Multicore Fiber for AI-Scale Data Halls

This technical report evaluates the carbon footprint and operational efficiency of high-density, preterminated passive optical infrastructure in AI data halls, specifically comparing 4-core multicore fiber (MCF) to traditional single-core fiber (SCF). The study focuses on a representative generative AI data hall architecture supporting 18,432 GPUs. It quantifies the carbon footprint of optical fiber and associated components, revealing that MCF can reduce the carbon footprint per GPU by up to 60%, decreasing the total data hall carbon footprint from 4,549 metric tons CO₂e for SCF to 1,792 metric tons CO₂e for MCF. The report outlines how MCF increases connectivity density without expanding cable bulk, addressing physical constraints in data center pathways. Additionally, it discusses the implications for skilled labor availability and deployment timelines, suggesting that MCF solutions can streamline network architecture and reduce installation labor by approximately 60%, thereby facilitating faster deployment in the context of growing AI infrastructure demands.