Google Tests AI Data Centers in Space with Satellite Launch

Google has taken a bold step toward space-based AI computing with the launch of a prototype satellite carrying its custom AI chips. The refrigerator-sized satellite, named MVP, lifted off from Vandenberg Space Force Base on Thursday aboard a SpaceX Falcon 9 rocket. This mission, part of Project Suncatcher, is a collaboration with San Francisco-based Planet Labs. The goal is to test whether Google's tensor processing units (TPUs) can operate reliably in orbit despite extreme radiation and heat.

Google has been preparing for this moment since February 2025, when it began radiation testing its chips at the Crocker Nuclear Laboratory in Davis. The company discovered that the original chip configuration provided more shielding than anticipated, leading to additional testing. To manage heat in the vacuum of space, Google developed a cooling system that transfers heat through conductive layers to a radiator panel. This allows the chips to run for about 15 minutes before needing to cool down.

The current launch is just the beginning. Google plans to send another pair of satellites next year and envisions constellations of more than 80 satellites flying in formation. The company is also exploring the possibility of a massive satellite roughly the length of a soccer field. However, significant challenges remain, including high-speed communication between satellites, managing heat, and dealing with space debris and solar storms.

Researchers have raised questions about the environmental benefits of solar-powered computing in space, considering the emissions from launching and deorbiting satellites. Google's radiation testing suggests its TPUs could survive a satellite's expected five-year lifespan, but the reliability may not be sufficient for large-scale AI training runs. Despite these hurdles, the launch of MVP marks the first step in determining the feasibility of orbital AI data centers.