Google has run an artificial intelligence model on a satellite in orbit for the first time, a step toward its ambition to build data centers in space. CEO Sundar Pichai said the first Project Suncatcher prototype responded to a command from Earth with a haiku, calling it a "hello world" milestone for the project.
The team sent a command to orbit, booted one of the satellite's Tensor Processing Units and asked the onboard Gemma model to write a haiku about being in space. The answer was downlinked as the satellite passed over Svalbard, Norway.

'Starlight feeds my core'
The poem read: "Blue marble below, / Data streams in silent grace, / Starlight feeds my core." Later tests included space-themed limericks and requests for telemetry on the satellite's payload systems.
Built with San Francisco-based Planet, the satellite launched Oct. 1 from Vandenberg Space Force Base in California on SpaceX's Transporter-18 rideshare mission. It carries four of Google's Trillium-generation TPUs.
Why put data centers in orbit
Google's long-term vision is a constellation of solar-powered, TPU-equipped satellites linked by laser communications that could supplement or eventually replace some Earth-based data centers. Pichai has pointed to energy as the main driver: the sun puts out about 100 trillion times humanity's total electricity production, and satellites in sun-synchronous orbits can draw solar power almost continuously.
The prototype is testing whether the chips can withstand launch, radiation and the challenge of shedding heat in a vacuum, and for now the TPUs run only in short bursts. Google plans to launch two more satellites with Planet in 2027 to test high-bandwidth laser links between spacecraft.
Google is not alone in the race. Startup Starcloud flew an Nvidia H100 GPU on its Starcloud-1 satellite in November 2025 and trained a small AI model in orbit.