Google’s experimental AI satellite sends its first model-generated response from orbit
Google says its experimental satellite has sent its first response generated by an AI model in orbit. The test is an early step in Project Suncatcher, an initiative exploring whether AI data centres could eventually operate in space.
The satellite’s Tensor Processing Units ran Google’s Gemma 4 model, which produced a haiku about space after receiving a prompt. Google also tried space-themed limericks and prompts using satellite telemetry. The company plans to test how the chips handle heat and radiation, with two interconnected satellites targeted for 2027.
In short
- Google’s satellite generated its first AI response in orbit.
- Gemma 4 ran on onboard Tensor Processing Units.
- Google aims to test linked satellites in 2027.
Explained simply
Project Suncatcher is Google’s research effort to explore whether computers used to run artificial intelligence could one day operate in space. Such systems need powerful chips, reliable communications and ways to cope with harsh conditions beyond Earth.
The test involved Tensor Processing Units, Google’s chips for AI tasks, running the company’s Gemma 4 model aboard a satellite. Google is also studying how these chips might withstand heat and radiation, and how satellites could work together to process information.
The idea matters because space based computing could offer new ways to handle AI workloads, but it would need to overcome technical and practical challenges. Google has said it is aiming to test two linked satellites in 2027.
AI Art Culture Space Technology
Read the full article at the source →
Originally published by The Verge as “Google’s satellite AI server transmits its first response to Earth”.