Researchers at Microsoft Azure's Fibre R&D lab, collaborating with the University of Southampton and spin-out Lumenisity, have demonstrated a hollow-core optical fiber with a record-low attenuation of just 0.091 dB/km at 1,550 nm. Published in Nature Photonics, the breakthrough marks one of the most consequential advances in waveguide physics in decades, promising to slash latency and energy consumption across hyperscale networks, submarine cables, and AI-driven data centers.
The fiber, a double nested anti-resonant nodeless fiber (DNANF), guides light through an air-filled core instead of solid silica. By almost eliminating material scattering and nonlinear effects, the design achieves near-vacuum light speed—cutting propagation delays by up to 47% compared to conventional single-mode fiber. For network architects, the implications are profound: longer spans without repeaters, dramatically expanded spectral capacity, and latencies that could enable new classes of distributed computing.
The Physics Behind the Record
For forty years, silica-core single-mode fiber (SMF) has been the backbone of global communications. The practical attenuation floor for SMF hovers around 0.14–0.16 dB/km at 1,550 nm, a constraint imposed by Rayleigh scattering within the glass. Hollow-core fibers (HCF) circumvent this by confining over 99% of the optical power in air. However, early HCF designs struggled with high leakage and surface scattering losses.
The DNANF geometry evolved from nested anti-resonant nodeless fiber (NANF) concepts. It employs a lattice of ultrathin, precisely spaced silica capillaries that create antiresonant reflections, trapping light within the hollow core while suppressing higher-order modes. Three critical design levers drive the record performance:
- Capillary wall thickness and spacing are tuned to exact anti-resonant conditions, minimizing confinement loss.
- Node-free construction eliminates contact points between nested tubes, reducing leakage pathways.
- Surface roughness inside the microstructures has been tightly controlled to limit scattering loss to just a few percent of the total attenuation.
Modeling and multi-kilometer spool tests confirm that the fiber supports robust single-mode operation—essential for coherent transmission—while achieving the headline 0.091 dB/km figure. This is not merely a lab curiosity: the team has already installed over 1,200 km of hollow-core fiber carrying live Azure traffic, with production reels drawn at the Romsey, UK facility Microsoft acquired with Lumenisity in December 2022.
Key Performance Metrics
The DNANF breakthrough delivers advantages across every dimension network engineers care about:
| Metric | DNANF Result | Silica SMF Comparison |
|---|---|---|
| Minimum attenuation at 1550 nm | 0.091 dB/km | ~0.14–0.16 dB/km |
| Low-loss bandwidth | 66 THz (loss <0.2 dB/km) | C-band (~4 THz) |
| Sub‑0.1 dB/km window | ~18 THz | Not available |
| Latency (group index) | ~1.00 | ~1.47 (glass core) |
| Chromatic dispersion | ~7× lower than SMF-28 | Standard values |
A 66 THz low-loss window represents a seismic shift. Conventional dense wavelength-division multiplexing (DWDM) systems crowd into the C- and L-bands, but the DNANF fiber opens up the E-, S-, and U‑bands for exploitation. If amplifier and transceiver ecosystems evolve to match, a single fiber could carry orders of magnitude more traffic, dramatically reducing the cost per bit.
Latency benefits are equally game-changing. Light in a hollow core travels 47% faster than in silica, slashing one-way propagation delay. For financial trading floors, distributed AI training, and real-time augmented reality render farms, every microsecond saved translates directly into competitive advantage or smoother user experiences.
Why This Matters for Data Centers and Long-Haul Networks
Attenuation compounds over distance. A 200 km link built with 0.14 dB/km silica fiber accumulates 28 dB loss, while the same span at 0.091 dB/km drops to just 18.2 dB. That nearly 10 dB savings can eliminate entire amplifier sites, trim power budgets, and simplify network design. In submarine cables spanning thousands of kilometers, the capex and opex implications are staggering.
Dispersion management also becomes far simpler. The DNANF's inherently low chromatic dispersion—about seven times less than standard SMF-28—reduces the digital signal processing burden on coherent receivers, potentially allowing for higher baud rates and lower power consumption per channel.
Nonlinear penalties, which plague high-power transmission in glass fibers, are virtually absent in the air-filled core. This feature not only benefits communication systems but also opens scientific applications, such as delivery of high-power laser pulses for industrial or medical use.
Microsoft’s Vertical Strategy
Microsoft’s acquisition of Lumenisity was not a speculative venture. By bringing the Romsey manufacturing line in-house, the hyperscaler controls the entire hollow-core supply chain—from preform assembly to installation—enabling rapid iteration that third-party fiber vendors cannot match. Azure’s backbone has already seen pilot deployments exceeding 1,200 km with hollow-core links carrying production workloads. Internal targets point toward 15,000 km of deployed hollow-core fiber across Azure’s global network, a figure shared in corporate presentations.
Vertical integration allows Microsoft to tailor fiber specifications precisely to Azure’s requirements: matching amplifier spacing to hollow-core span lengths, developing proprietary splicing recipes, and qualifying the fiber under real‑world mechanical stresses before offering it in public cloud regions. The hyperscaler can then market low‑latency “speed‑of‑light” connectivity between its data centers as a premium service, differentiating Azure in the competitive cloud market.
Engineering and Commercial Hurdles
For all its promise, DNANF faces a gauntlet of manufacturing and operational challenges before it can unseat silica fiber on a mass scale.
Manufacturing consistency: Sub‑micron control over capillary dimensions must be maintained across hundreds of kilometers of continuous draw. Any drift induces additional loss or mode‑profile distortions, and the current yield rates from the Romsey facility remain a closely guarded metric. Scaling to the tens of thousands of kilometers required for metro and long‑haul builds will demand a Moore’s‑Law‑like improvement in process control.
