Very, very common misconception, because of how often you see things/people in movies instantly freeze in space. But it’s just not remotely true.
The analogy the previous user gave is perfect; space is a thermos flask. It’s a perfect insulator.
To break that down a little more, you have to understand that heat moves in two basic ways; conduction and radiation. Conduction is when molecules agitate the molecules next to them. Radiation is when molecules give off electromagnetic energy.
The way a thermal camera works is that it sees the otherwise invisible infra-red light that hot things give off. That’s the radiation part of heat transfer. Radiation is, on the whole, a really slow, really bad way of moving heat.
Conduction is much faster, especially when there’s a big difference in temperature between the two mediums. That’s why you (average temp around 37C) can stand in a 21C room and feel really comfortable. You’re losing thermal energy, because the air touching your skin is colder, but you’re losing it at about the same rate your body naturally makes it.
But if you step outside into air that’s -20C, your temperature is going to start dropping very fast. There’s a much, much bigger difference in temperature now, so the heat transfer is faster. Also that air is probably moving because of the wind, which means the parts of the air getting warmed by the transfer from your skin are instantly replaced by fresh, cold air.
In space you have none of that. Just vacuum. There’s no molecules in vacuum to agitate. So aside from the very small amount you lose from radiation, heat just builds up. This is a huge problem for spaceships and satellites. They have to build in massive fins to help radiate heat away faster.
But it gets worse, because you know what radiates heat really, really well? The Sun. Which you are now exposed to, whenever you’re not directly in Earth’s shadow, with no atmosphere to absorb any of that incoming radiation. So the biggest problem for objects in space is rarely getting too cold, and far more often it’s getting too hot.
Introducing something that already has massive cooling requirements into that environment would be a total fucking nightmare.
I’m wondering if you could drag something into earths high atmosphere to conduct heat away from the data center but if Anathem and Seveneves taught me anything about orbital mechanics it’s that this would create shitloads of drag that would make keeping it in orbit very difficult.
Since you seem to actually know about this shit, how do you think it would be possible to cool this thing?
Long answer; You can if you’re willing to basically devote the entire economic output of a large country to the problem.
Here’s the thing, putting aside cooling, the entire notion of a data-centre in space is insane. Falcon Heavy is about the most efficient launch vehicle we have right now, and it still costs $1500/kg that you send up. A fully loaded data centre rack can weigh around 1,000kg. Almost all of that weight is that actual hardware in the rack; y’know, the computers and hard drives that are the data centre.
So, sending a single rack to orbit costs $1.5m. A very small data centre might contain around 20 racks. The ones being used for modern AI workloads and the like are more in the 50,000 - 100,000 range. But even if we keep this tiny, super boutique, only for data too important to keep on earth, you’re still looking at $30m just to put the actual hardware into orbit.
That sounds OK, but that is only a tiny fraction of our costs. This is all going to snowball massively. On earth those racks are cooled by massive industrial HVAC systems that each have their own standby generator as well as the astonishing amount of power they pull from the grid. That works because they can circulate cool air around the racks, blast it out into the atmosphere, then pump in fresh air that you cool in the HVAC. You have none of that in space.
So instead you’re stuck with radiating heat through massive heat sinks with massive arrays of fins. And you have to get the heat from each individual computer, with all their really hot components, out to the heat sinks. That means you have to liquid cool every single component in this orbital data centre. Thousands of CPUs, thousands of hard drives, all liquid cooled. Then your liquid cooling has to run through unimaginably large heat sinks and radiators. At a wild guess I would bet that the total weight of all this cooling equipment (heat sinks are solid metal, and liquids are heavy and hard to fly into space because they shift around) would probably be a hundred times that of the equipment being cooled. So you’re talking about billions of dollars just in hardware to orbit costs, across thousands of launches.
And then you have to actually assemble everything. That means you need engineers who are also trained to work in orbit (so, very highly paid), and you need to get them up there. Since there’s nowhere for them to stay during construction, that means they have to go up, do a few hours work, and then come back down. Eight hour EVAs are not unheard of, so in theory your guys can do a full shift up there, but holy shit you have just invented the world’s most expensive commute by many orders of magnitude. It takes months to years to get a data centre up and running, and that’s one that doesn’t have all of these added complexities. Plus, working in space is really, really slow compared to working on Earth. You’re in a clumsy suit, wearing clumsy gloves, in an environment where nothing moves likes it’s supposed to and where you can never put anything down because it’ll just float away. Building something like this would take years of daily launches. You can’t just pre-build the components and send them up either, because everything is so ridiculously heavy that even a small chunk would exceed the weight limit of any launch vehicle we have today.
Oh, and going into space is really taxing on the human body, so you’d have to give those engineers lots of breaks, meaning you’d probably need to cycle different teams in and out for this whole thing, so that runs up your costs even higher.
And then what happens when something breaks? Liquid cooling needs constant maintenance, it’s very fiddly stuff. And hard-drives fail. Your average data centre will be swapping out a few drives every day. Even a small one is going to need a drive replaced every few weeks or months. Every time that happens someone has to go up there. You can’t just call Ted and tell him to hop in his Civic.
