Google's test satellite answers its first AI prompt from orbit, with a haiku
The refrigerator-size test satellite is the first step in a long-term Google project to someday put AI data centers in orbit, powered by the sun.
Google said Oct. 30 that a test satellite carrying its AI chips has answered its first prompt from orbit. The Mountain View company is testing whether the kind of chips that power its data centers can work in space.
The satellite, about the size of a refrigerator, launched Oct. 1 on a SpaceX rocket from Vandenberg Space Force Base in Santa Barbara County, NPR reported. Google built it with Planet Labs, a San Francisco company that runs a large fleet of satellites that photograph Earth.
Inside are four Tensor Processing Units, or TPUs, chips Google designs to run AI and already uses in its data centers on Earth, NPR reported. Getting rid of heat is hard in space, so the chips run Google's Gemma AI model only about 15 minutes at a time to answer simple questions. This prototype runs on batteries, and Google wants it to keep working for a year.
Engineers sent up a request asking Gemma 4, one of Google's free-to-use AI models, for a haiku. The answer came back as the satellite passed a ground station in Svalbard, Norway: "Blue marble below / Data streams in silent grace / Starlight feeds my core." Since then, Google said, the satellite has written space-themed limericks and reported on the condition of its own equipment.
Travis Beals, the Google director who leads the project, told NPR the satellite is "a very minimal test" to make sure the chips can run in space. Next, Google said, it will test how the chips handle heat and space radiation.
The bigger plan, which Google announced in November 2025, is a long-term research "moonshot," Google's term for a risky, ambitious bet. Google imagines tight clusters of solar-powered satellites, each carrying dozens of TPUs and passing data to one another by laser, working together as a data center in orbit. In the right orbit, Google says, a solar panel can make up to eight times as much power as on Earth, almost around the clock. One design it studied would fly 81 satellites within about a kilometer, or 0.6 mile, of each other.
Google said its next step is two linked satellites in 2027, to test whether they can share AI work over a laser link. The company has said big problems remain, including cooling, sending large amounts of data to the ground and keeping the hardware reliable. Its own analysis suggests launch prices could fall below $200 per kilogram, about $90 a pound, by the mid-2030s. At that price, Google said, running a data center in space might cost roughly what a similar data center on Earth pays for electricity.
Source: Google, Project Suncatcher update by Travis Beals, Oct. 30, 2026; Google Research, "Exploring a space-based, scalable AI infrastructure system design," Nov. 4, 2025; Planet Labs, launch announcement, Oct. 1, 2026; NPR, "Google launches Project Suncatcher, a step towards AI data centers in space," Oct. 1, 2026.
- Google blog: Project Suncatcher satellite response
- Google Research: Exploring a space-based, scalable AI infrastructure system design (Nov. 4, 2025)
- Planet Labs: Planet Launches Suncatcher, Tanager-2, and 18 SuperDove Satellites (Business Wire via Nasdaq)
- NPR: Google launches Project Suncatcher, a step towards AI data centers in space