SpaceMoE: Towards Orbital General Intelligence with Distributed Mixture-of-Experts Inference

2026-05-16

AI Summary

Satellite networks face stringent constraints in memory, computational power, and energy, which hinder the deployment of general-purpose on-orbit intelligence. This work proposes SpaceMoE, a novel paradigm that introduces the Mixture-of-Experts (MoE) architecture to the satellite environment for the first time. By leveraging sparse activation and distributed inference, SpaceMoE enables efficient large language model inference under dynamic network topologies and tight resource limitations. To address unique challenges such as onboard battery degradation and thermal constraints, the framework innovatively redesigns three core mechanisms: expert placement, expert selection, and hidden state routing. These are further integrated with resource-aware scheduling and communication optimization strategies, yielding a scalable, efficient, and sustainable on-orbit intelligent inference framework tailored for satellite networks.

Abstract

As satellite networks evolve to support increasingly diverse services and artificial general intelligence (AGI), large language models (LLMs) are emerging as a critical foundation for future space systems. However, deploying LLMs on satellites is hindered by stringent constraints on onboard memory, computation, and energy. In this context, the mixture-of-experts (MoE) architecture emerges as a promising solution, leveraging sparse expert activation to enable scalable model inference. By harnessing the architectural advantages of MoE, this article provides a comprehensive overview of SpaceMoE, a new paradigm for distributed MoE inference in satellite networks. We first review recent industrial progress and emerging standardization trends that motivate the evolution toward space AGI systems. Then, we introduce the fundamentals and architectural evolution of SpaceMoE. Subsequently, we discuss three fundamental design problems in SpaceMoE, namely expert placement, expert selection, and hidden-state transmission and routing, highlighting how satellite-specific factors such as dynamic topology, battery degradation, and thermal limits fundamentally reshape their solutions. Finally, we outline promising research directions for realizing scalable, efficient, and sustainable on-orbit MoE inference in future satellite networks.

Problem

• satellite networks
• large language models
• mixture-of-experts
• onboard constraints
• orbital general intelligence

Innovation

• SpaceMoE
• distributed Mixture-of-Experts
• satellite networks
• on-orbit inference
• space AGI