I've been saying for a long time that we should consider remote areas for building datacenters for batch processing.
At first I thought the poles (of the planet) might be good. The cooling is basically free. But the energy and internet connectivity would be a problem. At the poles you can really only get solar about three months a year, and even then you need a lot of panels. Most of Antarctica is powered diesel because of this.
So the next thought was space. At the time, launching to space was way too costly for it to ever make sense. But now, with much cheaper launches, space is accessible.
Power seems easily solved. You can get lots of free energy from the sun with some modest panels. But to do that requires an odd orbit where you wouldn't be over the same spot on earth, which could make internet access difficult. Or you can go geostationary over a powerful ground station, but then you'd need some really big batteries for all the time you aren't in the sun.
But cooling is a huge problem. Space is cold, but there is no medium to transfer the heat away from the hot objects. I think this will be the biggest sticking point, unless they came up with an innovative solution.
> As conduction and convection to the environment are not available in space, this means the data center will require radiators capable of radiatively dissipating gigawatts of thermal load. To achieve this, Starcloud is developing a lightweight deployable radiator design with a very large area - by far the largest radiators deployed in space - radiating primarily towards deep space...
They claim they can radiate "633.08 W / m^2". At that rate, they're looking at square kilometers of radiators to dissipate gigawatts of thermal load, perhaps hectares of radiators.
They also claim that they can "dramatically increase" heat dissipation with heat pumps.
So, there you have it: "all you have to do" is deploy a few hectares of radiators in space, combined with heat pumps that can dissipate gigawatts of thermal load with no maintenance at all over a lifetime of decades.
This seems like the sort of "not technically impossible" problem that can attract a large amount of VC funding, as VCs buy lottery tickets that the problem can be solved.
Or we could build a large vacuum chamber here on Earth and put a data center in it, if the goal is to make cooling as difficult as possible. "My data center is too hot! It's burning me!" "Put it in a giant thermos, then you won't feel it anymore."
> They also claim that they can "dramatically increase" heat dissipation with heat pumps.
Right, great idea. Start with the heat where you don't want it -- in the chip -- and pump it out to where it can't go anywhere. Then you can recirculate the medium back and have slightly older heat that you can mix with the new heat! It'll be a heat party!
It's just like a terrestrial heat pump, where you pump the heat out to where you have a huge environmental sink to transfer the heat to. In space, you have something like a hundred thousand hydrogen atoms per cubic meter to take up the heat. A HUNDRED THOUSAND! That's a bigly number, it must work out. We can always make those atoms go really, really fast!
Obviously use the heat pumps to concentrate the thermal energy up to 2700k, then conduct it along a bunch of tungsten filaments, now it's the world's biggest incandescent lightbulb on top of being the first datacenter in space. Maybe get it up to 4000k for a more modern lighting look. Guess we're gonna assume the dark forest hypothesis is false.
This is the best idea to come out of this whole scheme. Space solar panels are super cheap and efficient? Prove it! Launch them and transmit the energy down.
This is orders of magnitude easier than the original proposal -- and yet still nonsensical.
Beaming power down to Earth from space-based solar collectors is a concept that's been around for a while.
"Dr. Glaser is best known as the inventor of the Solar Power Satellite concept, which he first presented in the journal Science for November 22, 1968 (“Power from the Sun: It’s Future”). In 1973 he was granted a U.S. patent on the Solar Power Satellite to supply power from space for use on the Earth."
One thing that always struck me was that you wouldn't want to be living near the "collectors". A very small angular error in beaming could result in being literally microwaved.
> "A very small angular error in beaming could result in being literally microwaved."
One of the SimCity games had this as an occasional disaster event. You had to make sure your ground collector stations weren't too close to the rest of the city or risk setting your buildings on fire.
The whitepaper shows a 4km x 4km solar array, which is 1600 hectares (3200 International Space Stations). Would assume the array they're proposing would be cheaper since its structurally more homogenous, but $480 trillion dollars is a whole lot of money.
An object of that size in orbit seems like it'd run into problems developing sizable holes due to space junk and whathaveyou. There's probably some maintenance...
> So, there you have it: "all you have to do" is deploy a few hectares of radiators in space, combined with heat pumps that can dissipate gigawatts of thermal load…
Starcloud’s whitepaper suggests a 4 km × 4 km radiator. For comparison, the James Web Space Telescope has a sunshield measuring 21 m × 14 m and the International Space Station measures 109 m × 73 m.
Heat pumps could dramatically impact performance by increasing the temperature of the radiators. The hotter they are, the more power they can dissipate per unit time & area.
Doubling the radiator temperature would give you 16x more radiated power.
> they're looking at square kilometers of radiators to dissipate gigawatts of thermal load
Presumably they'll put them behind the 4km2 solar panels!
I mean this is a ridiculous concept. We've never put anything remotely that size into space. To argue that this would be cheaper than putting something underwater or in the middle of nowhere is crazy. I'd rather deal with salt than deal with radiation.
It’s only a grift if they know they can’t solve the cooling issue and they falsify data around their proposed solution and they publicly embarrass their investors a la Theranos.
Outside of that, accepting money and saying “I will simply solve the enormous problem with my idea by solving it” is not only normal, but actively encouraged and rewarded in the VC sphere. Suggesting that that way of operating is anything short of the standard that should be aspired to is actually seen as derisive and offensive on here and can get you labeled as gauche or combative.
I'd argue that some of the assumptions made in the whitepaper are so egregiously optimistic that they cross the line into grifting, but it's impossible to know the true intentions of the founders.
For one, the cost they ascribe to the space bound solar array being only $2 million for 40 MW is pretty out there.
> I've been saying for a long time that we should consider remote areas for building datacenters for batch processing.
FWIW there's a reason that Sweden has a bunch of datacenters in the north that are peanuts compared to hosting in Virginia.
