Orbital datacenters subject to launch stress, nasty space weather, and expensive house calls William Gibson's Neuromancer holds up well after 40 years. One of the cyberpunk novel's concepts was an AI housed in an orbital datacenter (ODC) above the Earth. Today, startup companies and venture capital firms are hoping to turn orbital datacenters into reality to enable AI, believing that free power from the sun and cooling …
Many remote sensing satellites with IR sensors cool them down to quite low temperatures by having a radiator pointing out to space, which is about 3K so quite a delta for radiation cooling, but not as easy as Earth-style conduction/convection. Also you need a bit of effort to keep it pointing at cold space, not hot Sun (or warm Earth), and for stuff buried deep in the middle of the structure the heat has to be pumped to the outside-facing radiator, typically using a Stirling engine or similar.
Yep was going to say the same thing. The only question I have is are the people getting VC funding for this that ignorant or that crooked?
That they get a single dollar for this is further proof (as if any was needed) that extremely wealthy people did not get that way because they are smart. They were lucky, either by birth or circumstance. Because anyone putting even a single dollar into this venture deserves to have it conned away for being so Dunning-Kruger that they won't even seek the advice of someone with a clue who could tell that a datacenter in space is an impossibility.
This is the same lack of basic understanding of the physical characteristics of space that Musk displays with his 'genius' brain when it comes to terraforming Mars. Mars doesn't have a significant atmosphere not only because it is not very massive but also because it doesn't have much of magnetosphere to protect it from the solar winds. The atmosphere it did have was blown away and then there is the problem of radiation.
I guess like all rich assholes, he has no grasp on scarcity. He plans for humanity to continually dump resources into generating an atmosphere, like dumping fuel from a jerrycan into your rusted out fuel tank so you can drive a janky truck a little bit further. But he probably believes he could restart the core of Mars or something absurd like that.
Only viable on a dark side of a planet/moon, or hidden behind a ‘Sunshine movie’ heat shield. Otherwise you are in full glare of the Sun 24x7 and self-evidently heat conduction isn’t a thing.
Personally I’m waiting for Space X to propose a Foundation style Space Bridge/Space Elevator.
No, you're not wrong. Space is misunderstood, people think an unprotected person would freeze in space when they would actually over heat, the body cannot radiate faster than it generates heat.
To cool an orbital box of processors you'd need to concentrate the heat on the radiators and they'd need to be big and completely out of any direct or reflected sunlight. Possible but tricky.
One of the simple vacuum sensors is just a heater and a temperature sensor. The systems puts a constant power in to the heater and you measure the temperature, which increases as the pressure drops, because the heater cannot be cooled by convection, just radiation.
You'd need some type of coolant loop where the coolant is protected against freezing but reaches the highest possible maximum temperature, which can be transferred to the radiator. If we assume there's some sort of 'ideal' coolant that doesn't limit your design then you are limited by the maximum temperature your heat pump/compressor can achieve and the maximum temperature your radiator material can obtain without thermal expansion becoming an issue.
If you could get your radiator hot enough to glow a faint red (i.e. about 800K) then you could shed about 15kw/m^2. If you could somehow get it to 2000K you can bump that up another order of magnitude.
So I suppose if you can solve the problem of the coolant to use, the heat pump able to concentrate heat that much, transferring that heat to the radiator, and selecting a material for your radiator that gets that hot without issue, and build this all reliably enough that it will continue operating for years since no repair will be possible then you have only one problem left.
How to power this. Because if you create something able to glow white hot at 2000K and radiate away 150kw/m^2 into space then unless your solar panels are measured in tenths of a square kilometer all that clever engineering to maximize your radiator's capacity will have been wasted. You're going to have a launch a nuclear reactor into orbit, which I'm sure no one would have any objections to, and it too has to operate without maintenance - and will add to your cooling requirements.
> Esp.. When someone is ‘spaced’ as punishment.
Unless there's a passing spaceship running the Infinite Improbability Drive, of course.
Now, what was the telephone number of tht Islington flat where I once went to a very good part. There was this very nice woman. . .
Latency ...
Indeed. If you put your data center in Low Earth orbit the latency isn't too bad. But the satellite is only visible from a given location for a few minutes every now and then. If you put the data center in geostationary orbit, it's always visible from the same roughly 40% of the Earth's surface. But the round trip latency is around 600ms.
Not to mention obsolescence.
Given that the generational lifespan of the average bit of data center kit would make a mayfly feel like an old codger by comparison, who's going to use that floating pile of "legacy hardware" in a year or two?
We should double check the imaging from the Cassini Orbiter. Maybe that's how Saturn got its rings.
I don't think I've ever seen an article category more appropriate than this - off prem, indeed!
Can't see this happening, though. Cooling and maintenance aside, what do you do when it's time to upgrade or replace your fancy orbital servers? Simply deorbiting them would seem like a terrible waste.
I wonder if it would be cheaper in the long run to have the data centre in a crewed station? There would be people around to do maintenance, at least, but it still wouldn't be cheap.
Apart from the complete daftness of the idea of orbital data centres (especially in LEO) one "obvious"† way to overcome some of the drawbacks would to build the centres from small relatively cheap modules with the capability of assembling themselves into complete systems - and presumably disassemble themselves for upgrades or repairs.
Without too much thought I imagine CPU and GPU chips that dissipate >100W as heat would necessarily require convective cooling either by immersion in a fluid (gas or liquid) or some heat pipe technology (unless the chips were to glow incandescently white hot.:)
Handling and transporting fluids in an obligatory vacuum reliably for long periods would be a major engineering challenge in itself, I should think, without the thermal aspects of effective heat exchange to consider.
