New Hollow-Core Fiber Targets AI Data Centers

Hollow-core fiber guides light through a central channel filled with air instead of a solid glass core. Because light travels nearly 47% faster through air than glass, this design can significantly reduce data transmission latency. This latency reduction is critical for large-scale AI training, where massive volumes of data move between GPU clusters in what is known as "east-west" traffic. Network bottlenecks can leave expensive GPUs idle, and research from Meta has shown that early AI applications spent roughly a third of their time waiting on the network. AI data center networks require sustained high bandwidth and lossless, predictable connectivity to maximize GPU utilization. Current high-performance data centers are moving toward 400 Gbps and 800 Gbps Ethernet fabrics to handle these demanding AI/ML workloads. Yangtze Optical Fibre and Cable (YOFC) is a major China-based manufacturer that has been developing HCF technology for years. In June 2024, YOFC, in collaboration with China Mobile and China Telecom, launched the world's first 800G hollow-core fiber transmission test network and demonstrated transmission rates of 1.2 Tbit/s per wavelength. Despite its performance advantages, widespread HCF adoption faces hurdles. Manufacturing the complex microstructures is more difficult and costly than producing standard solid-core fiber. The technology also requires a new ecosystem of specialized connectors, splicing techniques, and amplifiers for deployment. Hyperscalers are already proving the technology's viability. Microsoft has deployed over 1,200 kilometers of hollow-core fiber to connect its Azure data centers in Europe. This allows them to connect sites over longer distances, treating them as a single operational unit and providing more flexibility in choosing locations based on energy costs or real estate.