Space is a vacuum. i.e. The lack-of-a-thing that makes a thermos great at keeping your drink hot. A satellite is, if nothing else, a fantastic thermos. A data center in space would necessarily rely completely on cooling by radiation, unlike a terrestrial data center that can make use of convection and conduction. You can't just pipe heat out into the atmosphere or build a heat exchanger. You can't exchange heat with vacuum. You can only radiate heat into it.
Heat is going to limit the compute that can be done in a satellite data centre and radiative cooling solutions are going to massively increase weight. It makes far more sense to build data centers in the arctic.
Musk is up to something here. This could be another hyperloop (i.e. A distracting promise meant to sabotage competition). It could be a legal dodge. It could be a power grab. What it will not be is a useful source of computing power. Anyone who takes this venture seriously is probably going to be burned.
The energy economics in space are also a bit more complicated than usually thought. I think Starlink has been using Si cells instead of III-V-based ones, but in addition to lower output they also tend to degrade faster under radiation. I guess that's ok if the GPU is going to be toast in a few years anyway so you might as well de-orbit the whole thing. But that same solar cell on Earth will happily be producing for 40+ years.
Also the same issue with radiative cooling pops up for space solar cells - they tend to run way hotter than on Earth and that lowers their efficiency relative to what you could get terrestrially.
That may be the plan, but this is also a great way for GDPR's maximum fine, based on global revenue, to bite on SpaceX's much higher revenue. And without any real room for argument.
Starlink and Falcon 9 have been an excellent pairing, Falcon 9 partially reusable rockets created a lot launch capacity and starlink filled the demand. Starship if it meets its goals will create more launch fully reusable supply by orders of magnitude, but there is not the demand for all that launch capacity. Starlink can take some of it but probably not all so they need to find a customer to fill it in order to build up enough to have the volume to eventually colonize mars.
Not going to read the article, because Data centers in space = DOA is common sense to me, however, did the article really claim cooling wasn't an issue? Do they not understand the laws of thermodynamics, physics, etc?
Sure, space is cold. Good luck cooling your gear with a vacuum.
Don't even get me started on radiation, or even lack of gravity when it comes to trying to run high powered compute in space. If you think you are just going to plop a 1-4U server up there designed for use on earth, you are going to have some very interesting problems pop up. Anything not hardened for space is going to have a very high error/failure rate, and that includes anything socketed...
> Not going to read the article, because Data centers in space = DOA is common sense to me, however, did the article really claim cooling wasn't an issue?
No. Nearly everyone that talks about data centers in space talks about cooling. The point of this article was to talk about other problems that would remain even if the most commonly talked about problems were solved.
It says:
> But even if we stipulate that radiation, cooling, latency, and launch costs are all solved, other fundamental issues still make orbital data centers, at least as SpaceX understands them, a complete fantasy.
Not disagreeing with you at all: that physics fact always come up. My honest question is: if it's a perfect thermos, what does, for example, the ISS do with the heat generated by computers and humans burning calories? The ISS is equipped with a mechanism to radiate excess heat into space? Or is the ISS slowly heating up but it's not a problem?
Massive radiators. In this photo[0], all of the light gray panels are thermal radiators. Note how they are nearly as large as the solar panels, which gives you an idea about the scale needed to radiate away 3-12 people's worth of heat (~1200 watts) + the heat generated by equipment.
The ISS has giant heat sinks[1]. Those heat sinks are necessary for just the modest heat generated on the ISS, and should give an idea of what a sattelite full of GPU's might require...
The equation has a ^4 to the temperature. If you raise the temperature of your radiator by ~50 degrees you double its emission capacity. This is well within the range of specialised phase change compressors, aka fancy air conditioning pumps.
Next up in the equation is surface emissivity which we’ve got a lot of experience in the automotive sector.
And finally surface area, once again, getting quite good here with nanotechnology.
Yes he’s distracting, no it’s not as impossible as many people think.
I'm convinced that >30% of this comes from ideas leaking out of fiction such as like Neuromancer, and percolating through the minds of wealthy people attracted to some of the concepts. Namely, the dream of being a hyper-wealthy dynasty, above any earthly government, controlling an extraterritorial Las Vegas Fiefdom In Space. (Which in the book, also hosted a powerful AI.)
