Data centers in space makes no sense

I would not assume cooling has been worked out.

Space is a vacuum. i.e. The lack-of-a-thing that makes a thermos great at keeping your drink hot. A satellite is, if nothing else, a fantastic thermos. A data center in space would necessarily rely completely on cooling by radiation, unlike a terrestrial data center that can make use of convection and conduction. You can't just pipe heat out into the atmosphere or build a heat exchanger. You can't exchange heat with vacuum. You can only radiate heat into it.

Heat is going to limit the compute that can be done in a satellite data centre and radiative cooling solutions are going to massively increase weight. It makes far more sense to build data centers in the arctic.

Musk is up to something here. This could be another hyperloop (i.e. A distracting promise meant to sabotage competition). It could be a legal dodge. It could be a power grab. What it will not be is a useful source of computing power. Anyone who takes this venture seriously is probably going to be burned.

As far as I can tell, Data centres in space only seem viable because their advocates insist on comparing them to standard terrestrial data centres.

And nobody ever calls them out on it.

Data centres which are optimised for reliability, redundancy, density, repairability, connectivity and latency. Most of the savings come not from placing the data centre in space, but the fact that advocates have argued away the need for absolutely everything that modern data centres are designed to supply, except for the compute.

If they can really build a space data centre satellite for as cheap as they claim, why launch it? Just drive it out into the middle of the desert and dump it there. It can access the internet via starlink, and already has solar panels for power and radiators for cooling. IMO, If it can cool itself in direct sunlight in space, it can cool itself in the desert.

The main thing that space gains you over setting up the same satellite in the desert is ~23 hours of power, vs the ~12 hours of power on the ground. And you suddenly gain the ability to repair the satellite. The cost of the launch would have to be extremely cheap before the extra 11ish hours of runtime per day outweighed the cost of a launch; Just build twice as many "ground satellites".

And that's with a space optimised design. We can gain even more cost savings by designing proper distributed datacenter elements. You don't need lightweight materials, just use steel. You can get rid of the large radiators and become more reliant on air cooling. You can built each element bigger, because you don't have to fit the rocket dimensions. You could even add a wind turbine, so your daily runtime isn't dependant on daylight hours. Might even be worth getting rid of solar and optimising for wind power instead.

An actual ground optimised design should be able to deliver the same functionality as the space data centre, for much cheaper costs. And it's this ground optimised distributed design that space data centres should be compared to, not today's datacenter which are hyper-optimised for pre-AI use cases.

It seems like every argument in favor of doing this is: "yeah sure but what if X was Y% cheaper?"

And some of us are reading these things and trying to be polite.

But at some point patience runs thin and the only response that breaks through the irrationality is some variation of "what if unicorns and centaurs had teamed up with Sauron?"

The limit of the ratio of useful:useless "what if's" approaches zero.

I'm convinced that >30% of this comes from ideas leaking out of fiction such as like Neuromancer, and percolating through the minds of wealthy people attracted to some of the concepts. Namely, the dream of being a hyper-wealthy dynasty, above any earthly government, controlling an extraterritorial Las Vegas Fiefdom In Space. (Which in the book, also hosted a powerful AI.)

Then they work backwards, trying to figure out some economic engine to make it happen. "Data centers" are (A) in-vogue for investment right now and (B) vaguely plausible, at least compared to having a space-casino.

It seems quite telling we are even discussing this.

Very confused by this plan. Data centers on Earth are struggling with how to get rid of waste heat. It's really, really hard to get rid of waste heat in space. That seems to be about the worst possible place to put a data center.

You guys clearly didn't read the full blog post where Musk mentions lunar mining. They're going to put an ASML machine on the moon and turns regolith into chips and solar panels automatically. Literally free compute

One way to work around the heat dissipation issues in space (and also on earth) is to move to computing systems that operate entirely at cryogenic temperatures to take advantage of superconducting circuitry.

I've heard stories that over a decade ago teams inside hyperscalars had calculated that running completely cryogenically cooled data centers would be vastly cheaper than what we do now due to savings on resistive losses and the cost of eliminating waste heat. You don't have to get rid of heat that you don't generate in the first place.

The issue is that at the moment there are very few IC components and processes that have been engineered to run at cryogenic temperatures. Replicating the entirety of the existing data center stack for cryogenic temps is nowhere near reality.

