Orbital Intelligence: Space-Based Infrastructure as the Next Frontier for Planetary-Scale AI

This paper evaluates the feasibility of orbital AI data centers—constellations of accelerator‑equipped satellites powered by space solar arrays and cooled by radiative heat rejection—to relieve terrestrial power, cooling, and siting constraints for large‑scale machine learning (ML) workloads. We formalize compute demand and terrestrial limits, present a techno‑economic framework that integrates launch cost trajectories, levelized cost of energy (LCOE), radiator sizing, and latency/bandwidth trade-offs, and simulate representative workload classes to identify viable early use cases. Results show that radiator area, launch $/kg, and workload latency tolerance are the dominant determinants of viability. Under optimistic launch‑cost declines and advances in lightweight radiator materials, orbital clusters can be cost‑competitive for delay‑tolerant, high‑energy batch training and archival processing; latency‑sensitive inference and interactive services remain best served terrestrially. We conclude with a phased roadmap, legal and governance considerations, and a research agenda addressing serviceability, debris mitigation, and cross‑border data sovereignty.