Microsoft’s Azure Fibre R&D team — working with researchers from the University of Southampton and the Lumenisity spin‑out — has published results showing a hollow‑core (air‑cored) optical fiber with record low attenuation of 0.091 dB/km at 1,550 nm, a broad low‑loss spectral window, and significant latency and dispersion advantages over conventional silica single‑mode fiber. (phys.org)
For roughly four decades, optical networking has been dominated by silica‑core single‑mode fiber (SMF). Improvements in silica attenuation have been incremental, and the practical floor for low‑loss SMF in the C‑band has generally been cited near 0.14–0.16 dB/km, a constraint that has shaped amplifier spacing, submarine cable economics, and transponder designs. Hollow‑core fibers (HCFs) aim to change that by guiding most optical energy through air rather than glass, reducing material scattering and nonlinear effects and bringing the effective group index closer to 1.0 — i.e., light travels nearer to vacuum speed. (networkworld.com)
The research reported by Microsoft Azure’s Romsey Fibre R&D lab and University of Southampton teams presents a double nested anti‑resonant nodeless fiber (DNANF) geometry — an evolution of nested anti‑resonant and NANF designs — that achieves ultralow loss and wide operational bandwidth by minimizing three dominant loss mechanisms: confinement (leakage), surface scattering, and microbending‑induced coupling. The authors tested multi‑kilometre spools and present lab and pilot deployment data to support their claims. (tomshardware.com, eprints.soton.ac.uk)
Microsoft and related press reports state that the pilot program involved over 1,200 km of installed fiber carrying live traffic and that Microsoft has set an internal deployment target reported in presentations of 15,000 km of hollow‑core fiber across Azure’s backbone in its rollout plans. Those deployment figures are significant as company statements but should be interpreted carefully: the research paper documents the technical measurements and pilot tests; fleet‑wide deployment metrics cited in presentations are corporate operational targets and have not been independently audited in the academic manuscript. Treat the 15,000 km as a stated corporate plan rather than a peer‑reviewed technical result. (tomshardware.com, networkworld.com)
At the same time, turning a lab milestone into a globally interoperable, field‑proven fiber standard requires coordinated ecosystem work — mass manufacture with tight tolerances, splicing and connector standards, amplifier and transceiver alignment, long‑term environmental validation, and neutral third‑party route studies. Those are significant and measurable engineering tasks that mean adoption will be phased and targeted initially.
Network architects should treat hollow‑core as a strategic emerging technology to pilot now, purchase selectively for high‑value routes, and track maturity indicators (production‑reel attenuation guarantees, splice yields, OTDR behavior, independent field studies) before relying on it for mission‑critical backbone redundancy.
Source: Telecompaper Telecompaper
For roughly four decades, optical networking has been dominated by silica‑core single‑mode fiber (SMF). Improvements in silica attenuation have been incremental, and the practical floor for low‑loss SMF in the C‑band has generally been cited near 0.14–0.16 dB/km, a constraint that has shaped amplifier spacing, submarine cable economics, and transponder designs. Hollow‑core fibers (HCFs) aim to change that by guiding most optical energy through air rather than glass, reducing material scattering and nonlinear effects and bringing the effective group index closer to 1.0 — i.e., light travels nearer to vacuum speed. (networkworld.com)The research reported by Microsoft Azure’s Romsey Fibre R&D lab and University of Southampton teams presents a double nested anti‑resonant nodeless fiber (DNANF) geometry — an evolution of nested anti‑resonant and NANF designs — that achieves ultralow loss and wide operational bandwidth by minimizing three dominant loss mechanisms: confinement (leakage), surface scattering, and microbending‑induced coupling. The authors tested multi‑kilometre spools and present lab and pilot deployment data to support their claims. (tomshardware.com, eprints.soton.ac.uk)
What the measurements actually say
Key headline figures
- Minimum measured attenuation: 0.091 dB/km at 1,550 nm — the headline, record figure reported in the team’s published work. (phys.org, tomshardware.com)
