1. Introduction

An anti-resonant hollow-core fiber measured at 0.091 dB/km at 1550 nm sits below the 0.1397 dB/km record held by solid silica, and both numbers are measured attenuations on drawn fiber rather than projections (measured, Nature Photonics 2025 for the hollow-core value; measured, OFC 2024 Tu2E.1 for the silica value). That single comparison is the reason a three-way media question exists at all in 2026. For four decades the selection problem was internal to one family: G.652 or G.654, larger effective area or smaller, and the answer changed the link budget by tenths of a decibel. The question now spans three physically different waveguides, and the differences between them are measured in factors rather than percentages.

Air guidance removes the glass from the optical path. The guided mode of a double-nested anti-resonant nodeless fiber overlaps the silica membranes by less than a part in a thousand, so the Kerr nonlinear coefficient falls by roughly three orders of magnitude, the group index falls to about 1.0027, and the Rayleigh scattering floor that bounds solid silica no longer applies. Standard single-mode fiber gives up all three of those advantages and returns something no laboratory result can match: several billion kilometres already in the ground, an installed base of transceivers, splicers, optical time-domain reflectometers, connectors and acceptance procedures, and a complete set of published ITU-T fiber parameter specifications that make a span buildable by two vendors who have never spoken. