YOFC Unveils Hollow-Core Fiber for AI Data Centers

Hollow-core fiber (HCF) transmits light through an air-filled channel, which allows signals to travel nearly 50% faster than through the solid glass core of traditional single-mode fiber. This architectural shift reduces latency by approximately 30%, a critical advantage for synchronizing the thousands of GPUs in distributed AI training clusters. This speed-of-light advantage translates to a latency reduction of about 1.5 microseconds for every 100 kilometers of distance. The technology isn't just about speed; it also promises lower signal loss (attenuation) and minimal non-linear distortion, allowing for more bandwidth and longer distances between data centers without the need for signal repeaters. Recent breakthroughs have demonstrated attenuation levels as low as 0.11 dB/km, which is better than the theoretical limit for conventional silica fibers. This enables data center operators to build facilities in more remote locations with access to cheaper land and renewable energy sources. This innovation directly challenges existing high-performance interconnects that are central to the custom silicon landscape, such as NVIDIA's NVLink. While NVLink provides ultra-high-speed, low-latency connections for GPUs within a server or across racks, HCF aims to extend that performance envelope over metropolitan-area distances, potentially altering the architecture of large-scale AI clusters. Several major cloud providers have already started deployments and formed partnerships to scale HCF manufacturing. The market for next-generation optical fiber, including hollow-core, is projected to grow significantly, from $137.4 million in 2022 to over $1 billion by 2031. Major players like Prysmian, Corning, and Fujikura are active in this space, alongside YOFC. Despite its advantages, widespread adoption faces hurdles, including manufacturing costs that are currently much higher than traditional fiber and the technical challenges of splicing the novel fiber into existing networks.