- Broad low‑loss window: Loss below ~0.2 dB/km across a ~66 THz spectral window according to the paper and corroborating press coverage. (phys.org, tomshardware.com)
- Subband low loss: Multiple outlets report a contiguous subband where loss stays below 0.1 dB/km (often cited as ~18 THz in press summaries). This is a narrower, practically useful slice of the broader low‑loss window that network designers will focus on when planning coherent transmission systems. (networkworld.com, tomshardware.com)
- Latency / group‑index advantage: The measured design reduces the effective group index sufficiently to claim roughly 45–47% lower one‑way propagation delay compared to silica SMF, i.e., light traverses the hollow‑core mode substantially faster than it would through glass. This is a physical consequence of guiding light in air and is supported by the team’s modeling and experimental group‑index measurements. (azure.microsoft.com, phys.org)
- Dispersion: Reported chromatic dispersion is dramatically lower (authors and press often cite a ~7× reduction compared with legacy SMF in the tested regimes), a benefit for coherent transceivers and DSP complexity. (networkworld.com, phys.org)
Why these numbers matter in practical terms
Every fraction of a decibel saved per kilometer compounds over long links. A link at 0.091 dB/km accumulates far less total loss than a 0.14 dB/km silica span, which can directly reduce the number of inline amplifiers (or their pump power), lowering both CapEx and OpEx for long‑haul and subsea builds. The group‑index advantage directly reduces propagation delay at scale, which is relevant to latency‑sensitive verticals (financial trading, distributed AI training, AR/VR). The broader spectral window opens engineering choices outside the crowded C‑band and could enable higher per‑fiber capacity in the future — but only if amplifiers, lasers, and transceivers evolve to exploit that spectrum. (networkworld.com, phys.org)The physics and the design: why DNANF works
Nested anti‑resonance and the DNANF lineage
The DNANF design builds on a lineage of anti‑resonant and nested‑tube hollow‑core fibers that use thin, concentric silica tubes to create antiresonant reflections that confine light in the hollow central core. Key levers are:- Precise control of capillary wall thickness and tube spacing to tune anti‑resonant conditions.
- Avoiding "nodes" (contact points) between nested tubes to reduce leakage pathways.
- Minimizing glass surface roughness inside the structure to limit scattering loss.
These geometric controls reduce leakage, surface scattering and susceptibility to microbending — the three dominant contributors to HCF loss historically. The paper’s modelling and the team’s production controls are focused on keeping these fabrication tolerances over multi‑kilometre draws. (amp.spie.org, eprints.soton.ac.uk)
Measured mode quality and single‑mode guidance
The DNANF approach emphasizes mode purity — suppressing higher‑order modes so that links behave like single‑mode systems in real networks. Recent lab work in DNANF variants shows very high higher‑order‑mode attenuation and strong single‑mode performance over kilometres, which is critical to maintaining coherent transmission fidelity and low bit‑error performance. (arxiv.org)Deployment context: Microsoft, Lumenisity and Azure’s plans
Microsoft acquired Lumenisity (a University of Southampton spin‑out) in December 2022, gaining both IP and a Romsey, UK manufacturing facility — a strategic vertical step that places Microsoft on the supply chain side as well as the consumer-of‑fiber. Microsoft has framed hollow‑core fiber as an infrastructure enabler for AI‑era workloads and low‑latency Azure fabric improvements. (blogs.microsoft.com, datacenterdynamics.com)Microsoft and related press reports state that the pilot program involved over 1,200 km of installed fiber carrying live traffic and that Microsoft has set an internal deployment target reported in presentations of 15,000 km of hollow‑core fiber across Azure’s backbone in its rollout plans. Those deployment figures are significant as company statements but should be interpreted carefully: the research paper documents the technical measurements and pilot tests; fleet‑wide deployment metrics cited in presentations are corporate operational targets and have not been independently audited in the academic manuscript. Treat the 15,000 km as a stated corporate plan rather than a peer‑reviewed technical result. (tomshardware.com, networkworld.com)
Practical upside for networks and data centers
