Google reports first AI response from an orbital TPU

During a ground-station pass, Google sent the satellite its first prompt, asking the onboard Gemma 4 open model to write a haiku about being in space.[1] The poem began “Blue marble below” and ended “Starlight feeds my core.”[1] Google also says it tried space-themed limericks and telemetry prompts that asked the model to report technical status across the payload systems.[1] The milestone followed commissioning work that included powering up the satellite’s host and TPU trays and running early machine-learning tests.[1] Google described the result as its first AI response from a data-center-class TPU in orbit.[1]

The prototype satellite carries four Google Tensor Processing Units, but it can operate only one TPU at a time.[3] TechCrunch reported that the satellite can turn on one TPU for about 20 minutes at a time and generates up to one kilowatt of electricity.[3] These constraints matter when judging what the demonstration establishes: it shows an AI response from the system, but does not by itself establish sustained or large-scale computing in orbit.[1][2][3] Google says the satellite was operating as expected after the company confirmed contact with it.[2] TechCrunch reported that commissioning took about three weeks.[3]

Google says its next tests will assess how TPUs perform under thermal stress and radiation exposure in orbit.[1] The company also says the satellite’s in-orbit data collection will examine how its TPUs handle spaceflight stress, radiation and thermal extremes.[2] Project Suncatcher is research into whether space could eventually host scalable machine-learning infrastructure, rather than evidence that such infrastructure is already in place.[2] Google identifies a pair of interconnected satellites in 2027 as a future milestone.[1] The prototype, built with Planet, reached orbit on SpaceX’s Transporter-18 rideshare mission; Google says a peer-reviewed paper about the research is available in the journal Joule.[2]