Auch wenn Technologieunternehmen voraussichtlich bis zum Ende des Jahrzehnts weltweit mehr als 5 Billionen US-Dollar für erdbasierte Rechenzentren ausgeben werden, argumentiert Elon Musk, dass die Zukunft der KI-Rechenleistung im Weltraum liegt – angetrieben durch Solarenergie – und dass die Wirtschaftlichkeit und Technik, die dafür sorgen, dass sie funktioniert, innerhalb weniger Jahre in Einklang gebracht werden könnten.
In den letzten drei Wochen hat SpaceX bei der Federal Communications Commission Pläne für ein Rechenzentrumsnetzwerk mit einer Million Satelliten eingereicht. Musk hat außerdem angekündigt, dass er plant, sein KI-Startup xAI mit SpaceX zu fusionieren, um orbitale Datenzentren zu betreiben. Und letzte Woche teilte er den xAI-Mitarbeitern bei einem Plenumstreffen mit, dass das Unternehmen letztendlich eine Fabrik auf dem Mond brauche, um KI-Satelliten zu bauen – zusammen mit einem riesigen Katapult, um sie ins All zu bringen.
„Der kostengünstigste Ort, um KI einzusetzen, wird der Weltraum sein, und das wird innerhalb von zwei, vielleicht spätestens drei Jahren der Fall sein“, sagte Musk auf dem Treffen des Weltwirtschaftsforums in Davos im Januar dieses Jahres.
Doch während Musk und einige andere Bullen argumentieren, dass weltraumgestützte KI innerhalb weniger Jahre kosteneffektiv werden könnte, sagen viele Experten, dass alles, was auch nur annähernd einen sinnvollen Umfang erreicht, noch Jahrzehnte entfernt ist – insbesondere, da der Großteil der KI-Investitionen weiterhin in die terrestrische Infrastruktur fließt. Dazu gehört auch Musks eigener Colossus-Supercomputer in Memphis, der nach Schätzungen von Analysten mehrere zehn Milliarden Dollar kosten wird.
Mehr lesen: https://fortune.com/2026/02/19/ai-data-centers-in-space-elon-musk-power-problems/
https://fortune.com/2026/02/19/ai-data-centers-in-space-elon-musk-power-problems/
14 Kommentare
The only reason to put a data center in space is to take investors‘ money
It makes no sense from a physics point of view due to cooling being one of the biggest datacenter / compute issues.
Cooling AND data transfer are both significantly more difficult IN SPACE.
Microsoft has already done experiments ON EARTH with sealed datacenter units run under water for cooling but fed with power and high speed data from land.
They proved the containers worked, completely sealed (what you would need for space even).
They didn’t end up going this way however as the project was deemed logistically and economically impractical for widespread use.
Considering that space would be WORSE in terms of deployment, data and cooling there should be no logical point to this.
Edit: radiation may also present an issue for compute in space as well.
Elon really is so ignorant that I don’t understand how he became a billionaire. I don’t understand how you could hear this concept and think this is a good idea.
Kyle released a video on this topic today: Space data centers are dumb -> https://youtu.be/-w6G7VEwNq0?si=Q-QNt8z6NI2zFhlP
Lack of power and cooling is the nail and since he owns a space launch system, that is his hammer.
But it is a horrible idea when you look at the power and cooling requirements, and we haven’t even gotten to the issue of shielding fragile electronics from radiation.
The great thing about datacenters is that they don’t need to be close to population centers or stuff like that so you can put them in the cheapest places.
Space is the most expensive place.
…and we all know how „reliable“ Musk is with his plans.
Fucking use your money to help other causes. Moron.
Why are we still paying attention to this pedophile?
A post I made on LinkedIn a couple of weeks ago:
I’ve been hearing more and more talk about „data centers in space.“ But there’s a physics problem that isn’t being discussed.
On Earth, data center cooling is straightforward. Fans blow air across heatsinks. Water circulates through cooling towers. Convection moves heat from hot components into cooler fluids that carry it away. Cheap, proven, and scalable.
In space, none of this works.
Space is a vacuum. No air, no water, no medium to absorb heat. The only option is thermal radiation: slowly emitting infrared photons from massive radiator panels. The entire ISS uses this approach to handle about 70 kW of waste heat. A single modern AI rack generates 40-120 kW. That figure will probably be more like 200kW in about two years.
Everyone thinks „space is cold“ because of movies and simple explanations. In reality, space „cold“ isn’t the same type of „cold“ we have on earth. The background temperature of space is ~2.7K, but temperature without a transfer medium is meaningless. A vacuum thermos keeps your coffee hot for hours using this same principle. Vacuum insulates. It does not cool. Heat in space needs to be radiated away, not actively disbursed like on earth. And that passive radiation is MUCH slower than active heat transfer.
For a modest 10 MW AI training facility (small by terrestrial standards) you’d need 30,000+ square meters of radiator surface, hundreds of tonnes of cooling infrastructure, and billions of dollars. The total system cost is 50-200x what the same facility costs on the ground. The cost of cooling would be more than the cost of the GPUs, the solar array, and the launch costs combined.
Elon Musk knows this. SpaceX engineers absolutely know this. Yet the narrative around space-based compute keeps growing, and it’s not hard to see why. SpaceX is heading toward an IPO, and every ambitious-sounding use case (Mars colonization, orbital data centers, point-to-point travel) inflates the total addressable market story that drives valuation.
Musk has a pattern of announcing physics-defying timelines and capabilities to move markets. The Cybertruck was supposed to be bulletproof and cost $40K. Full self-driving was supposed to be solved by 2020. The Boring Company was supposed to revolutionize transit.
Space data centers for AI are the same playbook. The engineering challenges are real and unsolved. The promises serve a financial purpose, not a technical one. Musk and the techbros that regurgitate his talking points only have one goal: artificially inflate the SpaceX IPO, especially since xAI has been absorbed by it.
Next time someone pitches you on orbital compute, ask them one question: how do you disperse the heat? If the answer doesn’t include specific radiator mass budgets, you’re not hearing engineering. You’re hearing a pre-IPO narrative based on lies.
I’ll post later about how to actually solve the AI problem, and why it is incredibly dangerous for people to continue to platform the lies of billionaire capitalists.
In space, no one can hear you scheme.
Even Stevie Wonder could see that this idea isn’t viable plus it’s what the 50th time Elon promises something?
The required surface area of solar panels and radiators practically guarantee Kessler Syndrome at any worthwhile scale.
I suspect that Elon wants to cause it to have an excuse for how much SpaceX will never deliver their part of the lunar mission.
SpaceX is the only company on this planet at building satellites at scale. No one else know that much about how much things cost or will cost.
Of corse you need to keep in mind that Elon Musk is quite often very optimistic about timelines to put it mildly