Google has reached an early milestone in its plan to explore space-based artificial intelligence infrastructure, after an experimental satellite successfully ran an AI model in orbit and transmitted its first response to Earth.
The satellite, equipped with Google’s Tensor Processing Units (TPUs), generated a short haiku about life in space after receiving a prompt. The test marks the first reported demonstration of the satellite’s onboard AI computing capabilities under the company’s Project Suncatcher initiative.
Google announced the achievement in a blog post on October 30, revealing that the satellite’s processors had been successfully powered on and used to run its Gemma 4 AI model.
AI model generates poetry from orbit
To demonstrate that the onboard system was working, Google asked Gemma 4 to write a haiku about being in space.
The model returned:
Blue marble below
Data streams in silent grace
Starlight feeds my core.
Google said it also tested the satellite with other space-related writing prompts, including limericks, as well as requests involving telemetry data.
Although generating a poem may appear modest, the demonstration was intended to show that the satellite’s AI hardware could successfully execute tasks while operating in orbit.
Project Suncatcher aims to move AI computing into space
The experiment is part of Project Suncatcher, Google's research initiative exploring the possibility of building AI data centres in space.
The concept involves placing computing infrastructure in orbit, where interconnected satellites could eventually process AI workloads beyond the limitations of conventional ground-based facilities.
Google launched the experimental satellite earlier in October 2026. Its successful activation and first AI response represent an initial step towards testing whether specialised processors can operate reliably in the space environment.
Radiation and temperature tests come next
Google said it will now conduct further experiments to assess how its TPUs perform under the demanding conditions of orbit.
The next phase will focus on thermal stress and radiation exposure, two important factors that could affect the reliability of computing hardware in space.
The company is also working towards its next major milestone: deploying two interconnected satellites in 2027.
That planned test will help Google investigate how separate orbital systems could work together to support computing tasks.
For now, Project Suncatcher remains experimental. The successful haiku demonstrates that Google's satellite can run an AI model in orbit, but further testing will be needed to establish whether space-based AI infrastructure can operate reliably at a larger scale.
Source: The Verge