Google has reached a milestone in Project Suncatcher by successfully executing an AI workload on a chip in low Earth orbit.
Researchers prompted an open-source model to generate a haiku in space, with the satellite beaming back the output during a ground-station pass in Norway.
As part of the Suncatcher project, the company managed to get its first artificial intelligence response from a special processor in low Earth orbit. During a communication with a ground station, the researchers asked the company's open-source model, Gemma 4, to write a short poem about its time in space. As the satellite passed over the Svalbard archipelago in Norway, it sent the poem back to Earth.
The first prototype satellite for this project was launched on October 1, in partnership with the company Planet and the SpaceX launch service. Google says that confirming that its processors can perform tasks reliably in space paves the way for the operation of two interconnected satellites in 2027.
The difficulties of space
The company notes that there are significant technical challenges to the project. Cosmic radiation and solar storms are hazards, while additional difficulties include cooling the systems without the natural airflow that exists on Earth and communication between the satellites.
↳
READ ALSO
HACKS.GR / RELATED
AI Unlocked
OpenAI fixed a bug in ChatGPT that could expose private conversations via a malicious link
↗
The initial tests are mainly to see what works and identify weak points for future missions. Google's plans include fleets of more than 80 satellites that would fly close to each other and communicate with each other, and there have reportedly been discussions about creating a giant satellite the size of a football field.
Competition and the future
Google is not the only company looking at using space for computing power. In January 2026, SpaceX applied for approval from the US government to launch up to a million satellites to create a network in space. Meanwhile, Jeff Bezos has predicted the creation of huge data centers in space within the next two decades.
>Stay connected with Hacks.gr
Set us as a preferred source on Google and find us on social media.
Preferred source
↗
end of article
Top comments (1)
On-orbit inference is compelling because it changes which data needs to travel back to Earth. The operational question is how models are versioned, monitored, and rolled back under bandwidth and power constraints; those controls will determine whether the demo becomes dependable infrastructure.