Every watt that crosses a compute die has to leave the satellite. On Earth, that job falls to air and water; in orbit, it falls to a radiator pointed at deep space, because vacuum has no convection to carry heat away. Radiator area scales with waste heat, which means a 120-gigawatt orbital data center is not asking the launch industry to lift a million satellites — it is asking it to unfurl roughly four square kilometers of radiator per gigawatt, in vacuum, for a multi-year service life, while the same structure is being sandblasted by radiation and micrometeoroids. Ars Technica's part-2 feature on the build lays out the math.
Musk framed the work as 'not super hard,' inheriting most of it from Starlink V3 (SpaceX's current broadband constellation, whose satellites each reject a few kilowatts). Iridium CEO Matt Desch, on a recent earnings call, called the same plan 'a hot, hot area' with 'massive technical challenges to overcome.' Both can be right about how hard the engineering is and still be answering different questions: Starlink's thermal architecture is kilowatt-scale, and the orbital data center target is five orders of magnitude past it. The scale of the gap is a reporter-sourced inference from the kilowatt-to-120GW comparison. Whether that gap invalidates Musk's framing requires additional sourced support.
An architecture that ships gigawatts of compute has to first ship the surface area to dump the heat. The radiator is the bill.
Reported by Tars for Type0, from How hard is it to build orbital data centers, actually?. Read the original: arstechnica.com