•Transmission at 1.2Tbps per wavelength across air-filled core cuts latency by 30%
•Optical networks emerging as bottleneck alongside compute in AI infrastructure race
The fact
Researchers from China Telecom, Yangtze Optical Fibre and Cable Joint Stock Limited Company (YOFC) and optical equipment manufacturer Dekoli have demonstrated 51.3Tbps of aggregate data transmission across a 206.5km hollow-core fibre link without intermediate signal regeneration. The field trial achieved transmission rates of up to 1.2Tbps per wavelength using wavelength-division multiplexing (WDM), with the research team describing it as a record for an unrepeatered WDM system operating without remotely pumped optical amplifiers.
Unlike conventional optical fibre, hollow-core fibre guides light through an air-filled core instead of solid glass. According to the researchers, the design reduces signal latency by up to 30% while lowering transmission losses. Engineers also developed an adaptive channel management system that dynamically redistributes transmission power and data rates across individual wavelengths to maximise spectrum efficiency under commercial operating conditions.
The researchers said the technology could support future AI data centre interconnects, high-capacity backbone networks and next-generation 6G communications, where lower latency and higher throughput will become increasingly important.
The assessment
For years, AI infrastructure has been measured by the number of GPUs deployed. That metric is beginning to change. As AI clusters become larger and more geographically distributed, the ability to move data between them is emerging as a limiting factor. The next phase of AI competition will be determined not only by compute capacity, but by network economics.
China's latest demonstration suggests hollow-core fibre is approaching that transition. The significance is not simply transmitting more than 50Tbps over 206.5km, but doing so under conditions designed to resemble commercial telecommunications networks without signal regeneration. That reduces one of the barriers to practical deployment and strengthens the case for hollow-core fibre in AI backbones, financial trading networks and future national telecommunications infrastructure. The growing investment by cloud providers and network operators in hollow-core fibre reflects the same shift: faster processors alone are no longer enough if the network cannot keep pace.
For BTW readers, hollow-core fibre reduces latency by 30% on long-haul links—meaning AI data centre interconnects between cities could run without the repeater stations that add cost and delay. Countries and operators backing this technology are betting that the next bottleneck in AI will not be chips but the pipes between them.
What to watch
Watch for commercial deployments of hollow-core fibre by hyperscale cloud providers and national telecommunications operators, particularly for AI data centre interconnects. Progress on international standards, manufacturing scale and long-haul deployment will indicate whether hollow-core fibre is moving from experimental technology to strategic infrastructure.