Forgot your password?
typodupeerror
Space

The Space-Based Data Center Hype Machine Is Already In Orbit (ieee.org) 195

IEEE Spectrum argues that orbital data centers remain far from economically or technically practical despite Elon Musk's prediction that space will become the cheapest place to run AI within a few years. Deploying SpaceX's proposed million-satellite constellation would require enormous increases in launch and manufacturing capacity, while cooling, radiation, maintenance, latency, orbital debris, and astronomical interference present major unresolved obstacles. Longtime Slashdot reader xetdog shares the report: Consider this: There are roughly 14,500 active satellites in orbit. Musk's Starlink constellation accounts for about two thirds of those. Both the launch cadences and satellite-manufacturing capacity would have to scale up astronomically to deploy a million orbital data center satellites. For context, there have been roughly 7,000 orbital launches in all of human history. To loft 1 million satellites into low Earth orbit on SpaceX's Starship, which is designed to carry up to 60 satellites per vehicle, would require 16,666 launches exclusively devoted to satellite deployments. Considering that SpaceX launched a record 165 orbital missions in 2025, even at 10 times that cadence, it would take a decade. And how long would it take to build 1 million satellites, given Starlink's current pace of around 4,000 per year and a generous tenfold increase in capacity? Short of a manufacturing revolution, try 25 years. Dissipating heat in space also requires enormous radiators. As IEEE Spectrum editor Dina Genkina noted, startup Starcloud has sent only one Nvidia H100 GPU into orbit, and "their radiator was too weak to let the chip run at full power." A single 700-watt H100 would require about 1.4 square meters of radiator area, while a 100-megawatt data center could need 2,500 radiators measuring 80 square meters each.

So, why are the hyperscalers hyping orbital data centers? Answer: because it's lucrative. "The Elon Musk part of it is honestly genius because he's got xAI building the data centers, SpaceX sending them to space, and Tesla building solar panels," Genkina says. "It's almost like he's paying himself."

The Space-Based Data Center Hype Machine Is Already In Orbit

Comments Filter:
  • So basically... (Score:4, Informative)

    by DrXym ( 126579 ) on Thursday July 02, 2026 @07:08AM (#66219592)
    ... it's just another pack of lies like everything else Musk hypes up.
    • Re:So basically... (Score:5, Insightful)

      by Anonymous Coward on Thursday July 02, 2026 @08:31AM (#66219652)

      Not lies, a carefully crafted scheme to defraud investors of their money

      • If that was the intent, it wouldn't really work due to Elon himself having more downside exposure than anybody, including all institutional investors combined.

        Besides, some of this argument is over engineering problems that have already been solved.

        • by ceoyoyo ( 59147 ) on Thursday July 02, 2026 @12:56PM (#66220036)

          What's the downside? SpaceX stock got pumped for their IPO. The money is made. As long as the hype keeps going they can raise more any time they want, or Elon could sell some of his shares. If it turns out to be unworkable, SpaceX (and subsidiaries) are back where they started.

          There aren't really any unsolved engineering problems. SpaceX can absolutely put a rack of nvidia GPUs into low orbit. We could have done that in the 70s. The argument is whether it's economical or not.

          Which is cheaper, putting a thousand square metres of solar panels, a rack of GPUs, a vacuum cooling system and propulsion in low orbit and incinerating and replacing it all every few years, or the panels, GPUs and a convective radiator that is ~50x more efficient on the ground and runs for twenty plus?

    • Re:So basically... (Score:5, Insightful)

      by evanh ( 627108 ) on Thursday July 02, 2026 @09:42AM (#66219728)

      The scary part is Musk isn't alone. The insanity of the LLM data-centre build out has got far too much money thrown at it already. There's no way it will be recouped. It can only end badly now.

    • It's part of the Golden Dome [wikipedia.org] anti-missile defense. The architects have decided it's important to process missile tracking data in space, rather than sending it down to earth for processing.

      They have a whole network of things up there [wikipedia.org] processing and communicationg.

      I haven't looked much at the architecture, but I suspect it's kludgy, completely lacking elegance.
      • by BranMan ( 29917 )

        If that is the case, then it all makes a whole lot more sense. There is no need for cost-efficiency when dealing with military equipment - the only question to answer is does it do what is needed for the GD system? And if what is needed is processing tracking data and generating firing solutions locally, without multiple trips back and forth to orbit, then making it work is the key, not cost or complexity (to some degree, even the DoD has limited funds).

    • I think satellite data centers are colossally stupid, but I suspect the larger problem is the public's gullibility for big lies.

      Now, which things ARE lies and which aren't has been delightfully co-opted by politics; what one puts on that list is *instantly* translated into political affiliation.

      I can think of 3 big lies that would immediately get me labeled "stupid maga fuck".
      I can think of 3 others that would likewise get me labeled "woke fag".
      Amusingly, putting all 6 in a list would be cognitively negativ

    • by gweihir ( 88907 )

      Yes. One wonders how this guy got rich. He is probably very persuasive, at least to weaker minds.

