Google Project Suncatcher's prototype satellite has produced its first AI output from orbit: a haiku about being in space, written by the Gemma 4 open model on a data center class TPU.
During a ground-station pass, Google's team uplinked the prompt to the onboard model. The satellite sent the poem back as it passed by Svalbard, Norway. Google calls the exchange its "Hello World" milestone and the first AI response from a data center class TPU in orbit. Follow-up prompts were space-themed limericks and requests for the technical status of the payload systems.
The Google Project Suncatcher satellite was built with Planet and rode SpaceX's Transporter-18 rideshare mission on October 1. Before the haiku, the team spent the last few weeks on a commissioning checklist: powering up the host and TPU trays and running early machine learning tests.
Next, Google says, a series of Project Suncatcher tests will evaluate how the TPUs perform under thermal stress and radiation exposure in orbit. In an earlier proton-beam test of its Trillium (v6e) TPU at 67 MeV, the high-bandwidth memory was the most sensitive part and began showing irregularities at a cumulative 2 krad(Si). Google puts the expected shielded dose over a five-year mission at 750 rad(Si), so the first irregularities appeared at roughly 2.7 times that dose. No hard failures traced to total ionizing dose showed up through 15 krad(Si) on a single chip.
Google says "some things can only be tested in space." Google frames the long-term bet as solar-powered satellites carrying TPUs and linked by free-space optical connections, likely flying in a sun-synchronous orbit with near-constant sunlight. In the right orbit, Google says, a panel can be up to 8 times as productive as on Earth.
Google says verifying that its TPUs can reliably run workloads in space paves the way for its next Project Suncatcher milestone, twin interconnected satellites in 2027. In November 2025, Google described a two-satellite learning mission with Planet, slated by early 2027, to validate optical inter-satellite links for distributed machine learning tasks.
