Hot AI in Cold Space: Thermal-Crosstalk-Aware Scheduling for Sustainable Orbital AI Clusters

This collection of arXiv papers covers system-level advances for large-scale AI and distributed computing with several security-relevant implications. Key topics: thermal-aware scheduling for Orbital Data Centers (Thermal-Load Balancing) that treats spatial cooling variance as a schedulable resource — failure or malicious manipulation could induce thermal throttling, premature hardware fatigue, and availability/DoS of orbital AI infrastructure. Moebius enables live switches between expert- and tensor-parallel layouts by resharing expert weights and KV cache across GPUs (fast GPU-to-GPU state移転