Google’s prototype data center satellite, launched in October 2026, successfully ran an onboard AI model in orbit and returned a creative response back to Earth—a haiku reflecting its experience in space.
- First AI output from a data center TPU chip in orbit
- Satellite uses stored solar energy to power AI computations
- Next phase includes linked satellites and laser interconnects in 2027
What happened
Google launched its prototype satellite, Suncatcher, in October 2026 as part of Project Suncatcher to test AI computing capabilities in space. Onboard the satellite, an AI model named Gemma 4, running on specialized TPU chips, generated a haiku about its surroundings, which was transmitted back to Earth during a flyover of Svalbard, Norway.
This historic event marks the first time a data center-class TPU chip has run AI in orbit and sent the output back to the ground. The satellite carries four TPUs but, due to a power limit of roughly one kilowatt, operates just one TPU at a time for about 20 minutes during each pass. Initial testing took three weeks, confirming the satellite’s systems operate normally in space conditions.
Why it matters
Running AI processing on orbiting data centers could revolutionize how data is handled globally by reducing latency and reliance on Earth-based infrastructure. This project demonstrates the feasibility of deploying powerful AI chips in space, utilizing solar power and managing technical challenges such as heat dissipation and radiation effects on memory and data integrity.
Google’s Suncatcher provides key insights into the robustness of TPU chips under orbital environmental conditions, where radiation can cause data corruption. Early ground tests indicate that running existing models in space is less vulnerable to errors than training new models, a critical distinction for operational stability.
What to watch next
Google plans to launch a pair of linked satellites in 2027 designed to generate more power and test laser communication links between them. These developments aim to enhance the capabilities and efficiency of space-based AI computing infrastructure.
The success of Project Suncatcher also depends on advancements such as the full reusability of SpaceX’s Starship rocket, which would support deployment and maintenance costs. Meanwhile, startups and competitors like Starcloud are also pursuing orbital data center technology amid some skepticism from industry players, including SpaceX, about the commercial viability of this approach.