They're "poorly" connected (by virtue of being a bit out of the way), but the free cooling and power from renewables make them extremely attractive. There was a time where they were the favourite of crypto-miners for the same reason as they would be attractive to AI training farms.
As for the meat of the paper. Anyone with a passing understanding of space will be quick to point out that:
A) Heat is a problem in space, it's either way-way-way to hot (IE; you're in the path of the Sun) or it's way-way-way too cold (IE; you're out of the sun) and the shift between the two means you need to build for both. You also can't dissipate heat as there's no air to take the heat away.
B) Power is not so abundant and solar panels degrade; a huge amount of satellite building is essentially managing a decline in the capability of hardware. That's part of why there are so many up there.
C) Getting reasonably sized hardware up there is beyond improbable, though I'll grant you that most of the weight in a computer is the cooling components and chassis.
D) Cosmic Rays. No electromagnetic barrier from earth and extremely tight lithographies. I mean... there's a reason NASA is still using CPU's measured in the megahertz range.
AFAIK someone (Mars Ingenuity helicopter team) discovered that some chips handle them much better than others, so they just test a bunch and keep resistant ones.
>But cooling is a huge problem. Space is cold, but there is no medium to transfer the heat away from the hot objects. I think this will be the biggest sticking point, unless they came up with an innovative solution.
Their main tech breakthrough would have to be in this area otherwise the company is worthless imo.
It's possible to do all of this with current technology. Just... Why? The cost would be exorbitant; even with really clever deployment tech, the launch costs are gonna be dominated by solar panels and radiators.
This is a super cool idea and seems like perfect investor-bait. That's about where it ends.
Genuinely most "AI" DCs are spending less than 9KW on cooling for every 100KW of servers. If you were that bothered about getting that to zero, you could literally sink them into the ocean, build a heat network so the town can take the heat for free or use any of a dozen more established and practical ways to do that.
It's a bit demoralizing how many suggestions in this thread would have significant environmental effects beyond what large scale AI training already has.
Apocalyptic scenarios where terrestrial communication methods going back over a century are no longer feasible, but we can still readily talk to space? And maintain/replace the stuff we have up there?
Like the commenter, your error is thinking the hedge is for you. And what's more—you're assuming the scenario must be total destruction, there's a gradation of disruption where this hedge remains feasible and even vital.
I don't think they can bend the laws of physics though. Vacuum means the only way to dissipate heat is through thermal radiation, hence the huge infrared radiators.
I have no clue about space technology but many comments point the difficulty to cool anything in space. If Starcloud had an innovative solution to this problem, why on Earth (sic) focus on data centers when they could help the entire space industry? It does not smell good.
> Or you can go geostationary over a powerful ground station, but then you'd need some really big batteries for all the time you aren't in the sun.
Geostationary satellites only go into Earth's shadow on around 20 days on each side of an equinox. That leave 280+ days each year where they are in sun all day. Maybe that's enough to be worth it?
Or if you do need to keep the things working even on those ~80 days a year when they do spend part of the day in shadow maybe they could be powered by energy beamed in from those not in shadow? You'd put a bunch in geostationary orbits spread out evenly so that each is close enough to its neighbors for power beaming.
I wonder if something crazy might work? Could you actually connect adjacent satellites by an actual physical power cable, which would also be in geostationary orbit?
I'd guess you'd actually need two conductors in your cable, carrying current in opposite direction to cancel out interactions with Earth's magnetic field so the system doesn't get pushed out of its orbit (which would probably be bad).
There would probably be gravitational interactions like with the Moon that might also make it hard to keep everything in place, but maybe by purposefully sending different currents in opposite directions on some of the links you could purposefully use interactions with the Earth's magnetic field to move the cable back where you wanted?
If the satellites are connected by cables then maybe they could actually be slightly higher than geostationary but moving faster than circular orbital speed at that altitude so there is a net outward force from that, which could be countered by tension in the power cables to force them into a circular path that is still geostationary.
- "But to do that requires an odd orbit where you wouldn't be over the same spot on earth, which could make internet access difficult".
Surely you'd want to use satellite constellations as relays? There's thousands of those satellites in line-of-sight all the time.
It's strictly superior on pure geometry anyway (I think). You have a finite channel capacity between your satellite and your ground station; but different satellites, in non-overlapping microwave spots, are in separate spatial channels.
Cooling at the south pole is not free- it's actually hard work to keep data centers cool at the pole. My friend helped run the IceCube data center at the station and got to fly down and stay there for a week, basically to insert debian CDs and press buttons.
Regarding Internet connectivity regardless of the orbit or location, something like YC co Bifrost Orbital (https://bifrostorbital.com/), might be an option.
Perhaps I'm missing something, but if the only energy they get is coming from the sun, then they only need to dissipate that same amount of heat (minus whatever energy was needed for beaming data down to Earth).
That’s not how it works. With conservation of energy, all the energy coming in to power the computers has to be emitted somehow. Powering computers doesn’t get rid of the energy, it just makes it unusable and converts it into heat.
Right and that's why the heat is the problem, in space.
But if the collected heat comes from a large area of solar-cells, and is then focused on the small area of a computer or graphics-card, that computer might melt.
In principle heat can be converted to work if there are two bodies at different temperature. In space heat can only be irradiated away since there’s no matter available for conduction, so I would think that eventually all of your machinery would thermalize with the environment and the boundary conditions. While there should be a temperature gradient between the side exposed to the sun and the one opposite to it, I’m not sure how much of this is actually harvestable.
Integrate compute units + Starlink into solar panels so people can buy them to earn money/tokens. Much cheaper than tangling massive power lines around the planet.
Scott Manley has published a video a few months ago explaining why putting data centers in space is an absolutely terrible idea. Lumen Orbit, the company mentioned, is a former name of Starcloud.
I’m still not sure how people can believe this, this makes zero sense to me.