Then they work backwards, trying to figure out some economic engine to make it happen. "Data centers" are (A) in-vogue for investment right now and (B) vaguely plausible, at least compared to having a space-casino.
That's not fair! Sometimes the ideas come from Snow Crash, which gave us the Metaverse because Zuckerberg wanted to cut a guy in half with a katana from a motorcycle.
I mean definitely, and they're not shy about admitting it. They see cool stuff in imagination-land, think it's cool, and work to make it a reality. Many people have worked to make the fantastical things shown in Star Trek.
I’ve come to think of interviews with people like Sam Altman as “freestyle science fiction.” They’re just saying stuff off the top of their head. Like you say, that often entails vague ideas from other sci fi percolating up and out, with no consideration of if they actually make sense. And like most freestyle, it’s usually pretty bad.
That is possible because DOGE and their comrades gutted the SEC and indirectly FINRA like a fish. The government is run by confidence men running crypto scams.
That’s how the CFO of OpenAI can essentially say “we need a Federal bailout”, and then turn around and say “lol just joking”.
A lot more. Everyone is chasing scifi ideas, ridiculous. This shows that even people with high IQ lack fantasy/imagination and creativity. They are intelligent robots.
So whenever I see here or anywhere else that your ideas mean nothing I just laugh at it. Of course, these come from people who are bland, doesn't have any imagination and they are not creative at all at all, but they have brute force, which is money.
One way to work around the heat dissipation issues in space (and also on earth) is to move to computing systems that operate entirely at cryogenic temperatures to take advantage of superconducting circuitry.
I've heard stories that over a decade ago teams inside hyperscalars had calculated that running completely cryogenically cooled data centers would be vastly cheaper than what we do now due to savings on resistive losses and the cost of eliminating waste heat. You don't have to get rid of heat that you don't generate in the first place.
The issue is that at the moment there are very few IC components and processes that have been engineered to run at cryogenic temperatures. Replicating the entirety of the existing data center stack for cryogenic temps is nowhere near reality.
That said, once you have cryogenic superconducting integrated circuits you could colocate your data centers and your propellant/oxidizer depots. Not exactly "data centers off in deep space" since propoxd tend to be the highest traffic areas.
Very confused by this plan. Data centers on Earth are struggling with how to get rid of waste heat. It's really, really hard to get rid of waste heat in space. That seems to be about the worst possible place to put a data center.
It’s a distraction as they suck out as much value from Tesla as possible before the music stops and they go bust. There are a few really big IPOs this year including SpaceX, which will likely trigger significant market volatility.
Indeed. I would go so far as to assert that, of all the ideas that have ever been proposed in the history of humanity, data centres in space is most certainly one of them.
Yeah he only micromanages (look at his old blog) every detail he has time for at an extremely successful aerospace engineering company, just an ideas guy.
> Yeah he only micromanages (look at his old blog) every detail he has time for at an extremely successful aerospace engineering company, just an ideas guy.
Have you ever spoken to someone who works at SpaceX? I have multiple friends in the industry, who have taken a trip through the company.
The overwhelming consensus is that - in meetings, you nod along and tell Elon "great idea". Immediately after you get back to real engineering and design things such that they make sense.
The folks working there are under no delusion that he has any business being involved in rocket science, it's fascinating that the general public doesn't see it that way.
Why are they doing any better than any other firm then? Why has Tesla been successful? Why is xAI pretty similar in terms of approach? My idea has less variables than yours. It also doesn't fly with his tendency to fire people.
What kind of the problem you're talking about compared to existing satellites? That is, all existing satellites generate power, and need to dissipate that power, and most of it goes to waste heat, and the satellites somehow do that successfully - what is the specific problem you're talking about, which can't be solved by the same means?
The numbers matter. The thermal budget a satellite is an tightly controlled thing. Large modern ones are in the order of a few to a couple of 10s of kilowatts, so something like a few to several low 10s of modern GPU compute power. Even with thousands of yet to be designed or launched satellites, it's going to have trouble competing with even a single current DC, plus it is in SAPCE for some reason, so everything is more expensive for lots of reasons.