But we still haven’t gotten to the biggest problem yet. Power. Data centres use a truly staggering amount of power, between the computers and the cooling. Right now data centres, on their own, account for almost 5% of all power usage in the US. That’s fucking insane. So you need to somehow power everything you send up there. Powering things like space stations and communications satellites works because we build them to be very, very efficient. Even communications satellites, which have to process huge amounts of data, use between 1,000 and 5,000 watts. A single server rack, by comparison, can consume between 5,000 and 10,000 watts. So that’s 2-5 communication satellites worth of power for one rack. And we said that our absolutely tiny data centre needs twenty of those (and, again, I really need to drive home how small that is; that’s not a data centre, it’s a single room in a low-end corporate HQ). There is absolutely no way you’re going to strap enough solar panels to this thing to generate the kind of power it needs. Not without increasing the weight and construction time by another factor of one hundred. So now you need nuclear power of some kind… Which generates huge amounts of heat. So now you have to radiate that heat. Which increases the weight and construction time by another hundred-fold.
When all is said and done, we’re talking about high billions to low trillions of dollars to build a data centre that could fit in an apartment. Why? What could be possibly be worth that? Even if you were to make that argument that someone has data so valuable that it couldn’t possibly be kept on Earth, that still doesn’t make sense. On Earth you could, for a fraction of that price, bury that data in a vault deep underground or put it on an island or store it deep in the arctic where the environment makes it difficult to even approach (and solves your cooling costs). And in all of those locations, with that kind of money to throw around, you could hire a small army to protect it. Whereas in space, ultimately your precious data is just sitting there, basically unprotected. If it’s worth that much, then it’s worth it for a state-level actor with launch capabilites to send a few guys up to steal it.
This is a wild pipe-dream cooked up by silicon valley tech-bros who didn’t consult a single engineer in the process.
Edit to add: In the article the company behind this claims they’re going to use robots to do all the construction, and that it will be powered by solar panels multiple kilometres wide. Again, given everything I just said about the cost of putting that much material in orbit, vs the actual benefits, there is literally no way the economics of that works. Sure, you can knock out some of the costs I’ve listed, but you’re still basically taking the cost of a tiny data centre and massively amplifying it for absolutely no benefit. At best I suspect they’re just trying to raise their profile by making sensational claims.
Re: 4
Very, very common misconception, because of how often you see things/people in movies instantly freeze in space. But it’s just not remotely true.
The analogy the previous user gave is perfect; space is a thermos flask. It’s a perfect insulator.
To break that down a little more, you have to understand that heat moves in two basic ways; conduction and radiation. Conduction is when molecules agitate the molecules next to them. Radiation is when molecules give off electromagnetic energy.
The way a thermal camera works is that it sees the otherwise invisible infra-red light that hot things give off. That’s the radiation part of heat transfer. Radiation is, on the whole, a really slow, really bad way of moving heat.
Conduction is much faster, especially when there’s a big difference in temperature between the two mediums. That’s why you (average temp around 37C) can stand in a 21C room and feel really comfortable. You’re losing thermal energy, because the air touching your skin is colder, but you’re losing it at about the same rate your body naturally makes it.
But if you step outside into air that’s -20C, your temperature is going to start dropping very fast. There’s a much, much bigger difference in temperature now, so the heat transfer is faster. Also that air is probably moving because of the wind, which means the parts of the air getting warmed by the transfer from your skin are instantly replaced by fresh, cold air.
In space you have none of that. Just vacuum. There’s no molecules in vacuum to agitate. So aside from the very small amount you lose from radiation, heat just builds up. This is a huge problem for spaceships and satellites. They have to build in massive fins to help radiate heat away faster.
But it gets worse, because you know what radiates heat really, really well? The Sun. Which you are now exposed to, whenever you’re not directly in Earth’s shadow, with no atmosphere to absorb any of that incoming radiation. So the biggest problem for objects in space is rarely getting too cold, and far more often it’s getting too hot.
Introducing something that already has massive cooling requirements into that environment would be a total fucking nightmare.
Thanks for the very thorough breakdown.
This seems SUPER problematic, hahahah.
I’m wondering if you could drag something into earths high atmosphere to conduct heat away from the data center but if Anathem and Seveneves taught me anything about orbital mechanics it’s that this would create shitloads of drag that would make keeping it in orbit very difficult.
Since you seem to actually know about this shit, how do you think it would be possible to cool this thing?
Short answer? You can’t.
Long answer; You can if you’re willing to basically devote the entire economic output of a large country to the problem.
Here’s the thing, putting aside cooling, the entire notion of a data-centre in space is insane. Falcon Heavy is about the most efficient launch vehicle we have right now, and it still costs $1500/kg that you send up. A fully loaded data centre rack can weigh around 1,000kg. Almost all of that weight is that actual hardware in the rack; y’know, the computers and hard drives that are the data centre.