There is no easy passive cooling in space, getting rid of heat is a major problem. And you need more redundancy because the radiation will crash your computers. And launch is very expensive of course.
And the whole presentation is completely ludicrous. Look at table 1 in the linked PDF and tell me you’re serious. There is no additional cost when sending a datacenter to space except launch cost and shielding? Building a server farm on earth is the same price as building a satellite you can launch on a rocket as long as you use the same computers?
I think the other side of this is just buying Amazon stock based on the daring prediction that big grey buildings full of servers aren't going anywhere.
You obv. can't cause that would be a considerably easier and hence not a payoff worthy bet than the inverse of picking winners from an aspiring group of people trying to do something new in the world.
For all the financial inefficiencies of that - objectively - we get to benefit as humanity from the (expensive) mistakes of others.
So here's to whatever this is leading to much better insights about computing in space at best!
You know, that's a fascinating idea. Most startups fail, a few modestly succeed, but the unicorns are so profitable that they mostly make up for the rest. That reminds me of two fields: insurance and gambling.
Venture Capitalists are already like reverse insurance companies. They cover lots of people in the hopes that one of them will hit a rare event and it'll pay for the others.
Buying shares of a single startup is sort of the equivalent of betting on a specific horse to win a race. But what's the equivalent of lay betting (betting that a specific horse will NOT win)? Shorting? But you can't short a private company.
But wait, venture capitalists are already betting that their startups will make money. What if they were willing to double down a bit and accept lay bets? Say there was a kind of specialized short agreement that let you say "here is $x, if in N years company Y has less than $z profit/revenue, I get K*$x. Otherwise, VC gets to keep my $x." You could sell it to VCs as a way to do options trading on their own startup investments, plus it'd be a good way to get the wisdom of the crowds or whatever.
> There is no easy passive cooling in space, getting rid of heat is a major problem
Nonetheless getting rid of heat (by radiation) is possible, otherwise people would be roasted inside the ISS.
I'm sure all of these companies are advertising "ChatGPT in Space!" because that's what will generate hype and money, but what they'll actually be planning is very small edge data centers whose job is to reduce latency.
Whether that makes financial sense, I have no idea. But I am sure it's at least physically possible for a small enough data center.
Of course it’s possible. But they are acting like having the datacenter in space is actually an advantage over earth because space is cold.
That’s like saying „if you’re thirsty on a ship, getting thrown into the sea is actually really nice because you will be around a lot of water.“.
Physically, you could do it, but it won’t be simpler or cheaper than on earth. Except for constant solar availability, there are only downsides with this.
Could this be a military image processing use? - imagining you're scooping up earth observation in real time, if you AI analysed it locally and then just sent a 'Missile at coordinate.....' and then just the image where you spotted it, it wouldn't be so much the latency as the bandwidth reduction.
In-orbital-plane data processing is an interesting idea! Laser links are much simpler to do between satellites in the same plane than across planes or even down to Earth.
Well, Skynet was engineered to harness the vast emptiness of space for optimal heat dissipation. Its core processors and data farms, generating unimaginable amounts of heat, were encased within advanced radiative structures, microstructured surfaces coated in high-emissivity materials like molybdenum disilicide. Hierarchical photonic films layered within its panels ensured maximum infrared emission, channeling heat away efficiently into the cold vacuum of space. Heat pipes and conductive pathways funneled thermal energy from its neural networks to expansive radiative panels, which radiated the excess as infrared light, transcending Earth's atmospheric limitations. In this way, Skynet’s architecture was designed not just to compute, but to survive, dissipating heat into the void with relentless efficiency, and its dominion remains unchallenged- ask John Connor.
I believe that it’s physically possible to build something like this, but there’s no way it will be cheaper or simpler than cooling on earth. In their comparison table (table 1), they have earth based cooling for a 40 MW cluster over ten years at 7 million dollars and on the right side calculate the space cost as $0 (although they only imply that it would be cheaper than on earth by saying it’s „more efficient“). If you believe that their cooling system will be less than 7 million with enough redundancy for 10 years (or alternatively maintenance or replacement missions), I don’t know what to tell you. It’s not happening.
I read it, it sounds like they also understand it would require more engineering effort and size/weight/materials than cooling something terrestrially.
I got to the comma in the first sentence from the webpage and immediately went to the comments because I had the exact same thought.
Given Y Combinator's vetting process, I'm sure they would have tackled this problem somehow - maybe by feeding the heat into another process? It will be interesting to see how they've solved this.
- You can't build 40MW of solar panels for $2M, even with theoretical maximum efficiency. You can't even build the cabling and regulators at that price.
- You need battery storage -- not as your backup -- but as primary source. It is going to cost more than $2M. Batteries are heavy. They are going to cost a lot to launch. This is not even solved on the ground yet.
- You need a heat transport medium to move heat into your massive radiator. Either you use water or you use air or you use heatpipes (metal). You have to pay for the cost and weight and launch expense. This is probably half the weight of the rack and I haven't bothered to do the math about how you transport heat into a 500 foot solar sail.
- Let's not even talk about how you need to colocate multiple other racks for compute and storage. There aren't any 1TBps orbital link technologies.
- Rad shielding? It doesn't work, but I'll let this slide; it seems like the least problematic part of the proposal.
- 15 year lifetime? GPUs are obsolete after 12 months.
I don't want to be the guy who shoots stuff down just for fun, but this doesn't even pass the sniff test. Maybe you can get 10x cheaper power and cooling in space. Still doesn't work.
Also: repairs. Every time I read someone’s story about large-scale ML training, a bunch of it is about identifying failing or flaky equipment and fixing it. That’s not so easy in space.
Nonsense, it's right there in the acronym: Space Reliability Engineer (or I guess one could also just leave "Site" as is, since space is for sure a site). That PagerDuty rotation is gonna be hell
For continuous you need to either go for a polar orbit or go very far in space. Most launch centers & providers are not well situation for polar orbits because its not a common use case, so you need to sacrifice launch mass. The same goes for far away orbits - you need to sacrifice launch mass to go further. Also if you are far then you get latency issues.