    • Re:So basically... (Score:5, Informative)

      by swillden ( 191260 ) <shawn-ds@willden.org> on Thursday July 02, 2026 @02:32PM (#66220220) Journal

      ... it's just another pack of lies like everything else Musk hypes up.

      Counterargument: Who would have predicted a few years ago that one private company would dominate global launch, launching more by every metric than the rest of the world combined, and -- all by itself -- triple the number of satellites in orbit in 7 years.

      Sure, 200Xing the satellite count is a lot harder than tripling the satellite count, about 66 times harder. But if Starship is successful (by no means a given, also far from impossible), SpaceX will reduce per-kg launch costs by 100X, maybe more.

      I'm skeptical... but I would also not just write it off as a "pack of lies". The things SpaceX is actively working on should make the launch part of it feasible. Will it be cost-effective? That's a harder question, and heat dissipation is the core thing that may make it infeasible.

      Also, the final paragraph of the summary seems to be confused:

      So, why are the hyperscalers hyping orbital data centers? Answer: because it's lucrative. "The Elon Musk part of it is honestly genius because he's got xAI building the data centers, SpaceX sending them to space, and Tesla building solar panels," Genkina says. "It's almost like he's paying himself."

      Yes, SpaceX will be incredibly lucrative if it owns the whole vertical stack, building, launching and powering -- but only if it works. If it doesn't work, and if orbital compute isn't cheaper than planet-bound compute, then SpaceX will have no buyers.

      The other possibility is that it's just a pump and dump, but that's not how Musk has ever worked in the past. Yes, he makes crazy promises, and delivers only half of them, and delivers years after the promised date, but those half-realized, years-late results are still often world-changing.

      • I credit most of SpaceX's success to CEO Gwen Shotwell. She keeps things going even when Musk is off on an irrelevant tear somewhere else.

        Unfortunately, Musk seems to be on a path to sabotaging her efforts. The SpaceX prospectus showed that xAI (which bought Twitter, because why not?) was the reason they posted a loss in the last fiscal year. Even with all the expenditures on Starship, SpaceX would have been profitable. Like every other major AI company, it is not at all clear that xAI can reach profitabili

  • Another bubble. Yippee-fucking-woo
    • More evidence we need to tax the rich, they have way too much money and are chasing really stupid shit. After you have 20 yachts then #21 doesn't have the same ego kick, so you look for pie-in-sky investments.

    • This isn't even a bubble. This is private equity and billionaires pipe dreaming. And I'm talking about the crack pipe.

      SSDs have a measurable lift in failure rates just being installed into aircraft due to cosmic radiation causing bit-shifts in the controller firmware. What is going to happen to this shit when it's even higher up, and harder to replace and service?

      These are disposable data centers, with orders of magnitude more complexity for shedding heat. It's never going to make sense until we can get

  • I need you to suit up and ride the next rocket to space. We need to upgrade the RAM on one of the space servers.
    • We don't build high-density datacenters with plans to EVER touch failing hardware. When a node breaks, it's taken off-line, not repaired. When an upgrade is needed, the datacenter is turned-over/forklift upgrade - no one runs into a google datacenter and talks about adding more RAM.

      At one time we talked about dropping watertight datacenters in the ocean - this is kinda the same idea, but it includes rockets! (And, relies on wireless technology and solar panels, ocean-based datacenters could be wired into a

  • It's the same reason Elon Musk made that stupid fucking tunnel. It's misdirection.

    So this lets them guzzle down all your water and electricity so that you have to start rationing both of them while they go to public and tell the public that they have a solution to the problem.

    As an added bonus Elon Musk can pretend SpaceX has a magic new customer that doesn't actually exist in order to justify the phony valuation while he extracts all the money from your 401k.
    • Do you not realize how silly that sounds? He wants to try and put datacenters in space in order to steal your water and power?

      The rest of the world is not out to get you.

    • by AmiMoJo ( 196126 )

      That's it, he is doing it to generate fake potential future business for SpaceX. In reality, what will probably happen is others enter the market and undercut Starlink and launching to LEO, and that business dries up. It's already happening with Tesla cars, which are constantly heavily discounted because rivals make better ones at lower prices. They are hanging on in the UK by somehow being a "prestige" badge along side BMW and the like... Actually I can see the connection there.

  • And they know it, radiation, heat and cost. Another problem is kessler syndrome and space debris. There is a limit to how much stuff we can put up there.

    • Re: (Score:2, Insightful)

      by sabbede ( 2678435 )
      Neither was half the other stuff people said he couldn't do but then did. If he thinks it's worth trying, let him. It's no skin off your nose.
      • Space Data Centers are in the same category as fully autonomous self-driving cars within eighteen months that he 'promised' in 2019.

        You can watch the 'Autonomy Day' video on YouTube. People financed Model 3's on the promise of renting them as robotaxis while they were at work.

        Physics is a hard stop on false promises.

