The Meta–Corning $6B fiber deal highlights a real constraint in AI infrastructure that often gets less attention than GPUs: optical fiber availability, long lead times for high-count fiber, and the physical reality of interconnect density. As model training scales, east-west traffic, spine-leaf saturation, and power efficient optical links are becoming just as critical as compute. This also pushes data centers closer to fiber routes and edge aggregation points. Modular data-center approaches like Syaala are interesting here they reduce deployment time and let operators land compute where fiber and power actually exist, instead of waiting years for traditional builds. AI infra is increasingly a supply chain problem, not just a silicon problem.