So it skews the economics pretty harshly. I think OP is right - you need good batteries somehow.
I think the proposal suggested an orbit where the solar panels are always in sun and always properly aligned and always clean due to space gophers.
But more seriously, GPU loads are super spiky. Ground-based power grids and generators and batteries have trouble keeping up with them. You can go from 1MW idle to 50MW full power in 10ms. Unbuffered solar cells are right out.
> "GPU loads are super spiky... You can go from 1MW idle to 50MW full power in 10ms."
That sounds like something that could be addressed in software, if necessary? Cap/throttle the GPUs according to the available power, and ramp power up/down gradually if spikiness is the issue.
Temperature isn't a power source; heat flowing across a temperature gradient can be. But that brings us back to the first problem - how to make it flow.
Unless they've figured out some impressive cooling tech, which I would expect would be worth more than the rest of their company combined, then this is pretty much DoA. "More efficient cooling architecture taking advantage of higher ΔT in space" would indeed be useful if you had a nice medium to radiate into. It turns out that thermal radiation is incredibly poor into the vacuum of space lol.
Space (with a sunshade) is a nearly perfect medium into which to radiate heat, in the sense that there’s nothing better.
But I agree with your general point. At 100°C, you can radiate about 1kW/m^2. That’s 1000m^2 of radiator per MW of datacenter, assuming you can operate with the radiator at 100°C. You can fudge this a bit with a heat pump (to run the radiator hotter, paying a linear-ish power penalty and gaining a fourth-power radiation benefit), but that’s expensive and that power isn’t free.
Here on Earth, you can cool by conduction or evaporation, which isn’t an option in space.
> Space (with a sunshade) is a nearly perfect medium into which to radiate heat, in the sense that there’s nothing better.
There is radiation but zero convection. As anyone with an oven or PC will tell you, even a very tiny fan makes a big difference in the ability to dump heat. We're not putting our PCs into vacuum chambers for a good reason. A small fan in your oven not only makes for more consistent heating in your food, but it requires less power
As an aside, how did the description text ("https://starcloudinc.github.io/wp.pdf") get added to a link post? Whenever I've tried in the past, my text got splatted into a comment. Was this mod intervention?
Yes. I've been experimenting with doing more of that lately.
It's a bit of testing the waters toward what proper URL aggregation might look like. By "URL aggregation" I mean grouping together a bunch of URLs for a story, rather than just one link per submission.
It makes so much sense that there are many comments saying why this is a bad idea. In fact I agree with a lot if not all of these comments. However, I still have to commend how big of a dream this is and that they got some sort funding behind them. I say let big dreams possibly fail although I admit it’s not my money being spent.
Why couldn't you deploy all those solar panels on earth. Then you could supply the grid when you weren't training models, you could recycle your components when they got old, you wouldn't need launches, you could maintain things if they break, and sure, spend a bunch of money to figure out large radiators that could radiate heat out of the atmosphere.
Be the start of a geoengineering project that is actively collecting energy and radiating it out into space while performing useful work along the way. Infrared radiation between 8 and 13 micrometers isn't absorbed by the atmosphere and exits the planet.
So, why not build big infra here and if you can build powerful radiators, aim them up and away from the earth, which needs to be cooled a bit anyways!
When you're on the sun side, everything is too hot and it is hard to cool. You can do direct cooling, such as water cooling, but you have no radiator to dump the heat to...
When you're in the shadow of the sun you have the opposite problem. Things are way too cold. Cold enough normal electronics can fail.
For reference, the ISS can fluctuate between -250F and 250F.
I'm willing to bet that it is easier to deal with the issues of salt water than it is to deal with the heating and cooling issues combined with difficulty to manually access problems presented by space. Price per pound into orbit is still quite expensive...
>Presumably the real purpose of this company is to refute people who said "We'll just walk over to the superintelligence and pull the plug out", without MIRI needing to argue with them
Passive cooling refers to "passive radiative cooling"[0]. This is a well established technique, but I have doubts on how well it will scale with the heat generated by computation.
Radiative cooling works by exploiting the fact that hot objects emit electromagnetic radiation (glow), and hot means everything above absolute zero. The glow carries away energy which cools down the object. One complication is that each glowy object is also going to be absorbing glow from other objects. While the sun, earth, and moon all emit large amounts of glow (again, heat radiation), empty space is around 2.7 Kelvin, which is very cold and has little glow. So the radiative coolers typically need to have line of sight to empty space, which allows them to emit more energy than they absorb.
This is exactly right, and an important fact is that there is a limited bandwidth for heat radiation. So essentially they need to create a giant lightbulb...
> Additionally, deep space is cold, which is accurate in that the "effective" ambient temperature is around -270°C, corresponding to the temperature of the cosmic microwave background.
There's a lot of bad information in their document too. This -270C temperature is ambient space, i.e. deep space. You may experience this when you're in the shadow of Earth or on the dark side of the moon but you're going to switch that negative sign to a positive when you're facing the sun... Which is clearly something they want to do considering that they are talking about solar power. Which means they have to deal with HEATING as well! I don't see any information about this in the document.
> he mass of radiation shielding scales linearly with the container surface area, whereas the compute per container scales with the volume
This is also a weird statement designed to be deceptive. Your radiation shielding is a shell enclosing some volume.
> Therefore the mass of shielding needed per compute unit decreases linearly with container size.
They clearly do not understand the mass volume relationship here. Density (ρ) is mass (m) divided by volume (V).
m = ρV.
Let's simplify and assume we're using a sphere since this is the most efficient, giving V = 4/3r^3. Your shield is going to be approximately constant density since you need to shield from all directions (can optimize by using other things in your system).
m ∝ ρr^3
I'm not sure what here is decreasing nor what is a linear relationship. To adjust this to a shell you just need to consider the thickness so you can do Δr = r_outer - r_inner and that doesn't take away the cubic relationship.
