Hollow-core fiber: Not just for low latency?

Optical fiber’s ability to carry petabit-scale data rates over thousands of kilometers at low cost is the foundation of today’s networks and the global digital economy. However, the requirements of emerging applications are beginning to stress the limits of conventional silica-core fiber (SCF). Hollow-core fiber (HCF) is evolving rapidly and could offer solutions to these problems.

The SCF we've used for the past 50 years has some specific limitations:

• Light travels roughly 33 percent slower through glass than through a vacuum, air or gas, resulting in higher latency compared to free-space or wireless transmission.
• Glass has complex signal absorption behavior, limiting the low-loss transmission window for the C and L bands.
• SCF is a nonlinear optical medium, leading to signal impairments such as stimulated Raman scattering (SRS), which becomes increasingly significant for multi-band transmission.

In contrast, HCF is made up of multiple glass tubes, with the optical signal traveling through the hollow center, usually filled with an inert gas. This design addresses the limitations of SCF while remaining compatible with today's coherent transponders.

Lower latency

Don't you just hate those 1.6 extra milliseconds it takes light to travel 1,000 kilometers along SCF vs through the air? Latency-sensitive users such as high-frequency traders dislike it even more and are willing to pay a premium to reduce it. Even in less-demanding applications, fiber route latency accounts for over 90 percent of the end-to-end latency, depending on distances. Interest from the low-latency market, despite HFC’s higher manufacturing cost, has accelerated innovation and helped drive production scale, which in turn is reducing costs.

AI workloads further broaden the addressable market. Large-scale AI data centers increasingly rely on distributed processing across multiple sites within a metro region or campus-scale cluster. Since AI requires the lowest possible latency, HCF allows more geographical diversity while maintaining low interconnection latency targets. As a result, the “scale-across” market for AI interconnect is much larger than the high-frequency trading segment. AI and cloud giants like Microsoft and Amazon have publicly discussed HCF deployments of tens of thousands of kilometers.

Beyond low latency for HCF

In recent years, HCF vendors such as Microsoft, following its 2022 acquisition of Lumenisity, and YOFC have reported strong results at industry events such as OFC and ECOC. Demonstrated attenuation has dropped well below the 0.14 dB theoretical minimum for SCF, with reported minimum losses around 0.05 dB/km and average losses near 0.08 dB/km across a wide wavelength range of approximately 18 THz. Importantly, these results remain above theoretical minimum loss for HCF, indicating more room for improvement. Additional performance milestones are expected at OFC 2026.

Lower loss in HCF allows for much longer unamplified spans. In terrestrial networks, amplifier spacing is typically 60 to 80 km, with "high loss spans" over 100 km possible using Raman amplifiers. In unrepeatered subsea links, Nokia has demonstrated reaches of approximately 600 km using high-power amplifiers and remote optically pumped amplifiers (ROPAs). Assuming a 0.05 dB/km loss for HCF, extremely long unamplified spans are possible using less exotic technology, reducing the number of in-line amplifier sites and lowering both capital and operational expenditures.

HCF can tolerate significantly higher optical launch power than SCF without incurring nonlinear penalties. Nonlinear effects such as SRS are approximately 1,000 times lower in HCF than in SCF, improving reach and increasing the effectiveness higher-modulation formats such as 128QAM or 256QAM.

Finally, HCF offers low attenuation across a much wider optical bandwidth than SCF. The 18 THz low-attenuation band reported at OFC 2025 represents approximately 50 percent spectrum than the combination of Super C and Super L bands in today's SCF, potentially increasing fiber capacity.

What comes next?

Significant work remains to make HCF economically and operationally practical at scale. Manufacturing costs continue to fall as volumes increase, but today’s installation ecosystem, including connectors, splicers and test equipment is still optimized for SCF.  

This challenge is similar to when single mode fiber (SMF) was first deployed in the 1980s. SMF was expensive and complex to install compared to the existing multimode fiber (MMF). As adoption increased, costs fell rapidly, and SMF ultimately became both cheaper and easier to install than MMF. A similar trajectory is likely for HCF.

Wideband, long-reach unamplified deployments are promising, but amplification will eventually be needed. HCF poses challenges for conventional optical amplification techniques such as erbium-doped fiber amplifiers (EDFAs) and Raman, both of which rely on silica-based gain media. Research activity in hybrid SCF/HCF amplification, cladding pumping, and wideband fiber amplifiers is well underway, with further progress expected to be reported at upcoming OFC events.

A hot topic at OFC

The range of HCF related papers announced for OFC 2026 is impressive: 35 papers directly covering the topic, including seven contributions from Nokia and Bell Labs. These papers are expected to showcase record-breaking transmission capacity, reach and bandwidth, along with system-level demonstrations featuring real-time, bidirectional and unrepeatered links. Practical deployment solutions will be a dominant theme, including splicing, diagnostics, compatibility with existing infrastructure and optimizing hybrid HCF and SCF network designs, including detailed financial analysis. And finally, we’ll see a range of new application domains, such as sensing, terahertz transport, power–signal co-transmission and submarine systems.

Hollow-core fiber has moved well beyond niche technology status. While low latency remains the primary driver for early deployments, HCF points the way towards even lower latency, increased capacity and much longer unrepeatered reach, positioning it as an important component of future optical networks.