Google's Project Suncatcher prototype satellite, built with Planet, ran the Gemma 4 open AI model on an onboard Tensor Processing Unit and returned its first response from orbit, Google announced on 30 October 2026.
Asked for a haiku about being in space, it replied 'Blue marble below / Data streams in silent grace / Starlight feeds my core', received as the satellite passed over Svalbard, Norway; Sundar Pichai called it a 'hello world' moment.
The satellite launched on 1 October 2026 on SpaceX's Transporter-18 rideshare; Google plans two interconnected satellites in 2027 to test optical links between spacecraft.
Google announces Project Suncatcher
Prototype launched on SpaceX Transporter-18
First AI response from orbit announced
Two interconnected satellites planned to test optical links
An application-specific integrated circuit (ASIC) Google built for neural-network machine learning. It handles huge volumes of low-precision arithmetic, giving more operations per joule than general-purpose GPUs — which matters where power is scarce, as in orbit.
Simple Analogy: A GPU is a multi-tool; a TPU is a blade made for one cut.
Runs Project Suncatcher; designs the TPU chips and the Gemma models
Earth-imaging satellite company that built the Suncatcher prototype with Google
Svalbard is a Norwegian Arctic archipelago between mainland Norway and the North Pole. Its high latitude lets ground stations there see polar-orbiting satellites on almost every pass, which is why the first downlink came through it.
GS Paper III > Science and Technology — AI, space technology
General Awareness > Science and technology in news
Google's ASIC for neural-network machine learning
Google's family of open AI models
Laser link joining spacecraft into a space-based network