HotCarbon'25 paper proposes Thermal-Load Balancing to cool space AI clusters

Terrestrial AI training consumes unsustainable energy and water, positioning Orbital Data Centers (ODCs) as a 'zero operational carbon' alternative. However, distributed LLM training demands sub-10μs communication latency, forcing ODCs into extreme physical density and triggering a critical 'Proximity-Thermal Paradox.' High-density systems suffer from thermal-fluid crosstalk (heat traps in shared cooling loops) and thermal-radiative crosstalk (mutual heating blocking deep-space radiators). Left unmitigated, this causes severe thermal throttling that degrades training throughput and induces thermal fatigue, drastically shortening hardware lifespans and generating premature space e-waste.

To address this, the HotCarbon'25 paper introduces the Thermal-Aware Heterogeneity Thesis, treating spatial cooling variances as a primary resource management dimension. The authors propose Thermal-Load Balancing (TLB), a software framework that dynamically migrates LLM workloads to the coolest available units based on instantaneous fluid temperatures or absorbed radiation. Analysis shows TLB resolves thermal bottlenecks to restore Model Flops Utilization (MFU) while simultaneously reducing physical thermal stress. Extending operational lifespan is crucial to amortize the massive embodied carbon of rocket launches, outlining a pathway to scale orbital AI without accelerating e-waste.