Google Launches AI Chip into Space with Project Suncatcher

Google is set to launch its first AI chip into space next week as part of Project Suncatcher, a research effort aimed at exploring whether space can host large-scale AI computing infrastructure. The mission, scheduled for liftoff on SpaceX's Transporter-18 rideshare launch, will see a prototype satellite carrying Google's Tensor Processing Units (TPUs) into low Earth orbit to test their reliability in extreme conditions.

The TPU chips have already been tested on Earth by simulating the radiation exposure they would face over five years in space. The previous experiment saw the chips show no hard failures up to a maximum dose of 15 krad(Si), with the most radiation-sensitive component identified as High Bandwidth Memory. To address heat dissipation issues, Google is testing a new design that combines heat pipes with radiators, which has proven successful in ground-based vacuum chamber testing.

Project Suncatcher is part of Google's long-term research to overcome the power demands of training and running large AI models, which are increasingly straining terrestrial power grids. The company believes solar panels in low Earth orbit can generate up to eight times more power than equivalent panels on the ground due to near-constant sunlight.

Google is not alone in pursuing orbital AI computing, with Nvidia-backed startup Starcloud and SpaceX also working towards their own data center ambitions in space. However, experts caution that space-based AI computing remains years from commercial viability due to high launch costs, engineering constraints, and satellite production bottlenecks.