Google Launches First Space Data Center Satellite Running Gemma 4 on TPU

Google's Project Suncatcher prototype satellite has successfully entered orbit and completed its first end-to-end AI inference, marking a critical shift from concept to reality for in-space computing. Equipped with a single TPU chip and the pre-installed Gemma 4 model, the satellite received ground instructions and returned a generated haiku, validating the closed-loop communication and computational capability.

Technically, the system is constrained by approximately 1kW of solar power, limiting the TPU to intermittent operation windows of about 20 minutes. Despite these early-stage limitations, the mission proves the feasibility of using orbital solar energy for large-scale AI inference. Theoretically, space-based solar power can generate up to eight times more energy than terrestrial sources, laying the groundwork for future low-latency, high-efficiency distributed AI infrastructure in near-Earth orbit.

This development signifies a major milestone for edge computing extending into space, offering new paradigms for autonomous spacecraft data processing. However, commercial viability remains distant, as challenges regarding long-term radiation shielding and thermal management must be resolved, with cost competitiveness expected only in the mid-2030s.