Orbital data centers, part 1: There’s no way this is economically viable, right? | “This is not physically impossible; it’s only a question of whether this is a rational thing.”
https://arstechnica.com/space/2026/03/orbital-data-centers-part-1-theres-no-way-this-is-economically-viable-right/
25 Comments
There’s no way for computer servers in space to cool themselves as quickly as on Earth. Convection isn’t available, only radiation. Which is far slower. Add to that: any semiconductors in space need to be hardened against radiation. That’s why space electronics usually aren’t just one generation behind, they’re several generations behind state of the art terrestrial tech. There is simply no way a chip like Nvidia’s H series or Vera would survive in a hard radiation environment.
By now folks should realize that when Elon Musk promises (soon!)
1. Living on Mars
2. Vacuum Train tunnels
3. Androids
4. Robotaxis nationwide
That he is deep in a K-hole and dreaming up ways to goose his stock price.
Sure, data centers in space aren’t physically impossible. They’re just incredibly stupid from just about every viewpoint you’d care to look at it from.
The solar arrays on the ISS produce about 120KW of power. The typical power used by datacenters range from about 20MW to 100MW. This would require approximately 420,000 square meters of solar panels for a 20MW data center. This would be a square that is approximately 6.5 football field on each side.
And then, you have to get rid of the heat. There’s no transfer medium in space, so all heat would have to be radiated. The radiators needed to dissipate the waste heat would also be massive.
It would take many many trips to build just one small datacenter, and then maintenance costs afterward would be massive as well.
It’s hard to believe serious people are actually considering this.
It feels at best like putting the cart before the horse. I’ve had people disagree with me on this. There are serious technical challenges, like coolant and getting enough material up there to matter. I feel like if you want to make a project of this magnitude, it would be a better bet to make a space based solar power station. That would get you a lot of juice that could be used for applications beyond AI, and solving the engineering hurdles would generate a bunch of useful spinoff technologies.
Again, what are they planning on doing about all the heat?
Yeah you CAN build one if you want to, there’s no physics forbidding it, and all necessary technology exists. But there’s no comparative advantage to ground based centers. The constant solar power is a relatively miniscule boon that doesn’t make up the associated extra costs at all.
Powering a data center in space would be a nightmare, cooling it would be a nightmare, maintaining it when stuff breaks down would be a nightmare. A project like this would cost 1000x more than just building the same capability on the surface.
The question is not “can you” but “why would you want to?”
It it *certainly* technically possible. There is nothing preventing anyone from constructing large foldable solar arrays and large foldable radiators – those things currently fly on many spacecraft, including Starlink sats.
It *could* be economically possible, but requires a large number of technological advancements in parallel:
1. You need launch cost to drop to around $100/kg. The theoretical limit for Starship is around $20/kg, but it obviously isn’t done or anywhere near that mark yet. Current best is Falcon around $1,000/kg. None of this is possible if Starship doesn’t work as planned.
2. You need sat power density to hit around 100 kW/ton. That requires a number of improvements:
* Chip junction temps need to rise to above 100 C sustained, which drops radiator area needed.
* Solar panels need to go > 1,000 W/kg. Current is around 300 W/kg
* Radiators need to go < 4 kg/m^2. Current is around 8 kg/m^2
3. You need the chips to not die
* Current failure rate is ~9%/year on the ground
* You need around CPU failure rate – 1.5%/year, but in Space with radiation
4. You need inter-satellite links to improve. Current laser links on Starlink handle 25 Gbps – **way** slower than connecting server racks with ethernet cables. You need around 1Tbps.
None of these are impossible. All are very hard.
Relevant Scott Manley video:
https://youtu.be/FlQYU3m1e80?si=nl97aOUwEOCUa9KW
Also, people are failing to understand the regulatory headwinds that data centers face. It’s not that space makes more sense. Its that the permitting process in the US is so difficult, that its actually more straightforward to put them in orbit. Localities simply don’t want data centers.
So I don’t disagree with all the assessments I’m seeing in this thread, just pointing out what’s pushing them in that direction.
Wouldn’t you be better to put data centres on the moon surface instead?
Although it still wouldn’t be ideal, at least you could have the heat dissipate in the cold surface of the moon.
In space, those orbital data centres are going to heat up extremely quickly with no way to shed all that access heat. You’d have to over engineer them to deal with the heat build up, and at that point you might as well just put them in the ocean.
It’s a ridiculous proposition that anybody with any familiarity of space tech can see. But the product isn’t actually data centers, it’s a financing scheme with a far fetched technology as a tagline. Looking at it that way, it makes a whole lot more sense.
Didn’t google or Microsoft play around with underwater data centers, and couldn’t make it work?
I’d think underwater is a lot easier than outerspace, with the benefit of passive cooling.
I’d rather an orbital data center than the one they’re lobbying to be built down the street from me
If this hinges on silicon getting cheaper, that will effect non space data centers too. There is room for compute in space but only for computers that actually need to be in space like communication satellites and telescopes. Running Gemini in space for people on earth is a dumb idea.
