It will be an impressive Kardashev II tech demo, but there are so many benefits to building sustainable AI infra on earth that outweigh the trade offs.
Power gen and delivery
Waste heat reuse
Stream of circular economy feedstock to catalyse the capital for processing ewaste
Phil says this is stupid because radiators are a "solved problem" but then stops short of extrapolating what that means.
The radiator surface area they specify is about the right size for the math to check out. You can make pretty much any physics "work" if you scale the right values on paper. The feasibility isn't about physics, its about economics.
For a moderately sized 100MW "data center" you'd need a fleet of close to 700 satellites to match its earthly alternative. That's an ISS's worth (the most expensive single structure ever built by humanity) of radiators, 700 times over, for a single "data center".
To put that into perspective that's about 73,000 m² or roughly 14 football fields worth of surface area
The costs to build such a fleet is insane compared to the the earthly alternatives we have yet to exploit and it assumes a lot of things that might not continue to hold true - like wattage per op staying constant.
It’s like someone looking at room-sized computers in the 50s and extrapolating that we are going to have a land crisis and starts imagining large floating sea platforms to host our computing future