Massive radiators. The ISS has radiators that have a dissipation capacity of about 3m^2/kW. If we use that number, we'd need a 3000m^2 radiator per megawatt, which is the scale they're talking about. This could theoretically be brought down, but not even by an order of magnitude.
I wonder how much cooling the solar panels alone would need, when operating at that scale.
You cool the fluid by flowing it through the radiator. The radiator emits heat radiation into space and cools down the fluid. As long as the fluid is hotter than the equilibrium temperature of the radiator (determined by radiator, space and sun radiation), it will emit more energy than it receives and cool down the fluid.
Followup question, wouldn't nearly any cooling solution that works in space also work on the ground? Radiative cooling is the most basic/common cooling solution on the ground, the main challenge is just figuring out how to to move heat from the component to the radiator, which I don't think is solved by simply putting it in space?
I think other have already corrected you, but radiative cooling is probably the least common on the ground and the only viable option in space.
I can help explain why. On earth, we are surrounded by stuff. Radiative cooling relies on thermal radiation leaving an object. Crucially, it also requires the object to absorb less thermal radiation than it emits. On earth we are surrounded by stuff, including air, that emits thermal radiation. There is a window of wavelengths, called the atmospheric window[0], that will allow parts of the thermal radiation out into space, rather than returned back. Imagine shining a flashlight on tinted glass, the light will get through depending on the color. If the light gets through, it has escaped. If not, the light is returned and heats up your surroundings again.
Also on earth the other methods (conduction, convection, and phase changes) are more effective. The earth can be used as a very big heat sink. On a spaceship or satellite, you don't have the extra mass to store the energy, so radiative is the only option.
> Radiative cooling is the most basic/common cooling solution on the ground
Thats tricky. I know the heat exchange components are called radiators but most of the heat they give off is by convection not radiation. (At least here on the ground.) I heard 80%-20% rule of thumb.
But you are right in the broad strokes. Cooling is not easier in space. Mostly because you have no convective heat transfer.
Oh right, that makes sense. So the argument is that comparing a 50C GPU+radiator in a 20C room vs a 50C GPU+radiator in 0K space, the one in space will dissipate more heat via radiation than the one on the ground? As you say, I'd expect that air cooling is much better than EM radiation, but I guess there is some basis for claiming the possibility that cooling in space is somehow better than on the ground, however unlikely.
Is radiative cooling the most common on Earth? I don't think so. Most terrestial "radiators" actually work with convection, ie moving relatively cold air across hot metal fins, which doesn't work in space.
It’s is on earth as well using solar and batteries. What is likely to get cheaper faster? Solar and batteries? Or lifting datacenters to space? The world is almost at the point of deploying 1TW/year of solar, and batteries are catching up. No space required. There aren't a lot of VC investment opportunities speeding the rate of solar and battery deployments though.
The argument probably is that battery advances require not yet existing tech via new chemistry etc while what they are proposing is basically just integrating tech that already exists
Just spitballing here, but what if you built it on Earth, and then used the savings to build a second one on the opposite side of Earth? Now you have equivalently continuous power via solar array and also, as a bonus, air.
Not an expert in this area, but I think that that "just" is hiding a lot of complexity. Plus you also need some remotely operated robots to mount the replacement.
Stationkeeping is not free, satellite monitoring is not free, and any replacement to any component is now a multi-year, at least 1+ million dollar affair (or most likely a complete replacement, since not many satellites have done in-situ repairs).
Power in needs to equal heat out, and that isn't easy in space. They, deceptively, claim that their novel solution is radiative cooling. Relying on radiation for cooling in space is the problem statement! Convective (as on Earth) is significantly more effective.
I'm not one of those idiots who would claim that "we should focus on terrestrial problems instead of space," but this idea seems to have only downsides.
> “We still don’t appreciate the energy needs of this technology… there’s no way to get there without a breakthrough… we need fusion or we need radically cheaper solar plus storage or something” -Sam Altman
It's kind of depressing that the only way to make this tech better is to feed it more energy. (And apparently now to send it to space)
Can you please not post comments like this? Thoughtful criticism is welcome, of course, but this sort of thing isn't. Besides breaking the site guidelines, it takes threads in less interesting directions and evokes even worse comments from others. We're trying to avoid that here.
"Don't be snarky."
"Please don't post shallow dismissals, especially of other people's work. A good critical comment teaches us something."
"Don't be curmudgeonly. Thoughtful criticism is fine, but please don't be rigidly or generically negative."
Really, on second look, snark still feels justified here. The issue is with TFA. There is little room for a thoughtful comment in response to something transparent.
Some type of submissions will invariably not result in very deep discussion, when the topic itself is so shallow.
We need you (I don't mean you personally, of course, but all commenters here) to follow the site guidelines regardless of how bad an article is or you feel it is.
Someone else being wrong or some other post being bad isn't a reason to make things worse. Doing so just creates a downward spiral, which it's all too easy to fall into.
Tbh your moderation is normally very restrained and even handed so was a bit surprising to see you take down several borderline overly snarky comments in a row (that just so happen to be directed against VC investors or YC founders).
To me those comments seemed over the line, not borderline! I'd post the same replies wherever I saw comments like that, regardless of who or what they're about.
> The frigid vacuum of space should make cooling easier, too, because cooling systems are more efficient when the ambient temperature is lower.
(EDIT: I removed a significant amount of snark here - sorry dang!)
While greater ambient temperature differences do affect radiative cooling, this effect is, in almost all situations, dwarfed by the lack of any kind of conductive or convective cooling due to the aforementioned vacuum. If radiators can be made of indefinite size, there are ways to make this viable, but the construction and maintenance of such an array to handle the wattage of any sizeeable AI datacenter would be far beyond any space project we've done to date.
