Google said on September 24 that its Project Suncatcher team is preparing an orbital test of Tensor Processing Units, the chips it uses for AI workloads. A prototype satellite developed with Planet is slated to fly on the upcoming Transporter-18 rideshare mission with SpaceX. Google described the flight as an early hardware experiment, not the launch of an operating AI data center. The mission is meant to collect data on how TPUs handle launch stress, radiation and temperature extremes. Google said engineers shook the satellite to mimic rocket vibrations and tested its Trillium TPUs with a proton beam at UC Davis while running AI workloads. The company reported encouraging results from those ground tests, but said some questions can only be answered in orbit. Cooling is another open problem: in a vacuum there is no airflow, so the team is testing heat pipes and radiators. Google also described future concepts in which clusters of satellites could communicate by laser, while stressing that scalable machine-learning compute in space remains a long-term research goal. Its September 24 account does not give a firm October 1 launch date or say that SpaceX co-developed the project. The test flight, if launched, will provide evidence about hardware survival; it will not itself show that orbital data centers are commercially viable.
Google plans orbital test of AI chips for Project Suncatcher
Google says a prototype satellite on a SpaceX rideshare will test whether its TPU chips can withstand launch, radiation and cooling challenges in orbit. This is an early hardware test, not an operating space data center.
Key points
- Google says a prototype satellite on a SpaceX rideshare will test whether its TPU chips can withstand launch, radiation and cooling challenges in orbit. This is an early hardware test, not an operating space data center.
Primary sources