Google Runs First AI Model on a Satellite, Receives Response From Orbit

Google has successfully run an artificial intelligence model on a satellite in orbit and received the generated output back on Earth, a milestone the company says marks an early step toward its long-term ambition of placing AI computing infrastructure beyond the planet.

The test is part of Project Suncatcher, Google's research effort exploring whether AI systems can eventually operate in space rather than depending entirely on terrestrial data centers. Google announced the successful demonstration on October 30, confirming that its experimental satellite had processed an AI request and transmitted the result back to the ground.

The satellite launched on October 1 aboard a SpaceX rocket and has spent the past several weeks undergoing testing. In the key experiment, Google sent a request from Earth asking the onboard Gemma 4 model to compose a short poem about being in space. The model processed the request using a Tensor Processing Unit, Google's specialized chip designed for AI workloads, and delivered a three-line verse that read: "Blue marble below / Data streams in silent grace / Starlight feeds my core."

The response was transmitted back to Earth as the satellite passed over Svalbard, Norway. Google described the exchange as its "Hello World" milestone, a term commonly used in computing to denote an initial successful test. According to the company, it was the first complete AI response generated by a data center-class TPU operating on an orbiting satellite.

Google said it has since carried out additional tests involving further poem generation and requests for information about the satellite's operating conditions. The company has not disclosed a timeline for deploying larger-scale AI infrastructure in space, and the project remains at the research stage.