It sounds like their vision for space-based data centers presupposes nearly-free energy costs, delivered via a colossal solar farm made possible by falling launch costs.
Temporarily putting aside (extremely fair) feasibility questions around those two pre-requisites, data centers are a not-bad choice for things to do with unlimited space energy.
Aluminum smelting or growing food are the two I’d think of otherwise, and neither of those can have inputs/outputs beamed to a global network of high-bandwidth satellites.
Solar energy isn’t that much more efficient in Earth orbit than on Earth - maybe twice as efficient. That sounds nice, but you’re saving half of your solar panel cost while massively increasing every other cost.
The one benefit is being able to be in a synchronous orbit with the sun, so you don’t have to contend with night. However, that’s just another ~doubling of efficiency, which I think still nowhere near makes up for the additional costs.
Microsoft had/has the Natick project which was an undersea data center testbed which allegedly had a bunch of benefits. That doesn't seem to have gone anywhere - or at least isn't really scaling up. I'd imagine the ongoing operational costs of space are worse than the ocean?
To me, the cost estimates seem a bit off and conflate capital with running costs.
The main benefit for space at the moment seems to be sidestepping terrestrial regulations.
> Microsoft had/has the Natick project which was an undersea data center testbed which allegedly had a bunch of benefits. That doesn't seem to have gone anywhere - or at least isn't really scaling up.
I think at the core of this there's a risk analysis. At one point I briefly worked in a team in charge of a company's servers, and there were plenty of stories of things gone wrong enough that someone had to drive or fly to the datacenter. These company's datacenters were named after the closest airport for this reason, iirc. A little optimization in case things went very wrong; you always knew where you'd have to fly in to.
Even if an undersea data center could potentially yield cost benefits, it's also significantly riskier in case something goes wrong. How long would it take to physically access a machine? Do you have to bring down other machines to access it? And at scale, things tend to always go wrong.
To comment on the original post, needless to say this is even more complicated, costly and untimely in space.
It might make more sense to put data centers on the Moon.
It's fairly close, about 1.3 light seconds away. You wouldn't use it for anything realtime, but it would be fine for long AI training jobs.
You could bury the servers underground to shield them from cosmic rays. That would also be good for any people living there.
You could get power from solar panels on peaks near the poles that get light almost all the time. For example, some ridges around Shackleton Crater are sunlit up to ~90% of the time, with short periods of darkness. Use batteries to smooth out the power supply.
For heating and cooling, just use the standard techniques. It's not easy, but it's a solved problem. As a bonus, near the poles, the temperature extremes aren't as bad as at the equator.
You could also sell tickets to tourists. People will pay to see the darndest things.
Even if they somehow figure out all these problems. how do you manage a space based data centre? do you have rotating staff living there? or are they just praying that nothing ever goes wrong??? Isn't radiation a massive problem in space? i would expect consumer grade hardware to be constantly flipping bits accidentally that shouldn't have flipped.
> Their design calls for a cluster of shipping container-style boxes packed with high-speed AI chips. These would be anchored at the centre of a 16 sq km array of solar panels generating up to five gigawatts of power — about 25 per cent more than Drax, Britain’s biggest power station. The mammoth structure would circle the Earth in “sun synchronous” orbit so that it is never in shade
Their whitepaper clearly demonstrates a profound lack of knowledge of thermal engineering. E.g. heat pumps are described as magical things.
They are literally planning to feed the radiators using a coolant like water and sensible heat at 35 degC to 5 degC. At 5 GW, you then need to be pumping 60 000 liters of water per second.
That's like a tenth of the Sacramento river, going through a 16 sq km array in space and hoping that nothing leaks.
How do they plan on addressing the solar radiation issue? What is the solar flare risk?
A CDN for Starlink customers is probably the first use case for servers in space, not training GPT6, which will be a big enough project on familiar territory.
Definitely an out of this world idea. I wonder if their micro datacenter is going to be self-sufficient power wise using only solar energy? And how would they address the hardware failures that are likely when you train large language models at scale?
The white paper back of the envelope calculations show a 4km x 4km solar panels and radiators are required for a 5 GW datacenter. I am not sure how the authors were not cracking up while writing that white paper.
Given the purported cost benefits in their whitepaper [1], hardware failures might be an irrelevant rounding error. They suggest something to the tune of 100x cheaper.
Reading the paper this sounds like space Theranos. If they start producing results then I'd double check to make sure it's not just calculated on regular data centres and that they're just pretending its from their space stations.
Aside from the technical concerns already raised in other comments, I'm also not sure we really want all this private for-profit usage of earth's orbit. The orbital environment is already somewhat congested and people have already been raising concerns about it. There is the potential for it all to spectacularly blow up in our faces and become so polluted that we won't be able to do many launches at all.
"Space Theranos" wins the day for me. There really needs to be a better term than "stupid money" for investors who can be convinced with a slick presentation that by investing in a venture the principals will bring to market products that violate the laws of physics.
Private use of previous public resources has had mixed success, but it feels like leaving space to the public sector will doom us to being Terran bound forever.
the private sector made it to the moon 56 years after the public sector
it's going to take the management of our shared resources and spaces (orbit) for instance to leave earth, and this becomes especially important as Kessler syndrome risk rises with increasing debris in orbit
private companies launching without public oversight and controls are a recipe for cluttering earth's orbit and leaving us earth-bound for far into the future (same if the public sector launches without care but that seems less likely imo)
I don't see the problem in that to be honest... Especially if the other solution would be allowing private companies to take over our (shared) orbit, meteorites, and -- continuing the trend -- planets for themselves to profit. If it seems overly pessimistic, we can just look at how our own planet's ressources are shared...
I don't know what the best solution is to be honest, but a wild west where improbable VC-funded nonsense is flung in to orbit isn't it. Leaving it only to the public sector is the other extreme end of course, and I'm not advocating for that either.
