a rant about “data centers in space”, with a particular focus on the moon

the tech press got off way too easy with cryptocurrency as a thing it can call bullshit on. as we’ve seen over the past couple years, it can only sometimes, just barely, manage to put together a sentence that is anything other than “wildly, unreasonably optimistic” about AI. and it’s completely unable to make any cogent points about other types of blatant hogwash like this incredibly insipid “lunar data center” I just found out about today, instead repeating bullet points from the company’s press release verbatim.

I have to go off about this, or else I will explode.

communication

the moon is 240,000 miles away from earth. this is a pretty uncontroversial fact, and it’s not slated to change any time soon. this means that your round-trip latencies to earth are, at minimum, about 2500ms; we’ve all seen TV broadcasts where the voice of someone reporting via satellite is delayed by a noticeable margin, and that’s just the delay to earth orbit and back.

this data is in a very inconvenient place, computer-wise, for doing anything with from the surface of the earth, and laws of physics mean it will always be very inconvenient unless someone invents an ansible. this means that any use for these data centers other than offline backup is going to require the computer processing that data to live in the same spaceship as it, and that means having to worry about cooling and power.

cooling

techspot’s article brags about the “naturally cooled solid-state drives¹” as if they’re some huge win for sustainability. the depth of thought going into this seems to be “space is really cold right? it must be good for cooling.” but the temperature of the heat sink is not the limiting factor for a computer system to reject heat. a microprocessor runs approximately as well (within a factor of maybe 2) at a temperature of 100K as it does at 300K, and — as demonstrated in that parenthetical! — you can get earth-based computers down to a pretty cold temperature with pretty barbarian means like a thermos full of liquid nitrogen, so there’s not much value in that².

in fact, sending a computer to space is the absolute worst thing you can do to cool it effectively. because there’s no air in space, a system can ultimately only reject heat to space through radiative cooling — shooting it away as photons through the process of black-body radiation — and not through conductive or convective heat transfer.³ (less obviously, you also have to be very careful about where you put space-based radiators; they have to be in shade all the time, or else the heat rejection of the radiator becomes negative because the sun can out-radiate you any day of the week. this means that a spacecraft with a certain surface area of radiator has to hide it behind an equally large, highly reflective, sun shade.)

the ISS can reject 70 kilowatts of heat with nearly 40 square meters of radiator surface area. a single standard-height datacenter rack can now generate as much heat, or more, as the entire ISS — the largest space-based structure ever built — can reject to space.⁴

for comparison, nvidia released a two-slot consumer graphics card yesterday that can reject almost 600 watts of heat. to the air. because it’s in an environment that’s full of air.

power

the reuters article quoted by techspot cites that there is “abundant solar power” available in space. fine. we’ll take that as granted without hand-wringing too much about power density or conversion efficiency. you can get about 1.3kW/m^2 of incoming power on solar panels at the top of the atmosphere, maybe 400 watts of which turns into electricity, and much of the other 900 watts turns directly into heat which now means you have to carry even more radiators up there — there’s a reason why the ISS has a separate active cooling system for the waste heat from its solar panels. so for a modest 10kW rack, you should bring about 25 square meters of solar panels up with you. (assuming none of them are ever damaged, which they will be, and they won’t suffer any degradation over their deployed life, which they will. you should design in some safety factor.)

the moon, however, is a terrible place to exploit solar power: it has a decent amount of gravity so you can’t build the panels as lightly as you can in orbit, it has just enough of an atmosphere that it’s covered in dust which blows around in storms, and just like every other planetoid (including earth!), every point on it spends half the time in the dark, which means you now need to overprovision your solar panels by 100% and carry enough battery backup to run your data center for 2 weeks at a time⁵.

from a power perspective, you would, ironically, be much better off — about 100% better off, to put a number on it — parking a data center in a polar orbit around the earth or moon, where there’s far less dust (but it’s spicier) and the panels will only be blocked during eclipses.

(un)natural disasters

one of the two most credible, which is to say least incredible, advantages cited for putting a data center in space is that there are no natural disasters in space.

this is blatantly false. everyone knows at this point, from movies like “Gravity”, that space is full of tiny particles of rock each with kinetic energy comparable to a bullet. the moon has seismicity. there is about 100 times as much ionizing radiation on the moon as on earth, which means 100 times as much radiation hardening to achieve the same bit-error rate from cosmic rays.

reliable access to stationary data centers in space from earth is also entirely dependent upon a network of relay satellites with line of sight to one another and line of sight to the customer; the moon isn’t visible from any given point on earth half the time (source: go outside and look at the moon for a while.) the kessler syndrome could start tomorrow (it could’ve started already), eventually dropping your data center’s effective uptime guarantee to one five. and with increased use of other extra-earth celestial bodies could come their own personal versions of the kessler syndrome.

