Google is sending its AI infrastructure somewhere data centers have never seriously operated before: orbit. On October 1, the company's first experimental satellite for Project Suncatcher is scheduled to lift off aboard SpaceX's Transporter-18 rideshare mission from California — a refrigerator-sized spacecraft carrying four Google TPUs that will run Gemini workloads hundreds of miles above Earth.

The mission asks one question: can AI hardware actually survive in space? Google has already stress-tested the chips on the ground — its Trillium TPUs were blasted with proton beams at UC Davis and survived radiation doses greater than a five-year space mission would deliver, and vibration tests showed they hold up under launch forces that can reach 50 to 100 times gravity. The orbital flight will collect the data no lab can replicate. The biggest engineering unknown is cooling: with no air in space, Google's heat-pipe-and-radiator design can reportedly only run about 15 minutes before the TPUs must shut down to let the radiators catch up.

This isn't about replacing Earth data centers anytime soon — experts caution the concept is years from commercial viability. It's about energy: satellites in low Earth orbit get up to eight times more solar power than ground systems, sidestepping the grids, land and water constraints strangling terrestrial AI buildouts. Google built the satellite with Planet Labs and plans two more launches in 2027 to test laser links between orbital satellites.

Sources