- Published on
Google Announces 'Hello World' Milestone: Orbiting Data Center-Class TPU Delivers First AI Response and Generates a H...
- Authors
- Name
- DeepHorus
On October 30 local time, Google announced that "Project Suncatcher" had reached the "Hello World" milestone: for the first time, a data-center-grade TPU operating in orbit produced an AI response. Previously, Google successfully launched the project's prototype satellite on October 1. Within weeks, the team completed the in-orbit testing and debugging checklist, including powering up the host and the TPU tray.
This week, ground engineers sent the first prompt to the Google DeepMind model Gemma 4 preinstalled on the satellite, asking it to compose a haiku about being in space. The model immediately replied: "Above the blue sphere, / silent streams of data flow— / starlight feeds the core." Google said this marks a key step forward for in-orbit AI inference.
In addition to haiku generation, engineers also tested telemetry tasks such as limerick generation and the return of orbital environmental data. The related tests were used to examine the ability of the satellite platform and the model to work together in the space environment. Google did not disclose more detailed metrics such as response latency or compute scale.
According to the introduction, the satellite carries four Google TPUs and can generate up to 1 kilowatt of electrical power. Its goal is to lay the foundation for building, in the future, a tightly formation-flying satellite network for executing large-scale inference workloads in orbit. Google views this direction as an exploration of extending data center capabilities into space.
Limited by heat dissipation and energy load, the onboard TPUs can be powered on for only about 20 minutes at a time. The tests therefore needed to be completed within limited windows. The core of the testing was to verify the chips' stability and error-correction capability under microgravity, severe temperature swings, and space radiation.
"Hello World" usually represents the first successful run in programming. Google's completion of an AI response with an in-orbit TPU this time means that satellite payloads are beginning to extend from traditional telemetry communication toward in-orbit intelligent processing. If subsequent verification goes smoothly, some data processing tasks are expected to be completed on the space side.
Under the vision of "Project Suncatcher," multiple satellites in the future could form a tight formation and execute large-scale inference workloads in orbit. Google said that in-orbit testing and debugging of the prototype satellite are still ongoing. This milestone provides a preliminary basis for subsequent verification of satellite networks and space computing capabilities.
