The thing that people miss in this is that the feature they’re seeking by putting servers in space is only to have servers outside of any jurisdiction, with the advantages that it might bring
Now that is actually smart
Wouldn’t it be cheaper to put it underground?
In either case the installation cost and infrastructure costs are excessive and the I/o is probably limited
Ridiculous, you can’t have cloud computing in space, there’s no atmosphere!
Considering the ludicrous price to put each pound of equipment into orbit, I’d like to invite them to send as much hardware as they can in to (high) geostationary orbit so they can find out how well a vacuum does NOT promote radiating heat
Edit: also forgot about solar radiation flipping bits. I love the idea of them having to reboot the machine (if they even can) remotely once ever 15 minutes
I don’t think the point is to really build datacenters in space. The point is to convince investors that it can be done in a profitable manner so some people can create a fake businesses out of it and siphon money off the system. Much like the same as trying to convince investors that LLM + more money = AGI
I also wonder if this is an entire red herring. There are increasing reasons for more compute in space, such as to pre-filter sensor data.
Is it to naive/optimistic to think no one is actually looking for a space datacenter to compute terrestrial loads, but they recognize the need for processing space loads?
See now you all are thinking.
The rich wouldn’t tell us this shit if it wasn’t going to be used as some spin/distraction whatever it is.
cause of ping?
Mr Musk has to justify that 1.75t valuation somehow
I love how his rationale is that manufacturers of natural gas generator parts are backordered o 2030, so instead of… I don’t know, spinning up more natural gas hardware or terrestial power generation, the easiest solution is to go from 11 attempts/0 successful launches of a space platform to tens of thousands of launches a year carrying unprecedented mass of bullshit into orbit…
Don’t data-centers require massive cooling?
Yes, and it’s easier to cool things on earth. In space, there’s no air to help you cool thinks off, you can only reject heat through radiation. Most spacecraft are carefully designed to reflect heat/light on surfaces facing the sun and radiate heat into empty space from surfaces that are shaded.
In space there’s no epa
What if you build it on an asteroid or moon or planet. Uranus is ~-225⁰C, right?
Yes I’d like to build data centres on Uranus one of the most distant planets in our solar system, and also one without a solid surface but who’s counting.
Yes, but the two-and-a-half hour lag each way would be a killer.
My understanding is that these “datacenters” would be used exclusively for model training, where latency doesn’t matter.
It is still an outrageously stupid idea for a zillion other engineering reasons, though.
It would need to have an atmosphere, so asteroids and most (all? Idk not an astronomer) moons are out.
Mars might be feasible at some point in the far future, but there’s still the lag problem of 3-20 minutes depending on time of year, so not very useful for anything user facing.
most moons
Pretty much every moon but Titan. Titan, however, would be excellent for heat dissipation. Long before generative AI was even a thing, scientists have speculated that Titan would be the perfect place for datacenters because low-temperature computation is so much more efficient.
Of course, building a datacenter on Titan would be a several-hundred-trillion dollar endeavor, so… good luck bootstrapping your way into that industry.
None of the moons in our solar system have atmospheres. Earths moon is too small to hold on to an atmosphere, and the Galilean moons of Jupiter are too cold for an atmosphere, the gases just freeze.
The best place would be either a space station in low earth orbit or of the L4 or L5 point. The data issue would be the problem though I suppose you could just use the data centres for training but not for active processing but then you would need to build data centres on earth for that.
Given that you’re going to build the earth data centres anyway you might as well do all of the processing on earth at the same time.
Naive question, but would bit-flip also be a problem without the atmosphere to shield (some) radiation?
Yes but also no. Bit flips will happen unless you have rad-hardened computers but apparently, bit-flips are not really too problematic for LLM training. I guess when correct answers are optional, correct buts are as well.
I can’t tell if “correct buts” is just a genius detail in this comment… Or a genius happy little bitflip accident.
Thats not a naive question at all. You’re totally right. The term to learn about this is “rad-hardened computing”. It’s a solved problem, but the solution involves a buttload of redundancy and extra silicon with huge performance reductions compared to non-hardened tech.
It’s less of an issue if you’re in the shadow of the sun but still quite a big issue.
So they would need to swallow up even more of our chip fab production and push ram and SSD prices even further through the roof for checks notes ah yes… the same functionality as they have on earth.
AI is already unprofitable because of the insane hardware requirements and the fact that no company has a “moat” so there is a race to the bottom pricing-wise… I can’t imagine anyone also then accounting for building space-hardened kit and getting it into space and dealing with shortened lifespan of the kit is ever gonna see a return.
All this just so that a chatbot can confidently tell people the wrong stuff
talks about this, in conclusion yes
There’s another problem that nobody mentions. Putting thousands of additional satellites into space would seriously increase the risk of Kessler Syndrome occurring.
Little bit of a nitpick but Kessler syndrome doesn’t care about how many satellites you have, and more about how many dead satellites you have hanging around on random orbits. You could put hundreds of millions of satellites in space as long as you had some sort of decommissioned program. You can always send up rockets if you can just move the satellites out of the way / know where they are.
Dead satellites do add a much larger risk than satellites that can be steered, sure. If we stopped steering all our satellites right now, I believe it’d only take a few days before a collision occurred.
However, every satellite in orbit adds to the risk, especially if a chain reaction starts happening and it becomes very hard to avoid the shrapnel flying around. Or if a once-in-a-century-type solar flare takes out a bunch of satellites.
