Google tests first AI exchange from an orbiting satellite
The Project Suncatcher prototype sent its first AI response from orbit after a launch on October 1. Google aims to verify if TPUs can run AI workloads in space.
On October 30, Google announced that its satellite prototype Project Suncatcher had produced its first artificial intelligence response in orbit, powered by a datacenter-class TPU. The test, conducted after a launch on October 1, aims to assess whether computing infrastructures can operate sustainably in space. For now, this is a research milestone, not a data center floating above our heads. But the signal is clear: Google is exploring a new frontier for AI computing.
A prototype designed to test the resilience of AI computing in orbit
Project Suncatcher is a Google research program studying the feasibility of a solar-powered machine learning infrastructure in orbit. The satellite was designed with Planet and launched aboard SpaceX’s Transporter-18 rideshare mission. According to the note published by Travis Beals, Senior Director of Paradigms of Intelligence, Google has completed the commissioning checklist, including powering on the host, TPU drawers, and initial machine learning tests.
The most visible demonstration involved sending a request from a ground station to the onboard open-source Gemma 4 model. The prompt asked to generate a haiku about life in space, then the response was sent back to Earth as the satellite passed over Svalbard, Norway. Google then tested limericks on the space theme and queries on embedded system telemetry. The core issue remains more serious than poetry: measuring hardware behavior under thermal stress and radiation.
Why Google looks beyond low Earth orbit
The next phase of the project is intended to verify whether the TPUs can withstand these extreme conditions over time. If the tests are successful, Google will better estimate the reliability of AI computations performed outside Earth’s atmosphere. The company says it wants to use these results to move toward a planned 2027 stage, with two interconnected satellites, to study distributed computing in orbit.
In this context, Project Suncatcher does not announce an immediate commercial product. It mainly shows that major cloud players are already seeking alternatives to the traditional constraints of data centers, including energy, cooling, and scaling. The bet is ambitious but also very concrete: if space can host useful computing, part of the current hardware architecture will need to be reconsidered.
By the numbers
- October 1 — launch of the Project Suncatcher satellite.
- October 30 — announcement of the first AI response in orbit.
- 2027 — next targeted phase with two interconnected satellites.
- 2 satellites — system Google plans to test next.
- 1 TPU — datacenter-class processor onboard the prototype.