Acronyms, initialisms, abbreviations, contractions, and other phrases which expand to something larger, that I’ve seen in this thread:
|Fewer Letters|More Letters|
|——-|———|—|
|[ASIC](/r/Space/comments/1s2ktff/stub/ocad8oi “Last usage”)|Application-Specific Integrated Circuit|
|[BFR](/r/Space/comments/1s2ktff/stub/oc9uq2x “Last usage”)|Big Falcon Rocket (2018 rebiggened edition)|
| |Yes, the F stands for something else; no, you’re not the first to notice|
|[ESA](/r/Space/comments/1s2ktff/stub/oc9uq2x “Last usage”)|European Space Agency|
|[FAR](/r/Space/comments/1s2ktff/stub/oc9uq2x “Last usage”)|[Federal Aviation Regulations](https://en.wikipedia.org/wiki/Federal_Aviation_Regulations)|
|[FCC](/r/Space/comments/1s2ktff/stub/ocakz95 “Last usage”)|Federal Communications Commission|
| |(Iron/steel) [Face-Centered Cubic](https://en.wikipedia.org/wiki/Allotropes_of_iron) crystalline structure|
|[LEO](/r/Space/comments/1s2ktff/stub/oc9uq2x “Last usage”)|Low Earth Orbit (180-2000km)|
| |Law Enforcement Officer (most often mentioned during transport operations)|
|[MMOD](/r/Space/comments/1s2ktff/stub/oc9qhan “Last usage”)|Micro-Meteoroids and Orbital Debris|
|Jargon|Definition|
|——-|———|—|
|[Starlink](/r/Space/comments/1s2ktff/stub/ocad8oi “Last usage”)|SpaceX’s world-wide satellite broadband constellation|
Decronym is now also available on Lemmy! Requests for support and new installations should be directed to the Contact address below.
—————-
^(8 acronyms in this thread; )[^(the most compressed thread commented on today)](/r/Space/comments/1s2ocv8)^( has 10 acronyms.)
^([Thread #12269 for this sub, first seen 24th Mar 2026, 20:40])
^[[FAQ]](http://decronym.xyz/) [^([Full list])](http://decronym.xyz/acronyms/Space) [^[Contact]](https://hachyderm.io/@Two9A) [^([Source code])](https://gistdotgithubdotcom/Two9A/1d976f9b7441694162c8)
It’s always going to be cheaper to submerge silicon in a watertight shipping container into the ocean for cooling and running some fibre to it than it will be to loft the same compute capacity into space. Tsiolkovsky’s tyranny remains. Juno has proven you can definitely harden against extremely hostile radiation environments, but can we do it on the scale a cost-effective data centre would need? Probably not, and certainly not cost effectively.
That doesn’t mean they won’t exist – I’m sure they will. But they’ll always be far weaker than terrestrial installations.
It’s not physically impossible, it’s very, very, very physically impractical though.
Each problem has a solution which spawns a tree of sub problems which recurses.
You can’t just launch a normal server rack into space. And expect it to run.
Furthermore, Imagine your rack has some Direct Attached Storage. 20 drives each with an expected failure rate of 1/10 years, you’re going to need to change a drive every 6 months. This applies on the ground too. Before you start adding in the failure rate of, phase change coolers, PV arrays, power circuitry, UPS systems, all those things a data centre technician usually looks after, and all the machinery an astronaut typically looks after on a space station need to be taken care of. And as we add complexity, the support requirements increase additively. Obviously, we probably expect very few failures in early operation, followed by a blitz of failures as time goes on rather than a linear failure rate, so the plan might be to build it to last to year 5 and then de-orbit, but that’s so much money, so much engineering time, so much waste to do the same thing that’s already a solved problem on earth.
As you say, it’s most likely economically inviable, even if you’re weighing it against a new DC for a large multinational built, for some reason, in the centre of Los Angeles, and powered by generators burning whale oil.
I think we may be missing the point. Moving AI data centers off planet achieve only one thing… it puts them far away and out of reach of any normal person or non-state actor.
Soon AI will drive a wave of unemployment and dissatisfaction the likes of which humans have never seen. Horses have, but humans haven’t. If these data centers are here on earth, with earthly connections and earthly resources… the angry mob will destroy them.
From orbit… any pizza-bix sized phased array will keep the data flowing from the data centers to the robots, agents, and systems they drive here on earth.
It isn’t about cost, or sense… its about security. From you.
There will be so much space debris as things stop working and replacements are sent out.
They will need to solve the issue of overheating.
I s’pose it’s the same level of “feasibility” as having ski slopes in the middle of the desert.
You can have them, it’s just a matter of how much money you want to throw at them.
While not physically impossible, we will quickly find it is fiscally impossible
With a full time IT crew on board with spare parts?
Hardware fails all the time.
They’re only economically useful as a tax write-off for the ultra-wealthy…..
Naah its nonsense. Its not a question of being possible, its a question of whats the point? You get ridiculous extra costs and complications but no benefit.
I’m going to install a raspberry pi in orbit and a honking big data centre on the ground, and then claim that the orbital one does all the work and isn’t subject to any jurisdiction on earth.
How close are we to the amount of junk flying around the Earth becoming a significant problem?