120kW worth of B200s run close to $5M. The chips are easily less than half the TCO of a terrestrial inference cluster.
SpaceX already builds and launches sats about a quarter of the mass of their tentative 120kW AI1 design for ~$1.4M.
Roughly half that cost is launch and expected to dramatically drop with Starship; the other half is expected to come down more modestly with economies of scale and maturing industrial processes.
And radiators and coolant pumps are cheaper than phased array antennas...
"No one who understands this stuff doubts that it's possible. They just doubt the economics."
Good, now do the math and stop pretending that doubting the economics is a more rational position than expecting it to work out.