> it's going to have trouble competing with even a single current DC
This looks like a valid argument to me, yes. Elon mentioned 1,000,000 satellites - I'm thinking about 3rd version of Starlink as a typical example, 2 tons, 60 satellites per Starship launch, 16,000 Starship launches for the constellation, comparing with 160 launches per year of today's Falcon 9...
The argument about problems of dissipating heat still stands - I don't see a valid counterargument here. Also "SAPCE" problem looks different from the point of view of this project - https://www.50dollarsat.info/ . Basically, out launch costs go way down, and quality of electronics and related tech today on Earth is high enough to work on LEO.
Even the buses for giant communications satellites are still at the single digit kilowatt scale. The current state of the art in AI datacenters is 500+ kw per rack.
So you're talking about an entirely different scale of power and needed cooling.
The reason we dont have a lot of compute in space, is because of the heat issue. We could have greater routing density on communication satellites, if we could dissipate more heat. If Starlink had solved this issue they would have like triple the capacity and could just drop everything back to the US (like their fans think they do) rather than trying to minimise the number of satellites traffic passes through before exiting back to a ground station usually in the same country as the source. In fact, conspiratorially, I think thats the problem he wants to solve. Because wet dreams of an unhindered, unregulated, space internet are completely unanswered in the engineering of Starlink.
I have actually argued this from the other side, and I reckon space data centres are sort of feasible in a thought experimental sense. I think its a solvable problem eventually. But heat is the major limiting factor and back of the napkin math stinks tbh.
IIRC the size/weight of the satellite is going to get geometrically larger as you increase the compute size due to the size of the required cooling system. Then we get into a big argument about how you bring the heat from the component to the cooling system. I think oil, but its heavy again, and several space engineering types want to slap me in the face for suggesting it. Some rube goldberg copper heatpipe network through atmosphere system seems to be preferred.
I feel like, best case, its a Tesla situation, he clears the legislative roadblocks and solves some critical engineering problem by throwing money at it, and then other, better people step in to actually do it. Also triple the time he says it will take to solve the problem.
And then, ultimately, as parts fail theres diminishing returns on the satellite. And you dont even get to take the old hardware to the secondary market, it gets dropped in the ocean or burnt up on reentry.
Principally speaking, as much energy as satellite receives from solar panels it needs to send away - and often a lot of it is in the form of heat. So, the question is, how much energy is received in the first place. We currently have some quarter of megawatt of solar panels of ISS, so in principal - in principal - we know how to do this kind of scale per satellite. In practice we perhaps will have more smaller satellites which together aggregate the compute to the necessary lever and power to the corresponding level.
The ISS's radiators weigh thousands of kilograms to radiate around 70 KW. He's talking about building data centers in space in the GW range.
Assuming he built this in LEO (which doesn't make sense because of atmospheric drag), and the highest estimates for what starship could one day deliver to LEO (200 metric tons), and only 1 metric ton of radiators per 100KW, that's 50 launches just to carry up the radiators.
Well the issue is that a lot of people believe that space is cold. If you will ask Google/Gemini what is a temperature of space, it will tell you:
The average temperature of deep space is approximately -270.45°C or 2.73 Kelvin), which is just above absolute zero. This baseline temperature is set by the Cosmic Microwave Background (CMB) radiatio...
Which is absolute nonsense, because vacuum has no temperature.
That's not how it works. Two bodies are in thermal equilibrium if there's no heat transfer between them: that's the zeroth law of thermodynamics. If you're colder than 2.73K in deep space, you will absorb the heat from the Cosmic Microwave Background. If you're hotter, you will irradiate heat away. So it does have a temperature.
It has nothing to do with the movements of atoms, but just with the spectrum of photons moving through it. It means that eventually, any object left in space will reach that temperature. But it will not necessarily do it quickly, which is what you need if you're trying to cool something that is emitting heat.
Well it isn't a perfect vacuum and it does have a temperature. But temperature is only a part of the story, just like how you go hypothermic a lot faster in 50 degree water than in 50 degree air.
but if you did use thermometer in space it would eventual read 2.73 kelvin right? so whats the issue? and also for a space based server it would have to deal with the energy coming from the sun
You guys clearly didn't read the full blog post where Musk mentions lunar mining. They're going to put an ASML machine on the moon and turns regolith into chips and solar panels automatically. Literally free compute
As a thought experiment, if humanity wanted to go all in on trying to move industrial processes and data centers off planet, would it make more sense to do so on the moon?
