Google is preparing the first in-orbit test of its custom AI chips, advancing a long-term effort to host machine-learning infrastructure in space. The initiative, dubbed Project Suncatcher, will fly a prototype on SpaceX’s Transporter-18 rideshare to assess whether Google’s Tensor Processing Units can withstand launch forces, radiation and extreme thermal conditions. Early ground tests, including vibration trials and proton-beam exposure at UC Davis, suggest Google’s latest Trillium TPUs can tolerate radiation doses exceeding a five-year mission.
Engineers are also working on space-specific cooling, using heat pipes and radiators in vacuum, and plan to evaluate high-bandwidth laser links for tight coordination among clustered satellites. A two-satellite laser interconnect demo is slated for 2027. Google says abundant solar energy in low Earth orbit could eventually support scaled AI workloads by networking multiple constellations, though core engineering hurdles remain. The research, conducted with Earth-imaging company Planet, underscores Big Tech’s push to find new power and performance envelopes for AI, even as practical deployment, reliability and cost questions persist.
Inside Project Suncatcher: Google’s bold bid to run AI computing in orbit
- Trending
- Comments
- Latest