        It's OK to back difficult challenges with no underlying physical impossibilities that's engineering. Radiating heat into space is a physics problem.

        I didn't believe the robotaxi promise then an

        • "It is not the critic who counts; not the man who points out how the strong man stumbles, or where the doer of deeds could have done them better. The credit belongs to the man who is actually in the arena, whose face is marred by dust and sweat and blood; who strives valiantly; who errs, who comes short again and again, because there is no effort without error and shortcoming; but who does actually strive to do the deeds; who knows great enthusiasms, the great devotions; who spends himself in a worthy cause
    • That and Russia has suggested destroying the orbit if Starlink doesn't stop enabling terrorist attacks by NATO proxies on its people.

      The AI cut-off dive bomb tactic is a blatant warcrime.

    • by gweihir ( 88907 )

      Indeed. Maybe a meaningless stunt can be done that will then become more space-debris within a few months or years. But that is it.

  • I look forward to Musk overcoming the limits of insolation in Earth orbit, the latency induced by the speed of light, and the Stefan-Boltzmann law.

    This is just as good an idea as the submarine he ordered built for cave rescue: it appeals to idiots who don't give any thought to the problem but think the proposed solution is 'cool'.

  • by smoot123 ( 1027084 ) on Thursday July 02, 2026 @10:46AM (#66219818)

    AI data centers are evolving rapidly. You want to be able to rip and replace gear all the time. That's expensive to do in orbit.

    Managing heat in a data center is a huge issue. It's very hard to dissipate heat in space. It's a lot easier to dump waste heat into the air or a river.

    Communication between nodes, racks, and rows is a fundamental limit to AI performance. Spacing rows hundreds to thousands of kilometers apart is going to add speed-of-light delays and bandwidth limits. I don't see how you maintain performance in that environment.

    All in all, this seems like an incredibly impractical idea. It really screams "I have rockets and want to find uses for them." But Musk is a smart guy, I have no doubt he's thought more about it than I have.

    • But it is easy to hype because most people don't understand the technical issues. And people are way more likely to hero-worship than to be rational. And it ties together all the tech pieces he has build (mostly), so it is fantastic amplification of his brand.
  • Didn't Heinlein write a novel in which someone built a data center on the Moon, out of off-the-shelf Chinese hardware?

    Ah, got it: Temu is a Harsh Mistress.

  • The company I work with can easily ramp up satellite production to the scale required. It's one of the reasons I got hired, my extensive manufacturing experience in electronics and solar and power systems pairs perfectly with the requirements.

    Perhaps the IEEE should spend some actual time with the companies that already have some of this hardware in orbit, with more going up soon.

  • They resell shit PV from China
  • While a BSc is the lowest level of academic Physics, it should be enough to easily determine that data-centers in space are a really dumb idea. Well. I guess he essentially bought that degree.

  • by jd ( 1658 ) <imipak @ y a h o o . com> on Thursday July 02, 2026 @02:47PM (#66220244) Homepage Journal

    The only way you can lose heat in space is through radiation. But radiation carries momentum. Not much per photon, but it was enough to cause the Pioneer probes to move in unexpected ways. This means you have to emit equal amounts of heat towards Earth and towards space. If your resultant is zero, then you're fine. You can even direct some of the heat backwards. It won't do a huge amount, but every bit of atmospheric drag you overcome, the less fuel you need to use to stay in orbit.

    So you basically need absolutely gigantic radiators behind the space-based data centre, located inside a parabolic dish that will generate drag of its own (not to mention a potential difference betwen the lower and upper sections).

    This is an insane level of complexity. You're better off parking it in a stable orbit between the Earth and the moon, so it's absolutely clear of atmospheric effects. You're still going to need radiators, but it's marginally better as you don't have to do quite so much directing of it. The latency would be horrible, maintenance would be next to impossible, and there's all kinds of other issues to consider.

    No, I don't think you can make this workable.

    However, space might be useful. This very same issue of heat only being radiated means that you can make wafers with much more even loss of temperature, no dust, bacteria, or dirt, and much lower gravity. If you were to make extremely high quality wafers (silicon or gallium arsonide) in space, then you should be able to make WSI processors, which should in turn reduce the demands that datacentres make.

    The time it would take to set all this up would be about the same time as it took for IBM to perfect its stacked transistor topology. Intel was talking 90 cores per wafer-scale CPU a few years back - the shrinkage in transistors since then plus the x10 density IBM proposes might push you to 1800 cores per wafer, provided you can get the quality high enough. Which, in space, is quite possible.

    You wouldn't need your datacentres in space. Your wafer-scale CPU plus packaging would be about the same size as a CD drive. You could pretty much dispense with datacentres at that point. A typical tower will have two spare bays. "Cartridge datacentres" could simply be plugged in as needed. A regular CPU-based cartridge for heavy general-purpose computing, a GPU-based cartridge for LLMs. Yes, home users would have power usage through the roof, but then it's no longer your problem.

How much net work could a network work, if a network could net work?

Working...