China has achieved a significant milestone in space technology with the successful deployment of artificial intelligence models operating directly in orbit through an advanced satellite network. The breakthrough demonstrates the nation’s growing capabilities in space-based computing systems.
The pioneering project, known as the “Three-Body Computing Constellation,” represents a collaborative effort between Zhejiang Lab and international partners. Initially launched in May 2025 with twelve satellites, the constellation has undergone extensive testing over nine months to validate its core functionalities including orbital networking, computational processing, AI model deployment, and scientific instrument verification.
Among the technological achievements are the operation of two massive AI models in space: an 8-billion-parameter remote sensing system and an equally sophisticated astronomical time-domain model. These represent some of the largest AI parameter models currently operational in orbit worldwide.
The remote sensing capability was demonstrated in November 2025 when it successfully conducted an infrastructure census across 189 square kilometers in northwest China, automatically identifying structures like stadiums and bridges despite challenging heavy snow conditions.
For astronomical research, satellites equipped with specialized cosmic X-ray polarization detectors implemented an AI model that classifies gamma-ray bursts with remarkable 99% accuracy while significantly reducing data transmission requirements and processing timelines.
A critical advancement includes the establishment of inter-satellite communication links among six satellites, forming the foundation for a comprehensive space network. When fully realized with over 1,000 planned satellites, the constellation is projected to deliver unprecedented computational power exceeding 100 quintillion operations per second.
According to Li Chao from Zhejiang Lab, this space-based computing architecture enables direct data processing in orbit, allowing refined information to be delivered immediately to end-users rather than raw data requiring extensive ground processing.