AI1

Four days before the largest IPO in history, SpaceX put a concrete product behind the "data centers in space" filing it had quietly made: AI1, a satellite built to be a node in an orbital GPU grid. The spec sheet is enormous — 150 kilowatts of computing power, a solar wingspan wider than a Boeing 747, and a co-announced thousand-acre factory in Bastrop, Texas meant to stamp out roughly 6,700 of these a year, from molten silicon to finished satellite. The target: a gigawatt of compute in orbit by late 2027, on hardware that has not yet flown.

The same pre-IPO filing that markets the technology describes it as "unproven" and warns it "may not achieve commercial viability." — SpaceX's S-1

The pitch and the legal filing tell different stories. The investor update says the physics needs "no magic"; the same pre-IPO prospectus, in its risk factors, calls the orbital-compute technology "unproven" and warns it "may not achieve commercial viability." The hardest number to defend is heat. A computer in vacuum can only shed waste heat by radiating it, and AI1's radiator is claimed to do so at roughly eight times the density of the International Space Station's — humanity's most advanced existing space-cooling system — across a panel a fraction of the size.

Underneath sits an inversion worth holding onto. The case for orbit is free sunlight and a cold sky to dump heat into, no grid queue, no water. But to save the cheap resource — electricity, around $75,000 a year per rack on the ground — you spend the expensive ones: launch mass, irreplaceable hardware, and radiator area, each of which a rival space firm's math says leaves orbital compute three to four times costlier per watt. The reveal is real, and so is the race it escalates; whether AI1 is a product or a prospectus illustration is the question the filing itself declines to answer.