That said, once you have cryogenic superconducting integrated circuits you could colocate your data centers and your propellant/oxidizer depots. Not exactly "data centers off in deep space" since propoxd tend to be the highest traffic areas.

They're gonna propose something dumb like ejecting coolant out into space as a disposable heatsink and then they're gonna spend a bunch of money trying to build a proof-of-concept but it will never go anywhere because it's really some kinda money laundering scheme or whatever the Hyperloop nonsense was.

As a thought experiment, if humanity wanted to go all in on trying to move industrial processes and data centers off planet, would it make more sense to do so on the moon?

The moon has:

- Some water

- Some materials that can be used to manufacture crude things (like heat sinks?)

- a ton of area to brute force the heat sink problem

- a surface to burry the data centers under to solve the radiation problem

- close enough to earth that remote controlled semi-automated robots work

I think this would only work if some powerful entity wanted to commit to a hyper-scale effort.

I bet they can already weaponize their satellites to prevent the launch of other satellites.

Putting data centers in space keeps them out of reach of humans with crowbars and hammers, which may have been a vulnerability for those robots Tesla is building.

AI data-centers use upwards of 100MW. The biggest solar panels in space could produce around 240KW. When they speak of AI data-centers in space what do they actually mean in realistic non theoretical terms and where are the materials for this coming from?

If the AI data-center used only 10MW then each could have two redundant SMR's assuming the cooling challenges have been worked out but then we could have nuclear reactor disposal and collision issues.

> Data centers in space only make sense if they are cost effective relative to normal data centers.

Author made a fatal mistake. By flying enough hardware in space, you can simply blot out the sun and steal their solar capacity. Drink their milkshake with a long straw!

What is this website?

The website insists that you let it record your voice in order to show you the dangers of AI. Is it trolling the visitor? https://civai.org/talk

How about we just make a giant heatsink that reaches into space instead. Then we can cool the whole planet. Coming up with crazy ideas is cheap, but the logistics are obviously impractical.

It possibly makes sense if you're preparing for war, harder to hit, harder to physically break into, beyond the range of nuclear EMP, and accessible from anywhere on earth.

I don't get the point at all of these. You:

- have very non-deterministic latency

- are located outside of a country that can protect you (ie China could disrupt your space data center)

- have to pay millions of dollars to swap out hardware

A month and a half ago: https://news.ycombinator.com/item?id=46286645

What data centers in space enable is protection for the compute of near-superintelligent AIs from the interference of humans.

As an alleged human, I'd like to preserve my option to interfere.

But it's absolutely amazing hype and memevestors love it.

Don't let common sense stop you from a good time.

I guess the xAI/SpaceX thing is mainly a financial move and they made up an interesting story to give it some context

Comments full of EDS. Everyone is a rocket scientist in here also.

To Steelman the topic, Musk’s whole alleged mission is to make humans a multi-planet species that can survive an earth killing event.

To that end, a small data center space isn’t about unit-economics, it’s a bigger mission. So the question we should consider is what can we put into space the further that mission. Can we put a meaningful sum of human knowledge out there for preservation? It sounds like “yes,” even if we can’t train ChatGPT models out there yet.

I'm not sure datacenters in space have to make suense to everyone, or from the perspective of earth.

Taking a creative step back, perhaps datacenters in space support something with Mars?

As much as that might not seem realistic, I also have to counterbalance it with operationalizing and commercializing SpaceX, Starlink and Tesla relatively quickly when so much stays at the R&D stage for so long.

It doesn't make any sense to me either, but there are lots of things like that where the other thing is harder. As an example, a thing people say online a lot is something like "Why do the techbros build self-driving cars instead of just putting it on rails for efficiency and then they could call it a TRAIN?"

The answer to that is that coordination problems are really hard. Much harder even than what are currently unsolved engineering problems. In fact, SpaceX can only launch from California because they have DOD coverage for their launches. Otherwise the California Coastal Commission et al. would have blocked them entirely. Perhaps the innovation for affordable space Internet is combining it with mixed-use technology.

The truth is that in America today self-driving cars (regulated by a state board run by bureaucrats) are easier to build than trains (regulated by every property owner on the train route). Mark Zuckerberg tried to spend some money evaluating a train across the Bay and had to give up. But Robotaxi service is live in San Francisco.

