Google Launches First TPU AI Chip into Orbit for Space Data Center Tests
Google has launched a prototype satellite built by Planet Labs carrying its Tensor Processing Unit (TPU) into orbit aboard a SpaceX rocket under Project Suncatcher. The mission will evaluate chip thermal management, power consumption, and AI model execution during 15-minute operational bursts in space. This marks Google's first deployment of custom AI hardware in space as tech companies explore orbital data centers powered by continuous solar energy. If feasible, space-based computing could help bypass terrestrial land, energy, and environmental constraints for future massive AI workloads. According to Google's peer-reviewed paper in Journal Joule, scaling a planned 81-satellite orbital cluster will require launch costs to fall to $200/kg by 2035, requiring an estimated 1,800 SpaceX Starship launches carrying 200-ton payloads. Google and Planet Labs plan a follow-up demonstration next year using two specialized satellites connected via inter-satellite optical laser links.
## BACKGROUND
Tensor Processing Units (TPUs) are proprietary application-specific integrated circuits (ASICs) developed by Google specifically to accelerate neural network machine learning workloads. In-orbit data centers position high-performance computing hardware on satellites to harness unfiltered, continuous solar radiation, but face major engineering hurdles including vacuum thermal dissipation, cosmic radiation, and high launch costs.