AIBRIEF
Built with Planet and flown on SpaceX's Transporter 18 rideshare, the Project Suncatcher prototype will measure how a commercial AI chip handles radiation and thermal extremes in low Earth orbit.
Google, Planet, and SpaceX have a commercial AI chip in orbit. The prototype's job is not to run machine learning in space. It will log how the chip handles radiation, thermal cycling, and cosmic-ray strikes over the mission.
Project Suncatcher launched aboard SpaceX's Transporter-18 rideshare and confirmed contact with its operators at announcement. Google built the satellite with Planet, the Earth-imaging company, and frames the mission as the first step in a long-term research moonshot to test whether machine learning compute can run in space.
The instrument on board is a Tensor Processing Unit (TPU), Google's custom AI accelerator. The prototype's deliverable is a log, not a model output: bit-flip counts from single-event upsets, thermal cycling data, and radiation dose measurements. A peer-reviewed paper in Joule documents the engineering behind the test. Independent technical coverage flags thermal management, not radiation tolerance, as the binding constraint for any future scale-up.
What remains unknown: the in-orbit experiment readout cadence, how long a commercial TPU can survive radiation, and whether the 100,000-TPU facilities the company envisions are physically or economically feasible. The first answer comes when the prototype stops working.