Satellites lined up for shared space computing power networks

by Riko Seibo
Tokyo, Japan (SPX) Dec 31, 2025
Researchers are outlining a space computing power network that would link communication and computation across satellite constellations to handle growing data processing demands in orbit. Over recent decades, satellites have expanded global connectivity and remote sensing, but many systems still move raw data to the ground for processing under tight contact windows.

Traditional satellite-ground links must operate within short visibility periods, share limited high-frequency spectrum, and address data privacy concerns when downlinking sensitive information. To reduce these constraints, the proposed space computing power networks, or Space-CPN, use edge computing techniques and onboard computing payloads so satellites can process and filter information in space instead of sending all data to Earth.

The Space-CPN framework extends terrestrial computing power networks into space by integrating the communication and computation capabilities of low Earth orbit, medium Earth orbit, and geostationary Earth orbit satellites. Within this architecture, operators can schedule computing tasks flexibly across different satellite nodes to support secure, low-latency, and accurate onboard intelligent data processing. In emergencies, GEO, MEO, and LEO satellites could function as space-based computing centers, while ground stations would take on high-demand or less time-sensitive computing tasks.

A central challenge for Space-CPN is the design of communication principles that support specific computation tasks instead of simply maximizing raw data throughput. The work highlights a robust information bottleneck principle that seeks to maximize mutual information between the computed result and the true label of a data sample to maintain accuracy, while minimizing mutual information between the extracted feature and the input sample to improve compression. This balance is intended to strengthen communication robustness without adding overhead to the system.

Onboard computing architectures in satellites must also operate under constrained energy and hardware resources, which drives interest in low-power approaches. The study points to neuromorphic computing, which mirrors how the brain combines memory and processing, as one candidate solution for spacecraft. It examines the use of spiking neural networks in orbit and proposes satellite federated and decentralized neuromorphic learning network architectures to support onboard training across distributed satellite nodes.

Resource allocation across a Space-CPN is another core issue because satellite networks are dynamic and uncertain, with changing connectivity, workloads, and link conditions. The article proposes robust optimization tools to match computation tasks to available links and processing units, including robust reinforcement learning for deploying satellite microservices and distributionally robust optimization for satellite task scheduling. These methods aim to quantify the relationship between task requirements and network resources so operators can maintain efficient utilization as conditions change.

Together, these elements position space computing power networks as a next step in merging communication and computing throughout space infrastructure. By rethinking task-oriented communications, adopting energy-efficient onboard computing paradigms, and applying robust resource allocation strategies, such networks could change how satellites process and transmit information and support more capable and autonomous satellite systems.

Research Report:Space Computing Power Networks: Fundamentals and Techniques