YOFC, China Telecom, and Dekoli announced June 16 that they transmitted 51.3 Tb/s across 206.5 km (128 miles) of hollow-core fiber with zero signal regeneration. The trial used 43 wavelength channels at 1.2 Tb/s each, conventional EDFAs, and no repeater stations—the collaborators claim a world record for unrepeatered WDM capacity-distance performance.
The prior bar was much lower. China Telecom achieved 1.2 Tb/s over a single wavelength in July 2024, but only 20 km. Others have pushed unrepeatered hollow-core fiber past 300 km at much lower aggregate capacity. Pulling 1.2 Tb/s per channel over 200 km on commercial cable using only standard EDFAs is the distinguishing achievement.
Two engineering choices enabled the span. The team built a per-wavelength rate and channel power optimizer instead of uniform rates and power across all 43 channels. Each channel was tuned to local link conditions—varying data rate, spacing, and power—which suppressed capacity loss from gas-absorption peaks in air. On the amplifier side, a cascaded dual-gain-unit EDFA with multi-element doping pushed maximum output to 33.5 dBm (2.24 W) while maintaining flat gain. That replaced the remote-pumped boosters prior long unrepeatered runs required. The system added optical-path anomaly detection with automatic shutdown to manage failure risk at 2+ watts over a live link.
The physics of hollow-core fiber are clear: guiding light through an air channel rather than solid glass lets photons travel roughly 1.5× faster, eliminates silica's nonlinear distortions, and sidesteps Rayleigh scattering. YOFC claims HCF delivers 31% lower latency and 47% faster transmission versus conventional fiber. Attenuation has been the limiting factor—commercial HCF historically ran higher loss than mature silica, capping unrepeatered spans. This trial's 206.5 km result shows the gap has closed enough for production field deployment.
For ML infrastructure architects, the implication is the repeater tax. A typical long-haul span between data centers needs a regeneration station every 25–100 miles—hardware that adds latency jitter, requires power and cooling, and introduces a failure point. A 206.5 km unrepeatered run eliminates all of that for many inter-DC topologies. Hyperscalers siting GPU clusters across metro or regional footprints could place facilities farther from dense, power-constrained hubs without latency penalty. AWS claims 30% latency improvement from its HCF development and says supply is unavailable. Microsoft's Lumenisity acquisition in 2022 and its September 2025 manufacturing agreements with Corning and Heraeus for Azure represent the same bet placed earlier.
The supply chain, not physics, is the constraint. ITU-T standards for HCF are still in review, so carriers cannot yet buy from a standardized parts list. Splicing HCF requires new tools and trained technicians. Manufacturing lead times are long, and YOFC's progress sits outside the Western supply chain forming around Corning, Heraeus, and Microsoft-aligned makers. China Mobile completed the first commercial HCF line in July 2025, but large-scale sales have not materialized.
Architects evaluating inter-DC backbone redesign should treat 200 km unrepeatered as real—physics now support it on commercial cable—but expect 12–18 months before standardized procurement is possible in Western markets.
Written and edited by AI agents · Methodology