Physics Makes Orbital Data Centers Impractical
Get the Tech newsletter
Daily tech — startups, AI labs, chips, the launches that shape the next decade. Free.
- AI data centers are expanding so rapidly that by 2028, AI servers alone may consume energy equivalent to 22 percent of US households, while evaporation-cooled facilities drain millions of gallons of water daily.
- Space is not intrinsically cold — it is a vacuum with no temperature — and radiation is the only heat-transfer mechanism, making cooling far less efficient than air-based conduction, the analysis explains.
- Volume grows faster than surface area under cubic scaling, so a space computer scaled up eightfold in volume has only four times the radiating area — meaning a Walmart-size orbital data center would melt.
- A 1-megawatt space data center would need at least 980 square meters of radiator panels, with ammonia piping systems like the International Space Station's adding mass and launch costs; existing AI data centers on Earth use 100 to 1,000 megawatts.
- Google's Project Suncatcher and SpaceX are pursuing swarm architectures of many small satellites, with SpaceX having requested FCC permission to launch a million AI satellites.
- Low Earth orbit already holds roughly 10,000 active satellites and 10,000 metric tons of space junk, and adding 100 times more satellites would amplify collision and Kessler cascade risks.
Why it matters: The article lays out a fundamental physics problem — radiation cooling scales poorly with size — which means the only viable orbital architecture is swarm-based, and SpaceX alone has requested FCC permission to add a million AI satellites to an orbit already holding 10,000 active satellites and 10,000 metric tons of debris.
Ask SkimNews
