New Fiber Tech Promises Lower AI Latency

Hollow-core fibre transmits light through an air-filled channel instead of a solid glass core. This allows light to travel up to 50% faster, approaching its vacuum speed of nearly 300,000 km/s, compared to roughly 200,000 km/s in traditional silica glass. This speed increase directly reduces latency by about a third. Standard fiber optics incur a delay of approximately 5 microseconds per kilometer, whereas hollow-core fiber cuts that down to about 3.46 microseconds per kilometer. Over a 1,000-kilometer distance, this can trim latency by 1.54 milliseconds. In high-frequency trading, where latency is a direct component of profitability, such reductions are critical. A single microsecond of advantage can determine the outcome of a trade, and the market for exploiting these latency advantages is estimated to be worth $21 billion annually. For AI workloads, the primary benefit is mitigating "tail latency" in data centers, where even the slowest data packet can idle entire clusters of expensive GPUs during synchronized training tasks. Distributed AI training requires inter-GPU latency below 10 microseconds to avoid computational stalls and improve job completion times. Yangtze Optical Fibre and Cable (YOFC) is a major global supplier, holding a significant share of the Chinese market alongside competitors like FiberHome and HTGD, and competing globally with firms such as Corning (USA) and Prysmian (Italy). The company's key clients include major state-owned carriers like China Mobile and China Telecom. Major tech firms are already deploying this technology. Microsoft acquired HCF vendor Lumenisity and is using the fiber to connect its Azure data centers, validating its reliability for mission-critical cloud and AI workloads. European provider euNetworks has also deployed a 45 km HCF link between two London data centers. [4