Microsoft and the University of Southampton’s Optoelectronics Research Centre have shattered a decades-old attenuation barrier in fiber optics, revealing a hollow-core fiber design that posts a verified loss of just 0.091 dB per kilometer at the critical 1550 nm wavelength. The work, accepted for publication in Nature Photonics and detailed in a March 27 arXiv preprint, leapfrogs the roughly 0.14 dB/km practical floor that silica glass single-mode fiber has been stuck at for more than 30 years. In a single stroke, the result reopens design possibilities for long-haul and data-center networks that had long been considered off-limits.

The team, which includes co-authors from Microsoft’s Azure networking organization, achieved the milestone by replacing silica’s glass core with a meticulously engineered air-filled microstructured tube. Light guided through air dodges most of the material scattering and absorption that plague standard fiber, traveling roughly 50% faster and suffering far fewer nonlinear penalties at high power. The payoff is not only lower signal loss per kilometer but also a broadband low-loss window spanning 66 THz—more than double the usable spectrum of conventional single-mode fiber—and a dramatic reduction in chromatic dispersion.

Record-Breaking Numbers

The new hollow-core fiber’s performance stands in stark contrast to the industry status quo. Standard single-mode silica fiber bottoms out at approximately 0.14–0.16 dB/km at 1550 nm, a limit dictated by fundamental Rayleigh scattering and infrared absorption in the glass. The 0.091 dB/km measured in the preprint represents a 35% reduction over that floor.

Equally important is the usable bandwidth. The team reports that attenuation remains below 0.2 dB/km over a 66 THz spectral window—against silica’s typical 26 THz—and dips under 0.1 dB/km across an 18 THz sub-band, as confirmed by Network World’s review of the characterization data. That extra headroom could allow network operators to activate wavelength bands far beyond the crowded C-band, multiplying capacity without laying new cables.

These improvements compound rapidly over distance. A 1,000 km span in hollow-core fiber would suffer only about 91 dB of loss, versus 140 dB or more over the same length of modern silica fiber. In practice, that means fewer inline amplifiers, lower pump power, and substantially reduced energy bills—a direct line to both capital expenditure (CapEx) and operational expenditure (OpEx) savings. Microsoft itself frames the technology as a way to slash the number of amplifier sites needed for Azure regional rings and backbone routes.

How Hollow Core Fiber Works

The fiber is a descendant of a design lineage that includes Kagome, single-ring tubular, and nested anti-resonant nodeless fibers. In this latest iteration, a series of nested, non-touching glass capillaries surround an empty central core. Each capillary’s wall thickness is tuned to create anti-resonant reflections that keep light tightly confined to the air core. By optimizing the nested geometry and capillary dimensions, the researchers minimized three dominant loss mechanisms: leakage into the cladding, surface scattering from microscopic roughness on the glass walls, and microbend-induced coupling to cladding modes.

Because more than 99% of the optical power travels in air, the overall group index approaches 1.0. That translates directly into a speed advantage over silica (index ~1.47). Microsoft’s Azure blog and the preprint both quote a 50% increase in transmission speed. In practical terms, a one-way trip from London to Tokyo would see propagation latency drop by roughly 30 milliseconds, an eternity for high-frequency trading platforms or distributed AI training clusters.

Microsoft’s Strategic Bet

The results are no mere academic exercise. Microsoft acquired Lumenisity, a University of Southampton spin-out that pioneered commercial hollow-core fiber, in December 2022. That deal handed the company a 23,000-square-foot manufacturing facility in Romsey, UK, a trove of intellectual property, and a direct line to the ORC’s deep bench of photonics talent. Since then, Microsoft has invested in scaling production and has co-authored the breakthrough paper with ORC researchers.

Trade outlets report that Microsoft has already installed over 1,200 km of hollow-core fiber carrying live Azure traffic, and that the company announced a target of 15,000 km at an internal conference. These figures come from corporate briefings, not independently audited field surveys, so they represent ambitious deployment goals rather than confirmed installations. Still, they signal an aggressive push. For context, subsea cables like Marea or Dunant run thousands of kilometers; a 15,000 km hollow-core build would be a transformative backbone investment.

The Caveats No One Should Ignore

Lab records are necessary, not sufficient. Several engineering and commercial hurdles stand between this preprint and a fiber that a telecom operator would trust to carry live traffic.

Manufacturing Tolerance. The nested anti-resonant design requires sub-micron precision in capillary wall thickness and lateral geometry. Maintaining that uniformity over 50 km or 100 km reels demands process control beyond what standard silica fiber draws require. The team’s admission that removing impurities expanded the low-loss window underscores the sensitivity. Even slight gas contamination or dimensional drift during the draw can spike loss figures.

