Orbital Data Centers: Why the Hype Outpaces Reality(spectrum.ieee.org) |
Orbital Data Centers: Why the Hype Outpaces Reality(spectrum.ieee.org) |
Just because Scott Manley refuses to call that out, so he can do another eight videos about it, don´t stop listening to somebody with the feet on the ground:
"Orbital Data Centers: Spacecraft Constraints and Economic Viability" - https://arxiv.org/abs/2604.27197
"Hot AI in Cold Space: Thermal-Crosstalk-Aware Scheduling for Sustainable Orbital AI Clusters" - https://arxiv.org/abs/2606.26150
"Above the Cloud: Building Data Centers in Space - Richard Campbell - NDC Copenhagen 2026" - https://youtu.be/eo7MEPgWGic
"Space Data Centers Are Dumb" - https://youtu.be/-w6G7VEwNq0
Once you understand this about Musk, you realize that everything he is involved with works that way.
If it ends up bankrupting them, even better.
Remember when globally competitive electric cars, re-usable boosters, catching a rocket with chopsticks, playing a fps game via a brain implant, and maintaining a satellite constellation at 480km LEO were also impossible?
From the SpaceX S-1: "For launches of our Starlink satellites, the Company does not recognize any inter-segment revenue, rather those launch costs are capitalized in satellites in Property, plant, and equipment, net." In plain terms: SpaceX's rocket division charges outside customers roughly $102 million per Falcon 9 launch — but it charges Starlink $0.
SpaceX appears to be heavily subsidizing Starlink in the launch cost sense.
There is a mix of clever and sketchy accounting going on in Muskworld making it hard to see which elements profitable/sustainable and what isn't.
The only reason I can imagine for space-based data storage is being out of reach of most police investigators. Recently, a very corrupt banker in Brazil was arrested and his phones were confiscated and, up to now, two have been cracked, with a big effect on the approaching election. If the thing is in orbit, it's a lot harder to confiscate it.
But I also now look at actual results, and they are not merely late, they are far less than promised.
Globally competitive EV? Not against the Chinese competitors (Hint: it's why he plans to merge Tesla into SpaceX). Self-driving so good cars would be assets you could run your own self-driving Uber-like service (and customers paid a $10K upcharge for the ticket to do it next year)? NOPE, Tesla can't even do their own self-driving reliably. And whatever happened to the Tesla solar roofs? Can't get one.
Making a car company profitable enough to justify the insane multiples? NOPE, will never happen. But he's selling Tesla now on the Humanoid Robot "vision".
SpaceX did better on self-landing boosters, but they still have serious issues with scaling it up, and competitors are catching up.
But the physics of orbiting data centers make them not absolutely impossible, but very uneconomical. Every problem supposedly solved in space is easier and orders of magnitude cheaper to solve on the ground.
If you haven't yet noticed the pattern, Elon is always selling the next big hype wave. It is always the NEXT thing that will justify the insane multiples. If you want to buy and hype meme stocks on the next-greater-fool theory, good luck; you will do well for a while as there are many fools around.
He still hasn't built a fully self-driving car. His last electric vehicle (the only one he directly designed) was a massive flop. The successful Tesla vehicles are losing market share because they're old and because of the CEO's odious politics and because he seems to have lost interest in EVs unless they're robotaxis.
Just recently he scammed Wall Street with a horrific overvaluation for a company whose only profits come from Starlink and whose biggest rockets blow up before they reach orbit.
Now he's talking about orbital data centers which make no sense to any engineer who understands thermodynamics (and sadly, many don't) and which are unnecessary because within the next 5 years 90% of inference will be done on local machines in users' homes because the hardware and the algorithms will be good enough to enable it and every Joe and Jane are fed up with surveillance capitalism.
Not really, thermal radiators are well understood and the size needed aren't really that unreasonable. Radiation hardening the GPUs is probably the single hardest problem along with actual launch costs.
>making no economic sense
Yes this is the real issue, Spacex would need to reduce the cost of launches by ~10x with Starship for it to ever be viable.
* https://images2.imgbox.com/64/4b/GhtUtq1m_o.jpg
look at how massive the solar panels and more importantly the thermal radiators are outside
compare to the size of the human habitat tubes
"AI" in space would demand as much if not even more for each node
Of course, that's offset by the enormous heat output of a datacenter. I'm not saying it's viable, just that you actually need to do the math instead of eyeballing the ISS radiators.
Now work is being done for increasing solar access (Starcatcher) and others are working on improving LEO refueling and repair capabilities, but I would say we're decade+ away from establishing any true compute infrastructure in orbit.
Short of launching an ASAT and the risk of space debris, nobody will be able to shut them down.
This is always a funny "gotcha" people bring up. Except the gotcha is always different somehow. Sometimes it's "the US subsidizes it" because NASA gets amazing deals on launches it would otherwise pay 10x or more for. Sometimes it's Starlink that subsidizes the launches. Sometimes it's the launches that subsidizes Starlink.
Maybe the answer is just what is obvious: vertical integration and economies of scale makes starlink/falcon 9 profitable. The combination is the thing.
The guy has to rent out his terrestrial data centers because he can't make use of them.
We probably don't need to pollute space with dead GPU's.
