New Optical Fiber Tech Aims to Boost AI Infra
At MWC Barcelona, Yangtze Optical Fibre and Cable (YOFC) is set to unveil a new Hollow-Core Fibre (HCF) solution. The technology is designed to provide the ultra-low latency optical communication needed to improve the performance of global AI data centers.
Hollow-core fiber (HCF) guides light through an air-filled channel, unlike traditional solid-core fibers that use glass. This fundamental design difference allows light to travel at nearly its vacuum speed, approximately 30-33% faster than in solid silica. This speed increase directly translates to a significant reduction in data transmission latency. For AI workloads, particularly distributed training across multiple data centers, this latency reduction is critical. Lower latency minimizes GPU idle time, accelerating model training and improving the overall efficiency of expensive compute resources. The technology also allows data centers to be located further apart, offering greater flexibility in site selection based on real estate or power costs. High-frequency trading (HFT) was an early adopter of HCF, where a latency improvement of just 1.5 microseconds per kilometer provides a significant competitive advantage. Financial firms have utilized HCF for last-mile connections and between data centers to execute trades milliseconds ahead of competitors. Yangtze Optical Fibre and Cable (YOFC) has been a key player in advancing HCF technology. In June 2024, YOFC, in collaboration with China Mobile, launched the world's first 800G hollow-core fiber transmission test network. More recently, at the OFC Conference in March 2025, the company announced it had reduced attenuation (signal loss) to a record-low 0.05dB/km, a significant step toward broader commercial viability. Beyond speed, HCF offers other advantages, including lower signal distortion at high power levels and reduced backscatter, which is about 10,000 times lower than in traditional fibers. This allows for higher-power data transmission over longer distances with less degradation, further strengthening the infrastructure for demanding AI applications.