Apparently no matter how ridiculous the idea, if it is pitched as enabling the wider deployment of AI, the pimp will be showered with gold; clearly most wouldn't even deserve a golden shower.
† admittedly for very dubious value of obvious.
"Apart from the complete daftness of the idea of orbital data centres..."
Just the sheer mass of stuff to put up there is bonkers. Starcloud's own white paper (linked to in the article) proposes a 5GW solar array using"thin film cells" that they claim have >1kW/kg performance. At that performance, that's 5000 tonnes of thin film solar cell to launch (for reference, the whole ISS is about 400 tonnes). Internet says Falcon Heavy currently costs about 1,500 USD/kg at 64 tonnes max payload to LEO. So that's 7.5bn USD and 78 launches. That's falcon heavy, if you want to use falcon 9, the cost goes up by almost double and the number of launches by a factor of 3.
Starcloud are actually proposing LEO for this. 7.5bn USD doesn't sound a prohibitively large cost for some super gullible investors. But 78 falcon heavy launches? Assuming a fairly ambitious rate of 1 launch every month, that's over 6 years of non stop, trouble free launches.
That's just to get your solar panels into space (low earth orbit, if you want them in GEO instead that's about 17bn USD and 187 launches!). For that, you haven't got anything to do with them yet. You haven't unfurled them, connected them together, found a way to keep that massive structure pointed the right way, or even unpacked them from the box yet. And when if you do all that, you still haven't put anything else up there yet to plug them into.
Starcloud are claiming that launch costs could come down to 10USD/kg and be done at a rate of 3 launches per day. Ludicrous. Their reference for the 10USD/kg value is a five year old tweet by none other than Musk himself.
I have this great investment opportunity for an orbital AI blockchain, leveraging cutting-edge quantum computer and encryption paradigms to deliver the highest levels of security and reliability. We already have the UK's CRESS programme signed up for its combined navigation, Internet, environmental monitoring and emergency broadcast (financed by advertorial) constellation, to be launched by MUSTARD II, with operational deployment scheduled for ten to twenty years' time. Negotiations with India, China, Japan and Donald Trump's freshest dump are progressing well.
This is the future. Get in now, get in big. Don't miss this latest chance of a lifetime! [Click here] to receive the full information pack at cost price (ahem).
"Radiation from the sun and random cosmic rays play havoc with satellite chips unshielded by Earth's atmosphere, resulting in faults and errors."
Well for an "AI", faults & errors are par for the course, so (unlike the cooling issue) radiation induced errors won't be much of an issue as cannot make the output quality appreciably worse than it already is.
I could see a couple of views. One, the AI people could blame any problems on radiation, and that is why the answer was so wonky. On the other hand, if the radiation hit caused a "good" "creative" answer, they'd extol the wonders of AI. It is almost too perfect. I actually think though radiation is the problem. Unless things get incredibly inexpensive to lob into space, like pennies per pound. Lead is heavy.
Anyone know how the HP mentioned in the article is shielded? I'm thinking the storage has to be an SSD, no way spinning rust is going to make that trip. But an SSD seems like it would be incredibly susceptible to radiation hits moving that buried charge around. RAM is almost certainly ECC. Or did they put the whole thing in a chunk of lead?
ODC only makes sense if you are also in orbit to use the data and supply the inputs there. Otherwise Microsoft's under water DC was a much better idea. Much cheaper house calls and you can reject as much heat as you can generate (nuclear ships/subs reject 1000's of MW across the various fleets) and we are good at making those kinds of heat exchangers really cheap and reliable.
The real issue with ODCs isn't all the space weather, micro meteoroids, space junk, and cyber threats.... Or what happens when N.K. pops a nuke in LEO for testing... It is power generation and how to get rid of all that heat! Space is a terrible heat sink and power sources are heavy and dangerous.
... before drag brings it down to Earth? Shoudn't be it much higher to avoid the need of orbit being restored frequently? Maybe they can move it upwards using it as a solar sail with the proper orientation - but woudn't the Sun sometimes push it also in the wrong direction?
In orbit, no, but as long as you don't want it interactive it's more feasible on the Moon. Provided you find a way to land things softly you can ship things bit by bit (pardon the pun), you have a hot/cold zone with a degree of persistense, you don't have high velocity objects whizzing around (although there is very abrasive grit there) and there is even scope to build some kind of maintenance facility. And you can plant a reactor there to power it.
Still wouldn't want the service contract, though.
Folks,
You know that in low earth orbit the 'day' and 'night' are each 45 minutes long? And in higher orbits you'd have less 'night' but more 'latency', and be a lot harder to get to?
So answer me this: Does it suspend processing when the sun is not visible. or does it use batteries? Batteries that go through a full charge/discharge cycle every 90 minutes...
Oh - and has anyone thought how reliable stuff will be if any loose electricly conductive fragment can float around near the boards?
They've had experience of satellites running in LEO for a few years, power and electrical safety have been pretty well sorted out for decades.
The bigger problem is secondary latency - For about 70% of the orbit the satellite will have the bulk of the Earth between it and a ground station. They'd either need multiple ground stations or multiple satellites to relay data to and from whichever one is over the ground station.
Although the primary latency would be greater they could do away with the secondary latency by sticking them a lot further out in geostationary orbit.
Doesn't actually matter as it'll never get off the ground. As has already been pointed out, the oceans provide better cooling, perhaps building datacentres on the bases of wind turbines would be a more (well sightly more) viable proposition.
Or maybe wave energy, Salter's duck is not all that good for powering the grid but for a single datacentre a few may do the job.
The Register 
Biting the hand that feeds IT
Copyright. All rights reserved © 1998–2025