The moon has:
- Some water
- Some materials that can be used to manufacture crude things (like heat sinks?)
- a ton of area to brute force the heat sink problem
- a surface to burry the data centers under to solve the radiation problem
- close enough to earth that remote controlled semi-automated robots work
I think this would only work if some powerful entity wanted to commit to a hyper-scale effort.
The elephant in the room for all lunar scenarios is lunar regolith. Even ignoring the toxicity to humans (big problem and will happen quite quickly for any humans there!), it will be a big long-term problem for robots and machinery in general.
Water on the moon is limited and difficult to collect, it wouldn't make sense to use it for industrial purposes. It's a very challenging thermal environment (baking during the day, freezing at night). But perhaps worst of all, every month there's a 14-day period with no solar power. Overall seems worse than low-earth orbit.
Probably a lot easier, but the moon looses a major selling point of data centres in space, namely reasonable latency. To be clear, I don't think it's a good idea. But I think that specifically the way Musk is trying to position it, the moon would be an even harder sell.
What if instead we moved it all to a closer rock that has even more water, even more materials to manufacture crude (and even advanced) things, even more surface, more protection from radiation, and even crazier still had significantly less launch costs?
Almost any reason why the moon is better than in orbit is a point for putting it on earth.
I think there's something to be said about imagining a future where we can keep the earth clean of all the nasty industrial processes we have grown accustomed to living next to. A big part about this proposed idea is that you could do a lot of manufactoring in space.
I have long theorized there will be some game changing manufacturing processes that can only be done in a zero gravity environment. EX:
- 3d printing human organ replacements to solve the organ donor problem
- stronger materials
- 3d computer chips
I do not work in material science, so these crude ideas are just that, but the important part I'm getting at is that we can make things in space without any launches once that industry is bootstrapped.
They're gonna propose something dumb like ejecting coolant out into space as a disposable heatsink and then they're gonna spend a bunch of money trying to build a proof-of-concept but it will never go anywhere because it's really some kinda money laundering scheme or whatever the Hyperloop nonsense was.
> Data centers in space only make sense if they are cost effective relative to normal data centers.
Author made a fatal mistake. By flying enough hardware in space, you can simply blot out the sun and steal their solar capacity. Drink their milkshake with a long straw!
AI data-centers use upwards of 100MW. The biggest solar panels in space could produce around 240KW. When they speak of AI data-centers in space what do they actually mean in realistic non theoretical terms and where are the materials for this coming from?
If the AI data-center used only 10MW then each could have two redundant SMR's assuming the cooling challenges have been worked out but then we could have nuclear reactor disposal and collision issues.
I assume the idea is to have the entire constellation be the data center in question. Laser back haul transceiver bandwidth is in the same order of magnitude of rack-to-rack bandwidths [1][2]. I could see each sat being a rack and the entire mesh being a cluster.
This is how Starlink works however, you would need orders of magnitude more compute than those router pucks. Orders of magnitude more power needs unless you combined a nuclear reactor to it. It’s just such a fever dream at this stage that he’s really doing it to muddy accounting and consolidate debts from Grok failures.
The only thing that keeps bouncing around in my thick skull is something "data-center sized" whatever that means to them could hold some interesting objects. 2 Peta Watt laser, Rods from God, Tactical nukes, Miniature Rail-gun to quickly eradicate other satellites, Off-Planet archives of stuff, Doomsday clusters of brainwave transmitters to shut off all the humans or force everyone to defecate at once.
Those are just some guesses. Some of those could also explain the "why" for SpaceX Falcon Heavy and it's future iterations. It can carry 63,800 kg (140,660 lbs) to Low Earth Orbit and that load capacity will only increase with future versions.
How about we just make a giant heatsink that reaches into space instead. Then we can cool the whole planet. Coming up with crazy ideas is cheap, but the logistics are obviously impractical.
