China Completes Orbital AI Constellation Test

- The initial launch for the constellation in May 2025 consisted of 12 satellites, which collectively provide 5 peta-operations per second (POPS) of computing power and 30 terabytes of onboard storage. The long-term plan is to expand this to a network of over 1,000 satellites. - During testing, the constellation successfully ran an 8-billion-parameter AI model in orbit, one of the largest ever operated in space, to analyze remote sensing data. This model was used to automatically identify infrastructure like stadiums and bridges through heavy snow cover. - A key capability demonstrated was autonomous classification, where the satellites identified astronomical events like gamma-ray bursts with 99% accuracy and terrestrial features with 94% accuracy, all without ground intervention. This shifts the paradigm from downlinking raw image data to sending pre-processed "answers." - The satellites are interconnected with 100 Gbps laser communication links, enabling distributed computing tasks where one satellite can capture data and offload the processing workload to another in the network. - This project is a prime example of Orbital Edge Computing (OEC), a model being pursued by other major players. Competitors in the space-based AI infrastructure race include Google's Project Suncatcher and SpaceX's plans to leverage its Starlink network for orbital compute. - The project is a joint effort between the state-backed Zhejiang Lab, commercial company ADA Space, and other partners, highlighting a hybrid approach to developing critical space infrastructure in China. - The explicit goal is to overcome the bottleneck of satellite-to-ground communication by processing data in-orbit, enabling real-time intelligence for disaster response, environmental monitoring, and defense applications.