Bend sensitivity: Hollow‑core fibers have historically struggled with macrobending losses. Even modest bends of a few millimeters radius can scatter light out of the core. While recent advances have improved bend‑tolerant designs, DNANF must still prove it can survive the twists, turns, and compressive loads of terrestrial duct installation and building riser pathways, where bend-insensitive silica fibers already perform reliably.
Splicing and connector ecosystems: The global fiber infrastructure—fusion splicers, OTDRs, connector end-face geometries—is optimized for solid‑core glass. DNANF splicing requires different arc discharge parameters, and the air core can collapse if heated improperly. New connector designs must seal the fiber end to prevent moisture ingress while maintaining optical alignment. Standardized procedures and certification programs are nascent at best.
Amplifier and subsystem integration: Exploiting the full 66 THz low‑loss window demands amplifiers that operate far outside the standard C‑band. While thulium‑doped and bismuth‑doped amplifiers are maturing, they are not yet cost‑competitive with erbium‑doped fiber amplifiers (EDFAs). Raman amplification may fill some gaps, but mixing multiple amplifier technologies in a single link adds complexity and cost. Coherent transceivers must also be tuned for the specific dispersion and nonlinear characteristics of DNANF, a software and firmware task that will require vendor‑hyperscaler collaboration.
Long‑term reliability: Field‑deployed fibers must endure temperature cycling, humidity, mechanical compression, vibration, and rodent attacks. The hollow microstructures must remain hermetically sealed over decades; any crack could allow water molecules to condense inside the core, dramatically increasing loss. Independent, multi‑year environmental testing—sinking loops on ocean floors, burying reels in swampy ducts—is essential before carriers entrust mission‑critical traffic to hollow‑core lines.
Standards and third‑party verification: ITU‑T G.652 recommendations define the baseline for silica SMF. No equivalent standard exists for hollow‑core fiber. Operators will demand consensus‑based attenuation and bend‑loss specifications, standardized OTDR event signatures, and route‑level proof‑of‑performance audited by neutral testing labs. Until such benchmarks are in place, large‑scale tender processes will remain biased toward incumbent fiber types.
Realistic Adoption Timeline
Hollow‑core technology will not flood the market overnight. Instead, expect a phased rollout:
- 2024–2025: Controlled pilots within hyperscaler and specialized financial networks. Interoperability testing with major transceiver and amplifier vendors. First commercial‑grade production reels with refined bend performance.
- 2026–2027: Wider vendor qualification for splicing equipment and connectors. Standardized test procedures published by ITU‑T or IEC. Limited commercial deployments on high‑value routes where latency or capacity premiums justify the cost.
- 2028 and beyond: Broad adoption possible if manufacturing cost curves bend downward, bend performance matches field requirements, and multi‑vendor amplifier solutions stabilize. Replacement of legacy silica in existing ducts will proceed selectively, likely starting with new greenfield builds before touching brownfield campus or metro networks.
Early wins will cluster where the network owner controls the physical plant and can capture the latency benefit directly—hyperscaler private interconnects, stock exchange links, and high‑frequency trading campuses. Telecom operators are likely to follow once third‑party verification and competitive pricing emerge.
Guidance for Network Architects
Practical steps for teams monitoring this technology:
- Treat DNANF as a premium overlay, not a drop‑in replacement. Hybrid architectures that keep hollow‑core for the longest, most latency‑sensitive spans while retaining bend‑tolerant silica for distribution and last‑mile segments will minimize risk.
- Demand production‑reel attenuation guarantees, splice‑yield data, and environmental aging reports before committing to route‑level purchases. Ask for OTDR signature datasets specific to your chosen fiber length.
- Run controlled pilots with vendor cooperation to measure real‑world splice losses, bend‑radius limits, and amplifier interoperability. Use these results to negotiate service‑level agreements.
- Engage transceiver and amplifier roadmaps early. If you intend to exploit the extended spectrum, align with suppliers on tunable laser, coherent ASIC, and amplification paths well before cable is pulled.
Market and Strategic Implications
Microsoft’s vertical integration gives Azure a first‑mover advantage in latency‑sensitive cloud services. If hollow‑core fiber delivers on its promises, Azure could offer “express lanes” between regions that no competitor can match without similar infrastructure. The broader telecom and submarine cable industry will watch closely but remain cautious, relying on a multi‑vendor ecosystem and standards bodies to validate the technology before large‑scale procurement.
Other hyperscalers will almost certainly pursue their own hollow‑core strategies—either through partnerships with alternative manufacturers or by developing in‑house R&D. The resulting competition could accelerate cost reduction and innovation, much like the fierce transceiver price wars of the past decade.
The Bottom Line
The 0.091 dB/km result is not incremental noise; it is a legitimate, physics‑rooted breakthrough with clear practical payoffs: lower loss, wider bandwidth, and near‑vacuum latency. Microsoft’s aggressive deployment targets signal confidence that hollow‑core can move from lab curiosity to commercial reality.
Yet the gap between a record‑setting spool and a carrier‑grade link spanning thousands of kilometers remains substantial. Manufacturing tolerances, field handling, amplifier ecosystems, and multi‑year reliability data are still works in progress. Network architects should view DNANF as a high‑potential tool for specific pain points—long‑haul hyperscale backbones, low‑latency trading connections—rather than a universal replacement for silica glass.
The next 24 to 48 months will be decisive. Controlled pilots, cross‑vendor interoperability testing, and independent field studies will determine whether hollow‑core technology claims a premium niche or evolves into the backbone fabric of the AI‑era internet.