> Our core innovation is a radiative cooling material that we’ve combined with a panel system to improve the efficiency of any vapor-compression based cooling system
A heat pump is a “ vapor-compression based cooling system” so that tech is an addition-to not an instead-of.
Whether it’s better probably depends on how expensive the additional efficiency is in practice.
> SkyCool’s Panels save 2x – 3x as much energy as a solar panel generates given the same area.
It possibly makes sense if you're preparing for war, harder to hit, harder to physically break into, beyond the range of nuclear EMP, and accessible from anywhere on earth.
Any country capable of producing nuclear warheads will also be able to toss up enough BBs and other small objects into LEO to wipe out most of Starlink and anything else in LEO. At least on Earth data centers in theory can be hidden and physically hardened. In orbit, even a crude rocket able to reach that plane can become a weapon of mass satellite destruction. Even if those orbits clear out in four or five years, by then whatever ugliness is going on down on the surface of Earth will likely have resolved one way or the other. Starlink is a great military asset for a superpower pushing around smaller states in ways that aren't an existential threat to them. In a real conflict, it's a fragile target beyond the strike capacities of much of the developing world but easily destroyed by any moderate level industrial nation.
Any country capable of producing nuclear warheads will also be able to toss up enough BBs and other small objects into LEO to wipe out most of Starlink and anything else in LEO.
Pakistan doesn't have a domestic orbital launch capability but it does have nuclear weapons.
Surprisingly, the United Kingdom doesn't have a domestic orbital launch capability at present though it has had ballistic missiles and nuclear weapons for many decades.
At present, I would say that building a basic implosion-assembled atomic bomb is easier than building a rocket system that reach low Earth orbit. It's a lot easier to build a bomb now than it was in the 1940s. The main thing that prevents wider nuclear weapon proliferation is treaties and inspections, not inherent technical difficulties.
Satellites. Are. Fragile. People really don’t seem to intuitively understand this. Earth based assets are orders of magnitude more difficult to attack simply by virtue of being able to be placed inside of fortified structures anchored to, or inside of, the ground. The cost to deploy hardened buildings at scale is peanuts compared to orbiting constellations.
They also fail to realize how devastating an attack a BB canister grenade would be in LEO. Nothing would stay in orbit. Eventually everything would collide and come down.
You don't need EMP for that. Few ASAT missiles will start the avalanche and turn orbits around Earth into shooting range. Good luck talking to your satellites with shredded antennas and solar panels.
The bigger issue: datacenters in space are disposable. All the extremely recyclable aluminum, silica - you extract it, manufacture it and instead of recycling it when it’s done you incinerate it in the atmosphere and scatter the ashes far and wide across the earth, the harder to recapture later.
You do this when the most fragile part in the system fails. Solar panels good for 25 years but the SSDs burn out after 2? Incinerate the lot!
This kind of thinking is late capitalist brain rot. This kind of waste should be a crime.
To Steelman the topic, Musk’s whole alleged mission is to make humans a multi-planet species that can survive an earth killing event.
To that end, a small data center space isn’t about unit-economics, it’s a bigger mission. So the question we should consider is what can we put into space the further that mission. Can we put a meaningful sum of human knowledge out there for preservation? It sounds like “yes,” even if we can’t train ChatGPT models out there yet.
When I was a kid, I had to go to CCD, a religious after school program for Catholics.
The whole time I was there it was a mental game of trying to steel man the contradictory or incoherent stuff, using my brain power to try and rewrite things to make sense.
After some years, I woke up and realized that’s what I was doing, and even if I could do it in my mind, that didn’t make the source material rational.
I do not politically align with Musk. I’ve always thought Tesla was important in popularizing electric cars while being a low-quality built product with repair and supply chain issues. I think The Boring Company is a joke. Twitter was a power-grab.
I also think SpaceX is societally beneficial, a good means to shake-up a stagnant industry and a humanity-wide area of interest.
If you think I’m a member of a religious cult, I respectfully suggest you evaluate what led
You to believe that itself.
The problem of datacenters in space and knowledge preservation/disaster redundancy are entirely disjoint.
Datacenters in space have a lifespan measured in years. Single-digit years. Communicating with such an installation requires relatively advanced technology. In an extinction level crisis, there will be extremely little chance of finding someone with the equipment, expertise, and power to download bulk data. And don't forget that you have less than a decade to access this data before the constellation either fails or deorbits.