- Fewer inline amplifiers and lower power draw: Reduced per‑km loss compounds over long spans and can reduce the number or power of amplifiers required. This directly lowers both capital and operating costs on long‑haul routes. (networkworld.com)
- Lower latency per km: The group‑index advantage (nearing the speed of light in air) shortens propagation delay, especially valuable for time‑sensitive workloads and where every microsecond matters. (azure.microsoft.com)
- Broader spectral options: A 66 THz approximate low‑loss window widens the bands that engineers could exploit, potentially multiplying usable wavelengths beyond the congested C‑band — but only if amplifiers and transceivers follow. (phys.org)
- Reduced nonlinear penalties: For high‑power transport (e.g., laser delivery) and high‑capacity coherent systems, lower nonlinear interaction with glass is beneficial. That has niche scientific and industrial uses as well as telecom benefits. (arxiv.org)
Engineering and commercial hurdles (what remains to be proven)
The technical milestone is real, but the path to widespread operator adoption has several nontrivial barriers:- Manufacturing tolerance and scale. DNANF geometries require sub‑micron control of multiple dimensions along continuous multi‑kilometre draws. Scaling from lab reels to tens of thousands of production kilometres without reintroducing scattering sites or geometric drift is a major manufacturing challenge. (eprints.soton.ac.uk, networkworld.com)
- Bend sensitivity and field handling. Historically, HCF variants have been more bend‑sensitive than modern bend‑insensitive silica variants. Until bend performance matches field expectations for ducts, MDUs, and cabinet routing, many deployment classes remain unsuitable.
- Splicing, connectors and test ecosystems. The global fiber ecosystem — fusion splicers, OTDR profiles, connector form factors — is optimized for silica cores. HCF will need certified splicing recipes, new connector guidance, and updated test standards to ensure consistent yields and predictable OTDR signatures.
- Amplifier and subsystem integration. Long‑haul and subsea networks rely on established amplification ecosystems (EDFAs, Raman stages, gain‑flattening strategies). Exploiting the DNANF’s wide bandwidth and low loss in real links requires amplifier strategies and transceiver components validated end‑to‑end. This is not trivial and has cost implications for undersea economics.
- Long‑term reliability. Field cables face temperature swings, water ingress, mechanical compressions and rodent attacks. The hollow microstructures must be validated through long‑duration, independent field trials before operators will trust them for mission‑critical backbones. Multi‑year environmental aging and route studies are still needed.
- Standards and independent verification. Industry standards must be updated to define acceptance tests for HCF — attenuation on production reels, splice losses, bend loss specifications, and aging tests — and third‑party route audits are necessary for operator procurement confidence.
Timeline and realistic adoption scenarios
The research accelerates interest and reduces technological uncertainty, but broad commercial adoption will be phased and measured in years, not months. Reasonable expectations:- Short term (12–24 months): Controlled pilots, interoperability testing with transceiver and amplifier vendors, initial production reels for managed, hyperscaler‑owned links.
- Medium term (24–48 months): Wider vendor qualification for splicing/connectors, standardized test procedures, limited commercial deployments for high‑value routes (finance, selected long‑haul backbones).
- Long term (4+ years): Potential broader adoption if manufacturing scale, price per km, bend tolerance and environmental reliability meet incumbent expectations. Mass replacement of silica in ducts and buildings is unlikely until bend performance and cost parity improve.
Guidance for network architects and procurement teams
- Treat DNANF/HCF as an emerging, high‑value option rather than a drop‑in replacement. Plan hybrid architectures that reserve HCF for the most latency‑sensitive or long spans and retain silica for tight‑bend and legacy duct runs.
- Require vendor qualification data on production reel attenuation, splicing yields, OTDR signatures and environmental aging before placing route‑level orders.
- Run controlled pilots with vendor cooperation to collect splice yields, bend loss behavior, and amplifier interoperability statistics under operational conditions. Document everything for SLA negotiations.
