that AI is taking all the electricity and water away from a town. Who the fuck gave the permission to have a data center be built that would impact the quality of life of the people living in an area? Why wasn’t the zillon dollar corporation responsible for this increase in power and water consumption?
And then because most people are using corporate ai with shitty , generic prompts, there’s never a chance to discuss using it as a technical art form or within it’s accurate placement in art history, which is something that would be intellectually stimulating to discuss.