Google reported on October 30, 2026, that its Project Suncatcher prototype satellite had generated and returned an artificial intelligence response using a Tensor Processing Unit in orbit. The experiment ran Google’s Gemma 4 model aboard the spacecraft and returned a haiku during a communications pass over Svalbard.
According to Google’s account of the orbital test, the team first commissioned the host spacecraft and computing payload, then began machine-learning experiments. Subsequent prompts produced limericks and responses concerning payload telemetry.
A Tensor Processing Unit, or TPU, is a processor designed for machine-learning computation. In this experiment, the onboard hardware ran an existing model in response to an uploaded prompt. The resulting text was transmitted back to Earth.
The spacecraft reached orbit on October 1 aboard SpaceX’s Transporter-18 mission. Built with Planet, the initial orbital prototype carries computing hardware for experiments in the operating environment of low Earth orbit.
Testing Computing Hardware for Space
Google’s prelaunch technical overview describes ground tests addressing launch vibration and radiation exposure. Engineers subjected the equipment to vibration along three axes. They also operated AI workloads during proton-beam testing at the University of California, Davis, Crocker Nuclear Laboratory.
Those tests examined the computing equipment under physical and radiation conditions relevant to spaceflight. The orbital mission adds measurements from the integrated spacecraft, including the behavior of the computing payload during actual operations.
Google plans further thermal-stress and radiation experiments with the prototype. Its next announced step is a pair of interconnected satellites in 2027. That planned experiment extends the program from operating a computing payload aboard one spacecraft to testing connections between spacecraft.
