Google's First Space Satellite Runs Gemma 4 On Orbit

Google’s Project Suncatcher has achieved a critical milestone by successfully launching its first prototype satellite, which activated an onboard TPU to run the Gemma 4 model. This completed an end-to-end inference loop, generating an AI response (a haiku) from orbit in reply to a ground-based prompt. The event marks the first time data-center-grade AI chips have delivered actual compute power in space, validating the feasibility of solar-powered orbital computing.

Cross-referenced reports indicate that the satellite is constrained by approximately 1kW of power and thermal limits, allowing the TPU to operate for only about 20 minutes per session. The test primarily verified hardware stability and error correction capabilities under microgravity, extreme temperature fluctuations, and radiation. While still in the proof-of-concept phase, the successful inference confirms the technical viability of running specialized AI chips in orbit, laying the groundwork for future tightly formed satellite networks.

This development signifies a shift from theory to practice for “space edge computing,” offering new paradigms for low-latency spatial data processing and autonomous spacecraft. If subsequent radiation and thermal stress tests succeed, Google plans to deploy a dual-satellite interconnected system by 2027 to build an orbital AI computing network. This could reshape global infrastructure, though commercial cost competitiveness is not expected until the mid-2030s.