Meanwhile people who actually care about preserving knowledge in a doomsday crisis have created film reels containing a dump of GitHub and enough preamble that civilizations in the far future can reconstruct an x86 machine from scratch. These are buried under glaciers on earth.
We've also launched (something like) a microfilm dump of knowledge to the moon which can be recovered and read manually any time within the next several hundred or thousand years.
Datacenters in space don't solve any of the problems posed because they simply will not last long enough.
Let's say there is an earth killing event, and let's say there is an outpost on Mars with some people on it. How much does it really matter that some humans survive, in light of the enormous catastrohophe that killed all life on earth? Is it a very worthwhile objective for our species to persist a while longer, or should we not just accept that also life itself will will die out on geological or astronomical time scales?
I would suppose there is a gap we face between true species-wide survival capability and where we sit today. I have no true idea how hard we must go to bridge that gap, but it’s quite hard and far.
I also see no reason to “lay down and die” as I feel is somewhat implied here. I think it’s a truly noble cause, but maybe I read too much sci-fi as a young lad.
No matter what anyone does, the universe will end, and reality will stop changing.
Everything dies. Deal with it.
Instead of empowering shithead grifters who promise you a way out, grow trees to create shade for people you will never know. You do that by improving things, not burning limited resources on a conman.
I’m not the right type of engineer to know and, hell, software largely isn’t engineering anyway…
Can you not provide any type of shielding at scale to wrap a (small, not Google tier) data center? To be honest my criticism with TFA is its focus on “you can’t do massive scale” rather than the premise entirely.
Yes, but the added mass makes it prohibitively expensive. Shielding is heavy and every kilogram of added payload results in a geometric increase in fuel load.
The rocket equation will kick your ass every time.
Musk's whole mission is to scam even more people. Unfortunately people still buy his bullshit even though he couldn't deliver on anything, and just converts one failure to hyping up his next idiotic product.
Is there any insight into how Starlink solved cooling? One 'expert' insisted that there is no reason to expect that data center satellites would generate any more heat than starlinks.
I am willing to bet the whole xAI/SpaceX merger is simply a ploy by Musk to evade releasing accurate historical information about SpaceX's finances. How much did it actually cost SpaceX to launch a kilogram of payload into space each year? How much is NASA actually donating them, per each year?
I mean, I still remember promises of $1000-per-kg for space launches, and how e.g. Gigafactory will produce half of the world battery supply, and other non-scientific fiction peddled by Musk. Remember when SpaceX suggested in 2019 that the US Army could use its Starship rockets to transport troops and supplies across the planet in minutes? I do. By the way, have they finished testing Starship yet, is it ready?
It doesn't make any sense to me either, but there are lots of things like that where the other thing is harder. As an example, a thing people say online a lot is something like "Why do the techbros build self-driving cars instead of just putting it on rails for efficiency and then they could call it a TRAIN?"
The answer to that is that coordination problems are really hard. Much harder even than what are currently unsolved engineering problems. In fact, SpaceX can only launch from California because they have DOD coverage for their launches. Otherwise the California Coastal Commission et al. would have blocked them entirely. Perhaps the innovation for affordable space Internet is combining it with mixed-use technology.
The truth is that in America today self-driving cars (regulated by a state board run by bureaucrats) are easier to build than trains (regulated by every property owner on the train route). Mark Zuckerberg tried to spend some money evaluating a train across the Bay and had to give up. But Robotaxi service is live in San Francisco.
So if there is an angle that makes sense to me it's that they anticipate engineering challenges beatable in a way where regulatory challenges are not.
Interesting insight. I can think of some objections, but they don't change your point.
I also checked out your blog and got 2 interesting articles in 2 tries. If you have some personal favourites and listing them is not a bother, I'd be happy to read them.
No no, let Musk cook. This definitely won't be SpaceX's Cybertruck moment, where they completely throw away their first-mover advantage by wasting five years chasing after the egotistical boondoggle of a delusional megalomaniac.