Bend Sensitivity. Hollow-core fibers have historically been more sensitive to tight bends than the best bend-insensitive solid-core fibers. Deployment in legacy ducts, building risers, and fiber-to-the-home enclosures would force installers to manage minimum bend radii carefully. Unless bend tolerance improves, applications that demand sharp turns—like last-mile delivery or multi-dwelling-unit wiring—are off the table for now.

Splicing and Connectors. The global fiber ecosystem is optimized for solid silica cores. Fusion splicers, field connectors, and optical time-domain reflectometers (OTDRs) rely on well-known insertion and reflection characteristics. Hollow-core fibers demand tailored splicing recipes, specialized connector designs, and new test procedures. Until splicing yield and connector reliability are proven at scale, operating expense could rise even as amplifier counts fall.

Amplification Strategy. Erbium-doped fiber amplifiers (EDFAs) are tuned for silica’s C-band. While hollow-core fiber can operate in the same band, the physics of coupling pump light into an air-core amplifier differs. The broad 66 THz window could be paired with Raman amplifiers, semiconductor optical amplifiers, or parametric amplifiers, but that ecosystem does not yet exist. For subsea and ultra-long-haul routes, the amplifier question remains open.

Long-Term Reliability. Field cables face temperature swings, moisture, mechanical stress, and rodent attacks. The hollow-core microstructure must remain stable under these conditions for 25 years or more—a timeline that demands extensive accelerated aging tests and third-party qualification. Lab attenuation at 20°C is only a starting point.

Where Hollow Core Will Strike First

Given the current realities, hollow-core fiber’s initial commercial beachheads are clear. Hyperscale cloud operators and financial trading firms can engineer end-to-end physical plants that accommodate bend constraints and splice requirements, and they value latency reductions enough to pay a premium. Microsoft’s own Azure backbone exemplifies this model. Short-reach data-center interconnects, where the operator controls cable trays and handling from server row to server row, are another natural fit.

Specialized applications such as high-power laser delivery and distributed sensing also benefit immediately. Because air-core fibers avoid nonlinear glass effects like stimulated Brillouin scattering, they can handle higher peak powers without signal degradation—an attractive feature for industrial and scientific customers.

Broader adoption in metro rings, cellular mid-haul, and fiber-to-the-premises will depend on achieving bend losses competitive with ITU-T G.657 fiber and driving per-kilometer costs down through volume manufacturing. Industry analysts expect that hybrid architectures, mixing hollow-core on long, straight spans and retaining silica for highly bendy endpoints, will dominate the transition phase.

The Road Ahead: Standards, Ecosystem, and Timeline

The preprint accelerates industry interest but does not eliminate the sequential steps required for commercial deployment.

  • Standards bodies (ITU-T, IEC) must define test procedures for hollow-core attenuation, bend loss, splice loss, and environmental aging that map to existing carrier acceptance criteria.
  • Transceiver and amplifier vendors need to publish interoperability guides that specify splicing protocols, connector types, and power budgets.
  • Neutral third parties must conduct long-duration field trials—months to years—to validate the 0.091 dB/km figure under real mechanical, thermal, and contaminant conditions.
  • Splicer manufacturers, cable makers, and installer workforces require certified processes and quality-assurance tooling to deliver consistent per-kilometer performance without prohibitive overhead.

These steps are not optional. Without them, even a Nobel-worthy fiber will remain a niche curiosity. Network architects and procurement teams should therefore track vendor qualification data, splicing recipes, and independent field studies before specifying hollow-core fiber on production-critical paths. Engaging early with suppliers on interoperability testing will also let organizations bake the new acceptance and warranty metrics into their procurement terms.

A Milestone with Measured Optimism

The 0.091 dB/km result is a genuine technical milestone—concrete, testable, and peer-reviewed. It demonstrates that hollow-core fiber can outperform the practical loss floor of silica while delivering latency, bandwidth, and dispersion benefits that conventional fiber cannot match. If the performance persists in production reels and field routes, the implications for long-haul and data-center networks are immense: fewer amplifiers, lower energy consumption, wider spectral choices, and latency slashed by half.

Yet practical deployment timelines are measured in years, not months. Manufacturing scale-up, bend performance, splice/connector ecosystems, amplifier integration, and multi-year reliability testing remain real bottlenecks. Early adoption will cluster in environments where the operator controls the installation and absorbs premium costs—Azure backbones, trading links, controlled data-center interconnects. The next 12 to 36 months will determine whether hollow-core fiber evolves from a premium niche into the backbone fabric of the AI-era cloud.

For Microsoft and the University of Southampton, the chip is on the table. The preprint’s closing line—that this work “potentially heralds a new era in long-distance communications”—is bold but grounded. Now the hard, unglamorous work of turning a 0.091 dB/km lab reading into a reliable, cost-effective, and widely deployable fiber begins in earnest.