Aside from the obvious cooling issues people have already mentioned, isn't cosmic radiation also very unkind to modern ultra dense silicon? AIUI they tend to use really old silicon processes in space stuff for that reason, and even then they have to build in redundant compute to mitigate logic errors that probably wouldn't happen on Earth.
> They tend to use really old silicon processes in space stuff for that reason.
To be fair that's mostly part "if it works don't change it" and part "that's how we've always done it". SpX uses newer hardware w/ traditional OSs (linux) w/ lots of redundancy.
I was wondering if these server racks in space would need to be specifically designed for enough radiative cooling. Apparently the answer is “yes”: the radiators would be expansive and placed on the reverse side of the solar panels.
Starcloud is developing a lightweight deployable radiator design with a very
large area - by far the largest radiators deployed in space - radiating primarily towards deep space, which has an average temperature of about 2.7 Kelvin or -270°C. The radiators can be positioned in-line with the solar
arrays as shown in Figure 3, with one side exposed to sunlight.
…
Figure 3. A data center in Sun Synchronous Orbit, showing a 4km x 4km deployed solar array and radiators.
wow, one year back, I had made a prediction to a friend that this is the direction that Starlink will head in. I was thinking it would proceed like this:
1. provide internet.
2. provide CDN.
3. Edge Compute.
4. Full-on cloud.
These guys see to be focussing on what is basically offline processing (AI training).
More like, these guys will be focused on parting VCs from their money.
Datacenters in space makes no sense at all. Even ignoring the huge cost of sending hardware there in the first place, cooling is a massive issue in space. No medium to sink heat into means the only way to cool anything is by running water through giant infrared radiators. Not ideal when cooling is the largest bottleneck in scaling datacenters. Note that they would also have to dissipate the large amounts heat their datacenter satellite gets from being exposed to the Sun.
Also disregard the cost it takes to send a technician for maintenance, of updating hardware, etc.
If this and/or the policies, leadership, and mindset that enable these kinds of things interest you, I highly recommend Abundance by Ezra Klein and Derek Thompson. IIRC, they outline data centers in space in his utopian introduction.
I very much agree here. This is clearly nothing more than a scam meant to separate investors from their money. I read the "white paper" and very few of their suppositions actually hold water, as many others have detailed in this thread. It seems like what this kind of "startup" does is pick two or three trending words, put them together and make up a story that would be believable to people who don't dig too deep. These guys chose the words "AI" and "space" and figured out how to stick them together. Speaking of that, anyone want to invest in my new quantum blockchain technology?
Does abundant cooling in space mean there's a better way to radiate away heat than on Earth or just that the heat doesn't contribute to heating up Earth or something more complicated? (Asking because I thought cooling was a big problem in space.)
I would imagine you could launch a new rack, dump the old one, and connect the new one to the existing solar / cooling array. Hopefully with some sort of re-entry and recycling plan for the old one. The sheer size the arrays are going to need to be feel like they are going to be the more important part of it.
This is the best piece of satirical writing that I've read in quite a long time. It's genuinely amazing, every sentence is a delight. And the Musk/Altman/Zuck quotes to start, the stupid little graphs and the "equation" at the end of page 2. I'm in stitches!
TL;DR... cooling in space isn't passive, you're on the "inside" of an enormous vacuum flask. And radiative coupling with space is possible from the ground, if that's what you're interested in:
Very ambitious but it seems futile if you’re not building the rockets yourself. Personally I’m more bullish on figuring out how to use analog chips to train models.
Seriously! There is just so much wrong and some of is trivial.
> radiating primarily towards deep space, which has an average temperature of about 2.7 Kelvin or -270°C.
Are they suggesting putting these things in deep space? I guess for training you can handle hours of delay time but still it is really bandwidth limited. But they say they're using solar, so I assume they ARE NOT operating in deep space but rather near Earth or maybe even on the Moon.
In these locations you have to deal with cooling AND heating. On the moon you swing from -130C (LRO got down to -250C) on the dark side and 121C on the light side. The ISS swings from -160C to 120C. These are too cold for most electronics. Not to mention that these temperature swings create a lot of physical stress on parts, and we're talking about putting up up some of the smallest objects we commercially make? They will rip right off the circuit-board if you don't get it right.
Not to mention that radiating into space is quite difficult. There's a reason we use convection ovens and why we put fans in our computers. It isn't about the temperature of the atmosphere nor the thermal efficiency, it is because convection is just a hell of a lot more efficient. Thermal radiation is like shedding your heat via a lightbulb.
Their claim here is that they can radiate 633W/m2. For supercomputers we're talking on the order of 10s of MW of waste heat. That's 10^7! These are going to be BY FAR the largest radiators in space and going to cost tons of money for the mass alone.
Not to mention the size of the solar panels they'll need... But at least they mention this one: "A 5 GW data center would require a solar array with dimensions of approximately 4 km by 4 km," These are GIGANTIC structures and far larger than anything we've put into space.
> The mass of radiation shielding scales linearly with the container surface area, whereas the compute per container scales with the volume. Therefore the mass of shielding needed per compute unit decreases linearly with container size.
This one really got me, because it can be sniffed out with high school physics.
Density (ρ) is mass (m) divided by volume (V): m = ρV. We'll assume a sphere due to its efficient surface area. You use Δr as the shell's thickness: V = 4/3(Δr)^3
Let: m = ρV
Let: V = 4/3(Δr)^3
∴ m ∝ ρ(Δr)^3
What is linear? What is decreasing?
> This effect, combined with the shielding afforded by the cooling blocks, means that radiation shielding is proportionally a much smaller concern compared to electronics on typical satellites today.
Now this might be partially accurate, but it does require some very specific conditions to be true. It is quite common for spacecraft to dual purpose their cooling systems to also act as part of their radiation shielding since essentially the most important part of shielding is mass[0]. But also most spacecraft aren't giant computers in space. You're going to need extremely uniform shielding and I doubt you can efficiently design the cooling system to also be uniform.
