LLMSpace: Carbon Footprint Modeling for Large Language Model Inference on LEO Satellites

2026-05-06

AI Summary

This work addresses the current lack of systematic assessment of the full lifecycle carbon footprint of large language model (LLM) inference on low Earth orbit satellites, particularly the overlooked embodied emissions from launch, manufacturing, and radiation-hardened hardware. We propose the first carbon footprint modeling framework tailored to space-based LLM inference, jointly accounting for operational and embodied emissions, auxiliary subsystems, radiation-hardened accelerators, and the unique prefill–decode workload characteristics of LLMs. Integrating real satellite platform specifications, GPU configurations, radiation-hardened hardware power models, and life cycle assessment methodologies, our framework enables the first fine-grained joint analysis of carbon emissions and inference latency. This reveals critical sustainability trade-offs among carbon footprint, latency, hardware design, and satellite lifetime, offering foundational insights for designing green AI systems in space.

Abstract

Large language models (LLMs) impose rapidly growing energy demands, creating an emerging energy and carbon crisis driven by large-scale inference. Solar-powered, AI-enabled low Earth orbit (LEO) satellites have been proposed to mitigate terrestrial electricity consumption, but their lifecycle carbon footprint remains poorly understood due to launch emissions, satellite manufacturing, and radiation-hardened hardware requirements. This paper presents \textit{LLMSpace}, the first carbon modeling framework for LLM inference on AI-enabled LEO satellites. LLMSpace jointly models operational and embodied carbon, peripheral subsystems, radiation-hardened accelerators and memories, and LLM-specific workload characteristics such as prefill-decode behavior and token generation. Using realistic satellite and GPU configurations, LLMSpace reveals key trade-offs among carbon footprint, inference latency, hardware design, and operational lifetime for sustainable space-based LLM inference. Source code: https://github.com/UnchartedRLab/LLMSpace.

Problem

• carbon footprint
• large language models
• LEO satellites
• inference
• sustainability

Innovation

• carbon footprint modeling
• large language models
• LEO satellites
• radiation-hardened hardware
• sustainable AI