- Engage early with transceiver and amplifier vendors to align component roadmaps to the intended bands if you plan to exploit the broader spectral window. Plan for multi‑vendor interoperability testing. (phys.org)
Balanced analysis: strengths, caveats and where to be cautious
Strengths- The 0.091 dB/km result is a genuine technical milestone that, if reproduced at scale and in field conditions, would represent one of the most consequential waveguide advances in decades: lower loss, lower latency and wider bandwidth potential than conventional silica SMF. (phys.org, tomshardware.com)
- The DNANF geometry addresses known leakage and scattering mechanisms with concrete engineering strategies and demonstrated multi‑kilometre spool testing. (eprints.soton.ac.uk)
- Lab and pilot results do not automatically equal production‑reel consistency. Scaling manufacturing while holding sub‑micron tolerances is costly and technically demanding. (eprints.soton.ac.uk)
- Reported deployment metrics (e.g., 1,200 km pilot live traffic, 15,000 km target) are operational claims tied to corporate programs and presentations; they are important signals but should be treated as corporate plans until independently audited in route‑level filings or neutral third‑party reports. (networkworld.com, tomshardware.com)
- Exploiting the broad 66 THz low‑loss window requires amplifiers, tunable lasers, and transceiver ecosystems that are not yet standard; systemic ecosystem shifts will take vendor alignment and time. (phys.org)
- Some widely quoted figures (notably the 15,000 km Azure rollout figure) appear in corporate presentations and press coverage; independent, route‑level confirmation is not present in the academic paper itself and remains a corporate deployment target rather than an audited installation statistic. Readers and procurement teams should treat such numbers cautiously.
Market and strategic implications
Microsoft’s path — acquiring Lumenisity, investing in the Romsey manufacturing facility, and integrating Azure engineering into R&D — is the kind of vertical approach a hyperscaler needs to secure supply, iterate rapidly and pilot new infrastructure end‑to‑end. If Azure can reliably field‑install DNANF at production yields and operationalize splicing and amplifier practices at scale, Microsoft could extract latency and capacity advantages directly valuable to cloud services and AI workloads. However, the broader telco and subsea market depends on multiple vendors, standardization work, and undersea economics — meaning other operators and cable consortiums will be watching, qualifying, and likely pursuing parallel development tracks. (blogs.microsoft.com, datacenterdynamics.com)The bottom line for WindowsForum readers and network practitioners
The DNANF hollow‑core result reported by Microsoft Azure and University of Southampton researchers is not incremental noise — it is a legitimate, physics‑rooted advance with clear practical benefits: lower loss at 1,550 nm (0.091 dB/km), wide low‑loss bandwidth, lower dispersion and near‑air group velocity. Those attributes make HCF a compelling tool for hyperscale backbones, latency‑sensitive point‑to‑point links and specialized scientific or high‑power applications. (phys.org, networkworld.com)At the same time, turning a lab milestone into a globally interoperable, field‑proven fiber standard requires coordinated ecosystem work — mass manufacture with tight tolerances, splicing and connector standards, amplifier and transceiver alignment, long‑term environmental validation, and neutral third‑party route studies. Those are significant and measurable engineering tasks that mean adoption will be phased and targeted initially.
Network architects should treat hollow‑core as a strategic emerging technology to pilot now, purchase selectively for high‑value routes, and track maturity indicators (production‑reel attenuation guarantees, splice yields, OTDR behavior, independent field studies) before relying on it for mission‑critical backbone redundancy.
Conclusion
This research marks a pivotal step in the decades‑long effort to move beyond the practical limits of silica‑core fiber. The DNANF design delivers measurable, repeatable laboratory evidence of lower attenuation, lower dispersion, and faster propagation — and Microsoft’s vertical play with Lumenisity gives the company a tangible runway to pilot and iterate in real networks. The result is both exciting and consequential: it widens the design space for long‑haul, hyperscale and latency‑sensitive networks while forcing the industry to confront a long list of deployment, standards and manufacturing challenges. The next 24–48 months will be decisive: controlled pilot deployments, cross‑vendor interoperability testing, and independent multi‑year field studies will determine whether hollow‑core technology becomes a premium niche or the backbone fabric of future long‑haul and AI‑era datacenter networks. (networkworld.com, phys.org)Source: Telecompaper Telecompaper