There are two very distinct kinds of AI workloads that go into data centres:
1. Inference
2. Training
Inference just might be doable in space because it is "embarrassingly parallel" and can be deployed as a swarm of thousands of satellites, each carrying the equivalent of a single compute node with 8x GPUs. The inputs and outputs are just text, which is low bandwidth. The model parameters only need to be uploaded a few times a year, if that. Not much storage is required , just a bit of flash for the model, caching, logging, and the like. This is very similar to a Starlink satellites, just with bigger solar panels and some additional radiative cooling. Realistically, a spacecraft like this would use inference-optimised chips, not power-hungry general purpose NVIDIA GPUs, LPDDR5 instead of HBM, etc...
Training is a whole other ballgame. It is parallelisable, sure, but only through heroic efforts involving fantastically expensive network switches with petabits of aggregated bandwidth. It also needs more general-purpose GPUs, access to petabytes of data, etc. The name of the game here is to bring a hundred thousand or more GPUs into close proximity and connect them with a terabit or more per GPU to exchange data. This cannot be put into orbit with any near-future technologies! It would be a giant satellite with square kilometers of solar and cooling panels. It would certainly get hit sooner or later by space debris, not to mention the hazard it poses to other satellites.
The problem with putting inference-only into space is that training still needs to go somewhere, and current AI data centres are pulling double-duty: they're usable for both training and inference, or any mix of the two. The greatest challenge is that a training bleeding edge model needs the biggest possible clusters (approaching a million GPUs!) in one place, and that is the problem -- few places in the world can provide the ~gigawatt of power to light up something that big. Again, the problem here is that training workloads can't be spread out.
Space solves the "wrong" problem! We can distribute inference to thousands of datacentre locations here on Earth, each needs just hundreds of kilowatts. That's no problem.
It's the giaaaant clusters everyone is trying to build that are the problem.
- Data centres need a lot of power = giant vast solar panels
- Data centres need a lot of cooling. That's some almighty heatsinks you're going need
- They will need to be radiation-hardened to avoid memory corruption = even more mass
- The hardware will be redundant in like 2 years tops and will need replacing to stay competitive
- Data centres are about 100x bigger (not including solar panels and heat sinks) than the biggest thing we've ever put in space
Tesla is losing market share (and rank increasingly poorly against alternatives), his robots are gonna fail, this datacentre ambition needs to break the laws of physics, grok/twitter is a fake news pedo-loving cesspit that's gonna be regulated into oblivion. Its only down from here on out.
Maybe instead of housing life, civilizations develop Dyson's spheres to house data centers. Solar panels on the interior, thermal radiators on the exterior and the data centers make up the structure in between. Combine that Von Neumann probes and you've got a fun new Fermi paradox hypothesis!
Don't combine it with von Neumann probes and you've solved the Fermi paradox: a civilization that puts that much work into computing power is either doing the equivalent of mining crypto and going nowhere, or is doing AI and is so dependent on it that they inevitably form a vast echo chamber (echo sphere?) that only wants to talk to itself (itselves?) and can't bear to be left out by adding the latency unavoidably added by distance.
tl;dr: civilizations advanced enough to travel between stars end up trapped by the resources and physics required to keep up with the Joneses.
I'd be curious to know simply how large the thermal radiator necessary to keep a typical GPU server cooled would be. Do they completely dwarf the server size? Can you do something with some esoteric material that is not particularly load-bearing but holds up well in space to get around some of these challenges?
Data centers in space make absolute sense when you want as close to real time analysis on all sorts of information. Would you rather have it make the round trip, via satellite to the states? Or are you going to build these things on the ground near a battlefield?
Musk is selling a vision for a MASSIVE government contract to provide a service that no one else could hope to achieve. This is one of those projects where he can run up the budget and operating costs like Boeing, Northrup etc, because it has massive military applications.
I can assure this author: strapping a company that lights money on fire (today, maybe not tomorrow) to a cash flow enterprise makes the IPO harder, not easier, in the absence of credible plan. The market speculates, but it’s not being completely irrational. I’d actually be surprised if we didn’t have factories or data centers in space one day.
Data centers in space are the logical progression from the multi trillion business of m2m and edge computing. It removes all physical limits to investment.
