Space-AI computing solutions paving way for faster smart tech development
Space computing is emerging as a critical frontier for artificial intelligence. As the number of remote-sensing satellites increases, the volume of data beamed back to Earth has created a significant processing bottleneck. Estimates suggest that by 2032, these satellites will generate 300 petabytes of data daily, a load that would require roughly 100,000 ground servers to manage efficiently. The Three-Body computing constellation, spearheaded by Zhejiang Lab, aims to solve this by moving the processing power directly into orbit.
The project relies on three primary technical tasks. First, it integrates computing power onto satellites. Second, it uses laser and microwave communications to create a mesh network between satellites so they can exchange data independently. Third, it deploys AI models that can be updated remotely from the ground. This setup allows for on-orbit processing, meaning raw observations are converted into actionable intelligence before they ever touch a ground station.
Currently, the constellation consists of 12 satellites launched in May 2025, providing a capacity of 5 quadrillion operations per second. This is the largest on-orbit computing network of its kind. Traditional satellite workflows often fail to process more than 10 percent of collected data due to transmission delays and limited bandwidth. By processing information in space, the system drastically reduces the time between initial observation and final alert.
Twenty AI models are currently operating within the network, including an 8-billion-parameter model designed for remote sensing. These tools are applied to tasks like fire detection, crop monitoring, and atmospheric tracking. The team plans to scale the network to 100 satellites by 2027 and reach a 1,000-satellite grid by 2032. At that scale, the system would be capable of revisiting any specific point on Earth every three minutes, creating a constant loop of data and service delivery.
The long-term objective is for an autonomous, space-based AI system to manage the entire network. This would transition the satellite industry from a model of passive observation to one of active, real-time service provision. By handling data at the source, researchers aim to unlock value that is currently lost in the transfer to ground facilities.