Google Tests Solar-Powered AI Chips in Low Earth Orbit

Google is embarking on an ambitious project to test whether machine learning hardware can run on solar power in low Earth orbit. The first flight of Project Suncatcher, a long-term research effort, will send four Tensor Processing Units (TPUs) into space next week aboard a SpaceX Falcon 9 rocket. The satellite, named MVP, is designed to measure the effects of launch vibration, radiation, and vacuum heat on Google's chips.

The project aims to create clustered, solar-powered satellites that can communicate with each other over lasers, using the Sun as their primary source of energy. Currently, AI usage consumes a significant amount of electricity and land, but the Sun is the largest energy source in the solar system. The first flight is a small-scale iteration of this concept, using 1 kilowatt of solar power to perform brief bursts of computation.

Planet Labs built the satellite's bus, which was modified to accommodate Google's chips. The company had already invested in Planet and asked to fly sooner than planned for two custom satellites scheduled for 2027. Ground crews physically shook the satellite to test its ability to withstand launch vibration, and it withstood more ionizing radiation than a typical five-year mission.