Google has demonstrated an AI model running on a satellite in orbit and sending its response back to Earth. The test, announced on October 30, asked the system to write a short haiku about being in space. The company describes it as the first orbital AI response produced by a TPU of the kind used in data centers.
The experiment was part of Project Suncatcher, Google’s long-term research into placing AI computing infrastructure in space. During a pass within range of a ground station in Svalbard, Norway, the team sent a prompt to Gemma 4, installed aboard the satellite. The model generated the poem onboard, and the satellite transmitted it to Earth. The team also tried further poems and prompts about the equipment’s technical status.
The prototype was built with Planet and launched on October 1, 2026, as part of SpaceX’s Transporter-18 mission. Its purpose is to gather data on how processors perform during flight and in the space environment. Google introduced Project Suncatcher in November 2025 as an investigation into networks of solar-powered satellites equipped with AI processors.
Operating computers in space presents engineering challenges. Without air, conventional fans cannot carry heat away from components, so heat must be conducted to radiators and released as thermal radiation. Google is testing heat-transfer pipes and radiators, while also studying how chips withstand space radiation and temperature conditions.
Google’s next planned step is to test a laser connection between two satellites in 2027. The company aims to determine whether separate machines in orbit can work together on AI tasks. The haiku test demonstrates one capability, but turning the idea into a large-scale service will require further testing of reliability, communications and economic feasibility.
Sources: Pplware.