What If We Moved Every Data Center to Space? Space Is Cold. That's the Problem
Alternaut
0:00 / 0:00
What If We Moved Every Data Center to Space? Space Is Cold. That's the Problem
20 просмотров · 3 дн. назад
Alternaut
6 подписчиков
20 просмотров · 3 дн. назад
A common pitch says space is cold, so cooling a data center up there should be free. This video checks that with one law of physics. In a vacuum there is no air or water to carry heat away, so nearly every watt a computer uses has to leave as infrared light, and the size of the radiator that does it depends on its temperature. We work from one square metre to a rack, a gigawatt and all the world's data centers, compare the numbers with the radiators of the International Space Station and with the design figures SpaceX has shown, and look at what Google's prototype and Starcloud's satellite have actually flown, as of October 2026.
CHAPTERS
0:00 The pitch: put every data center in space
1:13 Cold is a temperature, not a cooling rate
1:46 Heat can only leave as light
2:07 One square metre of radiator
3:07 A gigawatt of heat
3:21 The biggest surface isn't the radiator
4:09 The knob: how hot the radiator runs
4:54 The price of running hot
5:21 SpaceX's claim against the station
6:01 What has actually flown
6:30 Every data center on Earth
7:52 The cost and the launch price
8:20 Where it might work
8:46 What this video leaves out
9:05 So is cooling in space free?
9:35 Which arrives first?
How to read the claims (the measure behind every superlative and every "about"):
All radiator and solar panel areas are our own model for one gigawatt of waste heat, measured as flat surface area: radiator edge-on to the Sun, Earth's infrared included, emissivity 0.9, both faces radiating, low orbit. That gives about 0.7 square kilometres at 350 kelvin and about 1.4 at 300 kelvin, against about 3.4 to 4 square kilometres of solar panels (at 250 to 290 watts per square metre of panel), which is why we call the panels the biggest surface. The radiator area falls steeply as it runs hotter because the glow rises with the fourth power of the temperature in kelvin.
The world's 2024 data centers, about 47 gigawatts on average by the IEA's count, would need a radiator roughly the size of Manhattan's land area at 300 kelvin. That is an upper bound, because the count includes the ground's own cooling overhead.
Company figures (SpaceX, Google, Starcloud, Nvidia) are the companies' own claims or assumptions, not measurements, with the dates shown on screen. Every satellite application mentioned is pending as of this recording.
"Two to four times" is the spread of four outside cost analyses (BCG, Wood Mackenzie, Andrew McCalip, Thunder Said Energy), each seen through summaries; one other report, Bain's, puts a small system lower, at about one and a half times. They agree that orbit costs more today, not on by how much.
Radiation, debris and repairs are not covered.
The narration is a synthetic (AI) voice, never a person. The animation and sound effects were generated by us. Music: "Evening Glow" by Lee Rosevere, licensed under Creative Commons Attribution 4.0 (https://creativecommons.org/licenses/..., edited for this video (looped and level-adjusted), source: https://freemusicarchive.org/music/le.... Satellites, radiators and city areas are our own schematic drawings, not exact outlines; no map data is used.
Sources (full list in the first reply under the pinned comment):
NIST, fundamental constants (Stefan-Boltzmann): https://physics.nist.gov/cuu/pdf/wall...
NASA, ISS Active Thermal Control System: https://www.nasa.gov/wp-content/uploa...
Google, Towards a future space-based, highly scalable AI infrastructure system design: https://arxiv.org/abs/2511.19468
Starcloud white paper: https://starcloudinc.github.io/wp.pdf
FCC, SpaceX application (DA 26-113): https://docs.fcc.gov/public/attachmen...
IEA, Energy and AI (2025): https://www.iea.org/reports/energy-an...
BCG, space-based data centers cost outlook: https://www.bcg.com/publications/2026...
Andrew McCalip, space data centers: https://andrewmccalip.com/space-datac...
van Berkel, cost and network limits of space-based AI compute: https://arxiv.org/abs/2607.14172
US Census Bureau QuickFacts, Manhattan: https://www.census.gov/quickfacts/new...