Google Sends AI Chips into Space with Project Suncatcher

Google's 'Project Suncatcher' aims to harness space-based computing power for A.I. workloads. As part of this project, Google will soon launch its first satellite carrying custom-built Trillium A.I. chips aboard SpaceX's Transporter-18 rideshare mission.

The satellite, developed in collaboration with Earth-imaging company Planet, features the Tensor Processing Units (TPUs) that power models like Gemini on the ground. In space, these TPUs must withstand extreme conditions: intense vibrations and acceleration during launch, radiation damage from solar storms and cosmic rays, and heat without air for cooling.

Google claims its TPU chips can survive radiation doses exceeding a five-year mission duration, as tested at UC Davis's Crocker Nuclear Laboratory. The next step in Project Suncatcher is planned for 2027, when two satellites will test linking TPU clusters via high-bandwidth lasers.

Space-based computing offers an attractive solution to A.I.'s power appetite: satellites can generate up to eight times more solar power than Earth-based installations due to the absence of clouds, night, and atmosphere. However, computing in space is currently expensive, with launch costs estimated at $200 per kilogram.