Edit: Basically, the best way to prevent Kessler Syndrome from occurring, is to keep the number of satellites in orbit below the threshold where it could occur.
In the future cleaning up our LEO zone will be a problem like cleaning up the rubbish in the oceans.
At this point I feel we’d just be immunising the rest of the universe from human stupidity.
This isn’t true for low orbit items. They will come down on their own in ~5 years.
At the absolute worst case scenario, we’d be blocked or ~5 years. Maybe 10 years if they put it a little higher.
Collisions in LEO can chuck debris into orbits which intersect higher orbits. If one of those collides with something in in said higher orbits, you have a problem.
Any orbit resulting from a collision will pass through that collision point unless there’s another collision to change it’s velocity again. The higher a collision sends an object, the more likely the “orbit” intersects with more atmosphere to cause drag, or it might even collide with the ground without drag.
I sincerely doubt that a collision in low earth orbit is going to result in debris being flicked up into geostationary orbits, the energy differences involved are just monumental.
It’s possible it could go to a higher orbit, but we don’t have mega constellations in those orbits. I don’t know enough to know how far something could get flung up either, but I suspect if you’re in a 5y orbit, you aren’t reaching a 50y orbit area, and probably not even a 10y orbit area.
The idea of putting data centers in low Earth orbit sounds cool at first. It feels futuristic. It feels like something that should be efficient. It is not.
Yes, space is cold. Yes, you get a lot of solar power. Those are the two points everyone repeats. What they leave out is basic physics and cost.
Cooling in space is not free. There is no convection. Heat only leaves through radiation. That means giant radiator panels. AI racks throw off massive heat loads. The more compute you add, the more radiator surface area you need. That adds mass. Mass costs money to launch.
Even with companies like SpaceX driving launch prices down, it is still extremely expensive per kilogram. And servers are not permanent infrastructure. They get replaced every three to five years. You cannot economically upgrade racks in orbit the way you do in a building on Earth.
Then you have radiation. Either you harden the electronics, which makes them slower and more expensive, or you accept higher failure rates and build in heavy redundancy. Maintenance becomes a logistical nightmare. A failed power supply on Earth is a service call. In orbit it is a robotics problem.
Meanwhile hyperscalers like Amazon Web Services, Microsoft, and Google put data centers next to cheap power, fiber backbones, and cold climates. It is boring. It is practical. It works. Orbital data centers only make sense if we already have large scale industry in space. We do not.
And what really makes these threads irritating is the obvious rage bait framing. Throw up a clickbait title about AI destroying the planet or Big Tech trying to escape Earth and you attract people who already hate AI. The discussion stops being about engineering and economics and turns into ideological noise.
If someone wants to seriously debate energy efficiency or scaling limits, fine. But pretending near Earth orbit is some obvious solution is not serious analysis. It is a cool sci fi concept. It is not a rational infrastructure strategy.
To add to your point about logistical nightmare, Microsoft tried an underwater datacenter. Even right there, just a little bit underwater was absolutely not worth it.
Really? I would have figured the Rapture route would be workable with the right engineering. Especially given the massive amounts of borderline free cooling and non-existing regulatory environment if outside territorial waters.
Possible, but just not worth it. In their case it was barely underwater in some shallows. Go full Rapture without ADAM and it’s just untenable.
The whole point is that it is cool so that it can be hyped up like AGI and then sold.
You mean more socialism for the already rich.
It is consistently amazing to me how many people use the word socialism on this platform and have no idea what it means.
It’s a phrase often used by Martin Luther King Jr
Yes. It is indeed a phrase often used.
Maybe for a space based population a data center in space would work. This is just taking off site hosting too far.
“beyond the cloud” 🤮
Introducing: Microsoft Cosmos!
Send your data to heaven while we turn the planet into hell!
Plus the problem of having my data hosted, not just In another country, but In a different celestial body!
Ah shit, this is all for child porn again… Can the fucking tech industry just not do this every few months?!
Before even considering radiation damage, hopium $200/kg launch costs mean 15c/kwh electricity. The you add the cost of specialized panels and radiation emitters. At least 20x that of earthly systems.
Okay, but have you considered how cool it would be to put a data center in space?
What if I told you that we have to BEAT CHINA to space?
Ahhh China. What would US presidents do without China…
BEAT CHINA to space is for sure the magic words, but even better, what I told you to come up with an excuse to merge my space company with my AI company, and even though it is a paper transaction with no money changing hands, increase my wealth by $300B for the price I set!
hmmm… beat china does sound better.
China…CHINA!!
Gotta love the eternal threat of the “yellow peril”
Red Menace, Yellow Peril… Who’ve we got coming up for the Green Trouble?
Dumping heat in space is actually hard to do. You’d need huge radiators for radiative emission cooling.
Shame you can’t do some sort of thermoelectric power generation thingie with all the heat from these data centres.
You can’t turn pure heat into useful energy. Thermoelectric generators tap into the transfer of heat between a hot reservoir and a cold reservoir.
So assuming a water cooloant system, the water is heated and then the question how efficiently can you use this to generate energy? Even the simplest scenerio of pumping this water to a place where hot water is needed and would normally be produced by heating it with gas or elecetricity is a means of producing energy. Wouldn’t probably work here though as the water coolant system is a closed loop so you can’t have water leave the system. It still could pass through another reservoir of water to heat it up which then could be used for other purposes. But don’t know about specifics enough to guess whether or not this is feasible.