governments

the other most credible advantage cited is that there are no governments in space.

this is also blatantly false. the stock brain genius retort to the complaints about power and cooling above would be that, once we colonize the moon, we’ll strip-mine helium-3 out of the lunar regolith and feed it into our nuclear fusion plants; once we colonize mars, we’ll melt the ice caps and pump all the waste heat from our martian data centers into the lakes. none of these things — indeed, no meaningful resource extraction or wealth generation at all in space — will happen without one of two things happening first:

1. the surviving belligerent of the cold war, the united states of america, takes back the promise it made to the world: that it would not extend the frontiers of the cold war into space, and leave space part of the common heritage of mankind. the american government is itching for this, and has been openly talking about extending american rule to space for like the past 15 years, so nobody gets to pretend like this would be a surprise.
   
   end result: instead of an outer space without governments, we would have an outer space subject entirely to the US, china, and russia, maybe with slices held by japan, india, and the EU.
   
2. corporations gain extraterritoriality for their space ventures, and space becomes an environment where we replace the fraught human process of doing democratic politics with an iron-fisted corporate fascism probably enforced by private militaries. (you can’t violate the treaties against space weapons if you’re a non-state actor!) a broad swath of modern sci-fi media envisions a spacefaring future where space is under these conditions, thankfully mostly unaffectionately, so nobody gets to pretend that this would be a surprise either.
   
   end result: again, we would end up with an outer space subject to governments, but they would be weird microstates run by three guys who have the ability to move freely and a thousand of their reports who could be shipped back down to earth or kicked out an airlock if they mouthed off. finally, a world where the only thing keeping information from being available is if it personally embarrasses [checks my notes about who runs major space companies] elon musk.

and to drive this point home, from a governance perspective, data and compute in space mean absolutely nothing while there are no people living up there; governance is a relationship between human beings, not private property. if someone violates local law, there is nothing to stop an earth government from blocking access to their data at the relay satellite, or prosecuting them for transferring the data between earth-bound computers after it gets downlinked.

the most damning statement in this whole article is buried at the bottom, to allay people’s fears that our important lunar backups of the Imagine Dragons discography will be lost forever:

Thankfully, the data center will have a ground-based backup at a Flexential facility in Tampa.

so, what you’re actually saying is that the data will be subject to the laws of the united states of america.

conclusion

please. for the love of god. ask someone who has ever thought about space to report on space. stop assuming that just because something involves rockets, it will be the next big productive industry. these guys’ press release page is full of terrible, obvious AI art. their launch clients are a state which is more than happy to throw a few bucks to something which is propaganda for the space industry, a country which would love to have a cold wallet for the bitcoin they save from an economy composed largely of offshore gambling and tax avoidance, and Imagine Goddamn Dragons.

but then again, who am I to question them? they “have a firm belief in IDIC and a passion for GYSHIDO.”

1. back of the envelope, as of this writing, you can buy about 30-40 terabytes of earth-bound solid state drives at retail for the price of launching 1kg into low-earth orbit on a Falcon 9 ($6,000), a process which burns about 5 kg of fossil fuel (395,700 kg fuel mass on the falcon 9 / 22,800 kg of throw weight to LEO / (1 + 2.327) for the fuel mixture). on earth, you can cool 30TB worth of solid state drives with probably 50 bucks worth of industrial-grade fans and enough electricity to run one LED light bulb. but whatever. ↩︎
2. there is value in the low temperature of space for certain other things, though! for example, the optics of space telescopes are kept very cold because it reduces the noise on the camera sensor — but practical space systems are designed with a lot of attention paid to thermal management, keeping the heat-generating parts of the spaceship away from the parts that need to be cold, because of this. ↩︎
3. you can also theoretically cool it with a once-through fluid-cooling system, heating up a working fluid and then shooting it overboard, but now you’ve replaced the difficulty and expense of getting a ton of spacecraft into space with the problem of getting a ton of water — which is to say, 250 gallons, an amount of water that seems like a lot when you say it but would last a single once-through desktop computer water cooling system about an hour — into space, and then throwing it away almost immediately. ↩︎
4. you could maybe improve the heat rejection density of radiative cooling by using heat pumps to heat the working fluid being cooled by the radiators above the working temperature of the computers you’re cooling them with, then keeping the radiators on the other side of a heat break from the data center. I’m not sure how much this gets you, in the end. I’m not a thermodynamicist. ↩︎
5. there are other forms of space-borne power available which aren’t subject to the vagaries of that mysterious and poorly characterized phenomenon of “having to have line of sight to the sun for solar panels to work”, like radioisotope thermoelectric generators. not only has nobody ever built a 10kW RTG before, but it would mass about two metric tons if you did (making a very generous extrapolation), compared to the 50 kilogram mass of 25 square meters of solar panel. ↩︎