A hollow-core fiber cable just carried 51.3 Tb/s across 200 km

Julio Franco

Posts: 9,318   +2,252
Staff member
Forward-looking: Chinese fiber-optic giant YOFC has set a new world record for long-haul data transmission, pushing 51.3 Tb/s across 206.5 km of hollow-core fiber without any intermediate signal regeneration. The trial is the first field demonstration of a hollow-core fiber system capable of delivering 1.2 Tb/s per wavelength – on a live commercial cable, not a lab bench.

Conducted jointly with China Telecom and optical equipment maker Dekoli, the test ran on the world's longest cross-border commercial HCF cable. The result sets a new world record achieved without the signal boosters that long-haul fiber links typically depend on.

Unlike conventional fiber, which guides light through solid glass, HCF guides it through air. That structural difference matters: light travels roughly 1.5 times faster through air than glass, and the air core sidesteps much of the signal distortion that silica introduces over long distances. YOFC claims the technology delivers about 31% lower latency and 47% faster transmission speeds than conventional fiber.

The latency advantage isn't academic. YOFC already operates a trading link between the Shenzhen and Hong Kong stock exchanges that achieves round-trip latency of under one millisecond.

The 51.3 Tb/s figure wasn't achieved by brute-forcing every channel at max power. Instead, the team dynamically tuned each wavelength's data rate and power level individually, allowing the system to work around a particular HCF limitation: gas absorption peaks that occur when light travels through air rather than glass. The approach let them squeeze more usable capacity out of the full channel spectrum.

They also had to solve an amplification problem. Sending high-power optical signals through a live HCF link without damaging it required a custom amplifier design to maintain consistent signal strength across the full operating range while hitting a peak output of 33.5 dBm. Automatic fault detection, interlock shutdown, and alarm-linked safeguards were added on top to catch problems before they cause damage.

What separates this from earlier HCF results is the combination of capacity, distance, and amplification approach. Others have pushed HCF spans past 300 km before, but at much lower data rates.

Also, a 1.2 Tb/s per wavelength had been demonstrated over short distances. China Telecom showed it over 20 km back in 2024. What YOFC is claiming here is the combination: that throughput, over 200+ km, on a commercial cable, using only standard erbium-doped fiber amplifiers instead of the more exotic remote-pumped boosters typically required for long uninterrupted spans.

The timing isn't coincidental. As hyperscalers race to build larger GPU clusters, the fiber between those clusters (within and between data centers) is becoming a real bottleneck. HCF's lower latency lets operators site facilities farther apart without a speed penalty; it also extends the practical distance between interconnected data centers from roughly 60 km to around 90 km, opening up more options for land and power in markets where urban cores are already constrained.

China's HCF push is largely developing in parallel to the Western supply chain forming around the same technology. Microsoft acquired HCF pioneer Lumenisity in 2022 and signed manufacturing deals with Corning and Heraeus in September 2025 to expand production for Azure. AWS has built its own HCF and claims a 30% latency improvement over standard fiber – and says it wants more of it than suppliers can currently provide.

On the manufacturing front, YOFC has recorded lab attenuation as low as 0.040 dB/km, which falls below the theoretical loss floor of conventional silica fiber. Commercial production has reached consistent sub-0.1 dB/km – a threshold that makes real deployment viable.

Practical challenges remain, though. Splicing HCF to standard fiber still introduces losses of 0.5 to 2 dB in the field, compared to under 0.1 dB for a conventional splice. Field deployment requires new tools and training. And Amazon's AWS has been blunt: it's not demand holding back wider adoption, it's the difficulty and cost of manufacturing the fiber at scale.

This YOFC trial doesn't solve those problems, but it does demonstrate that backbone-scale HCF transmission is real outside of a lab. For a technology still working through its growing pains, that's a meaningful step forward – and a sign that China intends to be a major player in whatever comes next for optical networking.

Permalink to story:

 
Yet, no doubt stole the technology....
There is tons of intellectual property that China steals, but this is a hollow tube and the real miracle is on the manufacturing side. If they have the skill to manufacturer something within such tolerances to break records they deserve some credit. And considering the west just use patent laws to slow progress and sue people, I really couldn't give less of a damn if China did steal anything.
 
Oh wont someone think of the billionaires!
Western tech firms pay taxes to Western governments and employ skilled Western workers; when those firms can no longer compete due to IP theft, what do you think the consequences are?

Anyone who fails to understand that Chinese IP theft affects us all should be surgically sterilized as soon as possible, to prevent those defective genes from polluting the rest of the pool.
 
China, is NOT your friend.
Neither is the patent and copyright system in many cases*.
There should be something more reasonable in between the two evils.

Just shouting stolen technology doesn't make it true either. China with its massive population, focus on education and advanced industrial and technological sectors innovates/advances on its own just fine as well.
It still steals a lot, but there's tons of Chinese companies making new stuff and filing for their own patents as well. Or simply being industry leaders in many fields.

*Copyright lasts too darn long.
*Patents are given out to freely.
- Many don't even hold up in court, but it sure lets bigger companies bully smaller ones into spending bankrupting amounts in courts / prevent new players from entering a market.
- Or worse, let patent troll companies exist that don't produce ANYTHING yet make the costs for everyone that isn't them go up whilst employing no one besides overpaid lawyers.

Patent system is long due for an overhaul. imo they shouldn't be transferable (or under some strict set of rules that makes sense). Depending on what they're for should probably have their duration halved or even quartered. Should not be given out for trivial things like Apple patenting an electronic device in the shape of a rounded rectangle.

Perhaps that's a good use case for AI... prevent trivial patents from being awarded. It should be good for being aware of 'prior art' being trained on just about everything ever made.
 
Neither is the patent and copyright system in many cases*.
There should be something more reasonable in between the two evils.
Anyone who believes IP protection is "evil" should be spanked and sent to bed without supper. In the case of patents, they replaced the medieval system that hid knowledge in the minds of a tiny handful of guild masters for centuries, often to be lost forever.

With patents, individuals and businesses freely give the entire world their innovation --described in every detail to allow duplication -- in exchange for a brief, 20-year period of exclusive use. Britain's development of the modern patent system in 1623 is what sparked the industrial revolution and turned that tiny backwards island nation into the world epicenter of technology for two full centuries.

Just shouting stolen technology doesn't make it true either
No: all the tens of thousands of documented Chinese thefts make it true.

*Copyright lasts too darn long.
Now this of course is very true. Blame Sonny Bono and Disney for that. The earlier system of 28+28 years made far more sense. That doesn't invalidate the concept, however.
 
Back