Space-XNet: Distributed LLM Execution in Satellite Networks

A new framework called Space Network of Experts (Space-XNet) proposes distributed deployment of large language models (LLMs) across satellite networks. The approach targets mixture-of-experts (MoE) models, partitioning and mapping model components to satellites to overcome limited onboard computing and communication resources. This enables low-latency token generation by reconciling model architecture with network topology. The research, published on arXiv (2605.00515), addresses a key challenge for space data centers that harvest continuous solar energy. Companies like SpaceX and Google are investing in this vision for executing energy-intensive LLMs in orbit.