AI data centers require 36 times more fiber than standard servers

The rapid expansion of AI data centers has led to a surge in demand for optical fiber, far exceeding traditional telecom network usage. Major Chinese fiber manufacturers like Hengtong and FiberHome ar... Read original →

The explosive fiber demand from AI data centers is triggering a full-stack technological realignment—from advanced chip packaging to optical cable manufacturing. The reliance of 3nm nodes and CoWoS packaging on dense optical interconnects is forcing foundries like TSMC into deep partnerships with Corning. Meanwhile, the industry’s pivot to higher-margin G.657A fiber has starved legacy G.652D supply, inflating infrastructure costs. NVIDIA’s $300M bet on U.S. fiber capacity reflects a strategic de-risking move amid geopolitical tensions, yet new plants won’t ramp until 2027—leaving Hengtong and FiberHome as near-term bottlenecks, albeit under potential export control scrutiny. Cloud giants like Meta locking in multi-year deals signal that resource competition has shifted from chips to the physical layer. Over the next 18 months, optical fiber will become the ‘new oil’ of AI infrastructure: any supply delay will directly throttle GPU cluster deployment and accelerate silicon photonics commercialization.