The Cost and Network Limits of Space-Based AI Compute

This paper evaluates whether large-scale AI data centers in low-Earth orbit could become cost-effective alternatives to terrestrial facilities. It compares launch costs, power generation, cooling, radiation exposure, atmospheric reentry, and compute-network performance. A central architectural difference is the move from terrestrial Clos networks to space-based mesh networks connected by laser inter-satellite links. Using bisection bandwidth, bisection intensity, and roofline-style models, the authors conclude that LEO inference may be feasible, while training frontier-scale large language models is unlikely to compete economically or technically with terrestrial data centers.