That post does not appear to address or acknowledge any of these problems: 1) thermal management in space, 2) radiation degrading the onboard silicon, 3) you can’t upgrade data centers in orbit
*Data centers in space only make sense if they are cost effective relative to normal data centers*.
Disagree there are bunch of scenarios where Data Centers in space make sense. Like nuclear annihilation and having vaults across the globe to communicate and get back lost information because ground data centers would be wiped out by EMP from blasts.
Has it occurred to anyone that you can put computers underground? In this apocalyptic scenario you are describing, how do you expect the ground based command and control infrastructure to survive? Satellites are 100% reliant on ground based operations. That is a hard requirement. And if you put the command and control underground, might as well just skip the whole space based plan and just put the data underground.
It was not my intention to single you out, my apologies.
There is nothing wrong to imagine anything you like. But if you do it as a CEO, i personally consider that as fraud. Guess I'm weird and old-fashioned like that.
Space offers some unique benefits that enable computing that’s impossible or very hard to do on earth. E.g. Super conducting computing is possible, which can be thousands times to millions times faster than current CPU while using very little energy. When the satellite moves in the shade of the earth, temperature drops significantly. It can be low enough to enable superconducting. When the satellite moves under the sun, the solar panel can start charging up the battery to power the ongoing operation.
i don't understand? you won't insulate the craft from the sun? and you expect the craft to get rid of its heat just from being behind the earth for a moment?
What’s there not to like? Superconductors. Free electricity. No cooling necessary.
Put those three together and maybe it’s possible to push physics to its limits. Faster networking, maybe 4x-5x capacity per unit compared to earth. Servicing is a pain, might be cheaper to just replace the hardware when a node goes bad.
But it mainly makes sense to those who have the capability and can do it cheaply (compared to the rest). There’s only one company that I can think of and that is SpaceX. They are closing in on (or passed) 8,000 satellites. Vertical integration means their cost-base will always be less than any competitor.
No, just you. Superconductors don’t get hot. There is 0 resistance in superconducting mediums. Theoretically you could manufacture a lot of the electricity conducting medium out of a superconductor. Even the cheapest kind will superconduct in space (because it’s so cold).
Radiation may be sufficient for the little heat that does get produced.
Right. You build your computers out of superconductors, and they don't get hot.
Sadly, they also don't compute.
> Even the cheapest kind will superconduct in space (because it’s so cold).
Is this a drinking game? Take a drink whenever someone claims that heat is not a problem because space is cold? Because I'm going to have alcohol poisoning soon.
Let's see how cold you feel when you leave the Earth's shadow and the sun hits you.
Do you mean to suggest that computer hardware does not need to be cooled when it is in space? Or that it is trivial and easier to do this in space compared to on Earth? I don’t understand either claim, if so.
Even assuming that this la-la-land idea has merit, the equilibrium temperature at the Earth's orbit is 250 Kelvin (around -20C). The space around the Earth is _hot_.
There are people literally working on accomplishing this. I don’t understand what’s with the arrogance and skepticism.
Edit: Not trying to single out the above commenter, just the general “air” around this in all the comments.
I honestly believed folks on HN are generally more open minded. There’s a trillion dollar merger happening the sole basis of which is the topic of this article. One of those companies put 6-8,000 satellites to space on its own dime.
It’s not a stretch, had they put 5 GPUs in each of those satellites, they would have had a 40,000 GPU datacenter in space.
> There are people literally working on accomplishing this.
They're reinventing physics? Wow! I guess they'll just use Grok AI to fake the launch videos. Should be good enough for the MVP.
For the superconductivity idea to work, the entire datacenter needs to be shielded both from sunlight and earthlight. This means a GINORMOUS sun shield to provide the required shadow. But wait, the datacenter will orbit the Earth, so it also will need to rotate constantly to keep itself in the shadow! Good luck with station-keeping.
There's a reason the Webb Telescope (which is kept at a balmy 50K) had to be moved to a Sun-Earth Lagrange point. Or why previous infrared telescopes used slowly evaporating liquid helium for cooling.
> I don’t understand what’s with the arrogance and skepticism.
Because it's a fundamentally stupid idea. Stupid ideas should be laughed out.
I'm not talking about "stupid because it's hard to do" but "stupid because of fundamental physical limitations".