So, sending a single rack to orbit costs $1.5m. A very small data centre might contain around 20 racks. The ones being used for modern AI workloads and the like are more in the 50,000 - 100,000 range. But even if we keep this tiny, super boutique, only for data too important to keep on earth, you’re still looking at $30m just to put the actual hardware into orbit.
That sounds OK, but that is only a tiny fraction of our costs. This is all going to snowball massively. On earth those racks are cooled by massive industrial HVAC systems that each have their own standby generator as well as the astonishing amount of power they pull from the grid. That works because they can circulate cool air around the racks, blast it out into the atmosphere, then pump in fresh air that you cool in the HVAC. You have none of that in space.
So instead you’re stuck with radiating heat through massive heat sinks with massive arrays of fins. And you have to get the heat from each individual computer, with all their really hot components, out to the heat sinks. That means you have to liquid cool every single component in this orbital data centre. Thousands of CPUs, thousands of hard drives, all liquid cooled. Then your liquid cooling has to run through unimaginably large heat sinks and radiators. At a wild guess I would bet that the total weight of all this cooling equipment (heat sinks are solid metal, and liquids are heavy and hard to fly into space because they shift around) would probably be a hundred times that of the equipment being cooled. So you’re talking about billions of dollars just in hardware to orbit costs, across thousands of launches.
And then you have to actually assemble everything. That means you need engineers who are also trained to work in orbit (so, very highly paid), and you need to get them up there. Since there’s nowhere for them to stay during construction, that means they have to go up, do a few hours work, and then come back down. Eight hour EVAs are not unheard of, so in theory your guys can do a full shift up there, but holy shit you have just invented the world’s most expensive commute by many orders of magnitude. It takes months to years to get a data centre up and running, and that’s one that doesn’t have all of these added complexities. Plus, working in space is really, really slow compared to working on Earth. You’re in a clumsy suit, wearing clumsy gloves, in an environment where nothing moves likes it’s supposed to and where you can never put anything down because it’ll just float away. Building something like this would take years of daily launches. You can’t just pre-build the components and send them up either, because everything is so ridiculously heavy that even a small chunk would exceed the weight limit of any launch vehicle we have today.
Oh, and going into space is really taxing on the human body, so you’d have to give those engineers lots of breaks, meaning you’d probably need to cycle different teams in and out for this whole thing, so that runs up your costs even higher.
And then what happens when something breaks? Liquid cooling needs constant maintenance, it’s very fiddly stuff. And hard-drives fail. Your average data centre will be swapping out a few drives every day. Even a small one is going to need a drive replaced every few weeks or months. Every time that happens someone has to go up there. You can’t just call Ted and tell him to hop in his Civic.
But we still haven’t gotten to the biggest problem yet. Power. Data centres use a truly staggering amount of power, between the computers and the cooling. Right now data centres, on their own, account for almost 5% of all power usage in the US. That’s fucking insane. So you need to somehow power everything you send up there. Powering things like space stations and communications satellites works because we build them to be very, very efficient. Even communications satellites, which have to process huge amounts of data, use between 1,000 and 5,000 watts. A single server rack, by comparison, can consume between 5,000 and 10,000 watts. So that’s 2-5 communication satellites worth of power for one rack. And we said that our absolutely tiny data centre needs twenty of those (and, again, I really need to drive home how small that is; that’s not a data centre, it’s a single room in a low-end corporate HQ). There is absolutely no way you’re going to strap enough solar panels to this thing to generate the kind of power it needs. Not without increasing the weight and construction time by another factor of one hundred. So now you need nuclear power of some kind… Which generates huge amounts of heat. So now you have to radiate that heat. Which increases the weight and construction time by another hundred-fold.
When all is said and done, we’re talking about high billions to low trillions of dollars to build a data centre that could fit in an apartment. Why? What could be possibly be worth that? Even if you were to make that argument that someone has data so valuable that it couldn’t possibly be kept on Earth, that still doesn’t make sense. On Earth you could, for a fraction of that price, bury that data in a vault deep underground or put it on an island or store it deep in the arctic where the environment makes it difficult to even approach (and solves your cooling costs). And in all of those locations, with that kind of money to throw around, you could hire a small army to protect it. Whereas in space, ultimately your precious data is just sitting there, basically unprotected. If it’s worth that much, then it’s worth it for a state-level actor with launch capabilites to send a few guys up to steal it.
This is a wild pipe-dream cooked up by silicon valley tech-bros who didn’t consult a single engineer in the process.
Edit to add: In the article the company behind this claims they’re going to use robots to do all the construction, and that it will be powered by solar panels multiple kilometres wide. Again, given everything I just said about the cost of putting that much material in orbit, vs the actual benefits, there is literally no way the economics of that works. Sure, you can knock out some of the costs I’ve listed, but you’re still basically taking the cost of a tiny data centre and massively amplifying it for absolutely no benefit. At best I suspect they’re just trying to raise their profile by making sensational claims.
Great to learn about how this shit actually would work, thank you for taking the time write up such a thorough response!
Puts the idea into perspective for me.