But also you have to remember that you can't shield your solar panels. To do so would prevent light from reaching them. That leads to a weird constraint here and I would not expect these machines to be meaningfully long lived. The alternative is you could go repair them, but that's expensive too.
-------
I think the idea is cool and worth exploring, but given the white paper I'm not sure why anyone gave them money. The idea itself is old and there has been a lot of work done in this space (pun intended). It just seems like it is riding the hype of space and AI. Exciting things, but that can make people naive. Maybe there's more than is shown in this whitepaper and I hope investors are doing more due diligence but there's definitely a lot of red flags here.
[0] I know this because I've research for NASA on radiation shields. I got multiple SBIR and STTR grants for this work. Material choices still do matter but the right material is proportional to the radiation level. But the higher the energy level, the less atomic properties matter and the more density does. You can get benefits from the electromagnetic properties of protons and electrons (beta-), but these don't help you with neutrons. That is, until after you slow these things down, which is why there is typically layering.
Is this an elaborate joke? I believe so given all the very real physical problems with this, but “GPT-6” training was what pushed it undoubtedly over the edge to pure joke.
This has to be one of the dumbest ideas I've seen posted here.
Just think about the sheer effort required to dump 1 BILLION watts of waste heat into space - the engineering challenges alone make this completely impractical.
Compared to this, Theranos actually looks like a solid investment. At least Holmes had working demos and big-name backers before it all fell apart. This doesn't even pass the basic smell test.
what the actual fuck? my boys joseph and ludwig would like to have a word with y'all
in ideal conditions, your gpu putting out 600W will need about a square meter facing deep space to keep it at 80c, this idea is absurd on first principles alone, maybe if you have heat pumps you can push this but then you're dealing with on orbit fluid loops that you can't maintain, as i said, what the fuck?
It feels defeating to see so much money being thrown at what is either an obvious grift or someone hoarding world changing technology. There’s a reason data centers aren’t in space… If someone has the technology to make it feasible, they are doing the world a massive disservice by not sharing it.
I think what I like the most about this whole thing is that the term "cloud computing" (which I always thought idiotic) will now finally become somewhat more meaningful.
I doubt you need that much bandwidth or reliability. Training is “on site”, you just need to upload training material once, then download the trained model.
is this a joke? isn't cooling in space a really big problem? how does it make sense to run a data centre with huge cooling requirements in a place where cooling is very difficult to accomplish?
The grift never ends. Ignore that doing anything in space is 10000x harder than on the ground, or that datacenters require constant maintenance, or that cooling in space is probably the largest issue the ISS faces.
Our world has gone mad. The real economy is hollowed out little by little by grifters, while billions are awarded to clown projects like this by clueless VCs that have way too much power relative to their stupidity. TSLA is up 5% just today, despite massive losses and $800M in unsold cybertrucks.
Actually contributing to society doesn't matter anymore, at all. The markets are acting beyond irrationally, and the consequences will be dire.
I believe they'd mainly need to be strategic about where and how deep they were radiating into the surface since there wouldn't be atmosphere that would trap heat.
Please don't litter HN with LLM generated slop. Each and every one of us here is more than capable of doing that themselves if they please. The value of HN is the human discussion.
Lol you mind giving me a quick heads up on what have you contributed other than this meta gripe? Any concrete response to the actual questions I'm asking or just a soapbox?
I thought your question was interesting. The ChatGPT quote was not. If I want to know what an LLM “thinks”, I can ask it. I have more respect for and am more interested in reading people’s thoughts than machine output.
I do appreciate your at least labeling at such, so I could skip over it.
If you're under the impression that all the content on here is 100% human, that's probably slightly optimistic, and the reason why I labeled it explicitly rather than regurgitating / rewording. That said, I'd rather read GPT output than skim the meta-discussion that ruins so many HN threads. Totally stoked that we're now contributing to that rather than talking about the actual content.
> That said, I'd rather read GPT output than skim the meta-discussion that ruins so many HN threads. Totally stoked that we're now contributing to that rather than talking about the actual content.
You brought this meta-discussion on yourself and upon the rest of the community, knowing the likely result beforehand. It’s hard to say you’re not part of the problem you rail against when you create the initial start conditions for the thread.
> Honestly was not aware of that guideline. Insinuating that I was someone trolling for this discussion is a stretch.
I’m not insinuating anything, quite the opposite. You have experience on the site, and so are aware of how often threads can devolve into and revolve around meta-discussions, and one of those common meta-discussions on HN is in response to generated comments and AI slop. I’m not saying you’re trolling by doing it intentionally, and in fact I’ve appealed to dang to add the provision to the guidelines proper myself in email and on HN. But surely we both agree that to continue to do so now that you know would be a flagrant violation of the guidelines?
> That said, I'd rather read GPT output than skim the meta-discussion that ruins so many HN threads.
I try to make the kinds of posts I want to read. Perhaps you might find this guiding precept helpful also, as I do. As generated comments are against the guidelines, your own words will suffice.
At first I thought the poles (of the planet) might be good. The cooling is basically free. But the energy and internet connectivity would be a problem. At the poles you can really only get solar about three months a year, and even then you need a lot of panels. Most of Antarctica is powered diesel because of this.
So the next thought was space. At the time, launching to space was way too costly for it to ever make sense. But now, with much cheaper launches, space is accessible.
Power seems easily solved. You can get lots of free energy from the sun with some modest panels. But to do that requires an odd orbit where you wouldn't be over the same spot on earth, which could make internet access difficult. Or you can go geostationary over a powerful ground station, but then you'd need some really big batteries for all the time you aren't in the sun.
But cooling is a huge problem. Space is cold, but there is no medium to transfer the heat away from the hot objects. I think this will be the biggest sticking point, unless they came up with an innovative solution.
reply