sean goedecke

A technical analysis debunking the claim that heat dissipation makes AI datacenters in space impossible. While vacuum eliminates convection and conduction, radiative cooling via large radiator panels is physically viable. Estimates suggest ~250,000 square metres of radiator area would be needed for a 100MW datacenter — roughly 250x the ISS radiator size, requiring 100–500 Starship launches. The conclusion: space datacenters are wildly impractical but not impossible, and cooling is actually a relatively minor challenge compared to total mass requirements like solar panels and hardware.

AI datacenters in space do not have a cooling problem

I think the author ignores that “impossible” is often a shorthand for “doesn’t make sense” and not necessarily a comment on wether something will work at all. On the other hand, we should, of course, be explicit and accurate with what we want to express in a public discourse.
In general I still don’t think orbital data centers are feasible for multiple reason. First and foremost, that servicing and maintaining them is prohibitively expensive for the foreseeable future.

And what about rays and radiation, solar wind etc. … far more issues than just heat