So if there is an angle that makes sense to me it's that they anticipate engineering challenges beatable in a way where regulatory challenges are not.

Is there any insight into how Starlink solved cooling? One 'expert' insisted that there is no reason to expect that data center satellites would generate any more heat than starlinks.

you know you're looking at some hard analysis when they use the number "gazillion". can I get that one in scientific notation?

I am willing to bet the whole xAI/SpaceX merger is simply a ploy by Musk to evade releasing accurate historical information about SpaceX's finances. How much did it actually cost SpaceX to launch a kilogram of payload into space each year? How much is NASA actually donating them, per each year?

I mean, I still remember promises of $1000-per-kg for space launches, and how e.g. Gigafactory will produce half of the world battery supply, and other non-scientific fiction peddled by Musk. Remember when SpaceX suggested in 2019 that the US Army could use its Starship rockets to transport troops and supplies across the planet in minutes? I do. By the way, have they finished testing Starship yet, is it ready?

This is written by someone that is not in aerospace that thinks terrestrially.

Engineering is always a question of tradeoffs.

Launch costs are dropping, and we’re still using inefficient rockets. Space elevators & space trains, among others, can drop this much more, the launch costs are still dropping, even using rockets, maybe we’ll never get to elevators & trains the costs will drop so low!

Radiation shielding is not required for VLEO or LEO, and phenomenally more capable aerospace processors are near - hi Microchip Inc! There are many other radiation solutions coming, no doubt with nuclear power.

Satellites can be upgraded at scale, though for many things, it does not make $ sense to upgrade them, but fuel , reaction wheels, solar panels, among other things do make $ sense to replace.

Latency was technically solved in 1995 & 2001 with the first laser comms missions NASDA’s ETS-VI kiku-6 and ESA’s Artemis , and Laser crossbars for comms are common. A full laser TDRS no RF is not yet extant but soon. Earth to deepspace was just demonstrated by ESA.

Cooling can be significantly improved due to lower launch costs, heat piping, RTGs, TEGs, and thermoradiative cells, not to mention sunside solar and darkside inline radiators

Furthermore, it is very likely that as neuromorphics with superior SWaP emerge, we could see very different models of space based computation.

Economic tradeoffs should drive many of these decisions as I’m not discussing the other applications of datacenter in space

If we won't stop what he is doing with grok and ai-generated CSAM, he will be completely free from oversight up there.

The bigger issue: datacenters in space are disposable. All the extremely recyclable aluminum, silica - you extract it, manufacture it and instead of recycling it when it’s done you incinerate it in the atmosphere and scatter the ashes far and wide across the earth, the harder to recapture later.

You do this when the most fragile part in the system fails. Solar panels good for 25 years but the SSDs burn out after 2? Incinerate the lot!

This kind of thinking is late capitalist brain rot. This kind of waste should be a crime.

No no, let Musk cook. This definitely won't be SpaceX's Cybertruck moment, where they completely throw away their first-mover advantage by wasting five years chasing after the egotistical boondoggle of a delusional megalomaniac.

It's lala land nonsense.

- Data centres need a lot of power = giant vast solar panels

- Data centres need a lot of cooling. That's some almighty heatsinks you're going need

- They will need to be radiation-hardened to avoid memory corruption = even more mass

- The hardware will be redundant in like 2 years tops and will need replacing to stay competitive

- Data centres are about 100x bigger (not including solar panels and heat sinks) than the biggest thing we've ever put in space

Tesla is losing market share (and rank increasingly poorly against alternatives), his robots are gonna fail, this datacentre ambition needs to break the laws of physics, grok/twitter is a fake news pedo-loving cesspit that's gonna be regulated into oblivion. Its only down from here on out.

There are two very distinct kinds of AI workloads that go into data centres:

    1. Inference
    2. Training

Inference just might be doable in space because it is "embarrassingly parallel" and can be deployed as a swarm of thousands of satellites, each carrying the equivalent of a single compute node with 8x GPUs. The inputs and outputs are just text, which is low bandwidth. The model parameters only need to be uploaded a few times a year, if that. Not much storage is required , just a bit of flash for the model, caching, logging, and the like. This is very similar to a Starlink satellites, just with bigger solar panels and some additional radiative cooling. Realistically, a spacecraft like this would use inference-optimised chips, not power-hungry general purpose NVIDIA GPUs, LPDDR5 instead of HBM, etc...

Training is a whole other ballgame. It is parallelisable, sure, but only through heroic efforts involving fantastically expensive network switches with petabits of aggregated bandwidth. It also needs more general-purpose GPUs, access to petabytes of data, etc. The name of the game here is to bring a hundred thousand or more GPUs into close proximity and connect them with a terabit or more per GPU to exchange data. This cannot be put into orbit with any near-future technologies! It would be a giant satellite with square kilometers of solar and cooling panels. It would certainly get hit sooner or later by space debris, not to mention the hazard it poses to other satellites.

The problem with putting inference-only into space is that training still needs to go somewhere, and current AI data centres are pulling double-duty: they're usable for both training and inference, or any mix of the two. The greatest challenge is that a training bleeding edge model needs the biggest possible clusters (approaching a million GPUs!) in one place, and that is the problem -- few places in the world can provide the ~gigawatt of power to light up something that big. Again, the problem here is that training workloads can't be spread out.

Space solves the "wrong" problem! We can distribute inference to thousands of datacentre locations here on Earth, each needs just hundreds of kilowatts. That's no problem.

It's the giaaaant clusters everyone is trying to build that are the problem.

How else would you secure skynet against Sarah Connor?

We were supposed to be on Mars right now, but I guess data centers in space are nice too. Kinda disappointed they aren't on the moon.

Next up, the Boring Company gets imaginary contract for underground datacenters, is now valued at $500B.

how much latency would a minecraft server in space have?

Facts.

Just do the basic thermal heat transfer math.

I'd be curious to know simply how large the thermal radiator necessary to keep a typical GPU server cooled would be. Do they completely dwarf the server size? Can you do something with some esoteric material that is not particularly load-bearing but holds up well in space to get around some of these challenges?

Admiral Grace Hopper is famous for using a length of wire to explain to others what a nanosecond was.

https://www.pbs.org/newshour/world/pentagon-embraces-musks-g...

Data centers in space make absolute sense when you want as close to real time analysis on all sorts of information. Would you rather have it make the round trip, via satellite to the states? Or are you going to build these things on the ground near a battlefield?

Musk is selling a vision for a MASSIVE government contract to provide a service that no one else could hope to achieve. This is one of those projects where he can run up the budget and operating costs like Boeing, Northrup etc, because it has massive military applications.

Like I’ve always said love him or hate him Elon Musk is a SPACE OIL SALESMAN!

I can assure this author: strapping a company that lights money on fire (today, maybe not tomorrow) to a cash flow enterprise makes the IPO harder, not easier, in the absence of credible plan. The market speculates, but it’s not being completely irrational. I’d actually be surprised if we didn’t have factories or data centers in space one day.

[dead]

Data centers in space are the logical progression from the multi trillion business of m2m and edge computing. It removes all physical limits to investment.

Counterpoint: https://x.com/CJHandmer/status/1997906033168330816

(If you can't xcancel it yourself your hacker card is revoked.)

*Data centers in space only make sense if they are cost effective relative to normal data centers*.

Disagree there are bunch of scenarios where Data Centers in space make sense. Like nuclear annihilation and having vaults across the globe to communicate and get back lost information because ground data centers would be wiped out by EMP from blasts.

If you read these comments carefully, you see that they can all be summed up as:

"That Musk guy is so naive to think you can put data centers in space, what a doof".

Similar comments were probably made regarding electric cars, reusable rockets, buying Twitter, and so on.

Space offers some unique benefits that enable computing that’s impossible or very hard to do on earth. E.g. Super conducting computing is possible, which can be thousands times to millions times faster than current CPU while using very little energy. When the satellite moves in the shade of the earth, temperature drops significantly. It can be low enough to enable superconducting. When the satellite moves under the sun, the solar panel can start charging up the battery to power the ongoing operation.

What’s there not to like? Superconductors. Free electricity. No cooling necessary.

Put those three together and maybe it’s possible to push physics to its limits. Faster networking, maybe 4x-5x capacity per unit compared to earth. Servicing is a pain, might be cheaper to just replace the hardware when a node goes bad.

But it mainly makes sense to those who have the capability and can do it cheaply (compared to the rest). There’s only one company that I can think of and that is SpaceX. They are closing in on (or passed) 8,000 satellites. Vertical integration means their cost-base will always be less than any competitor.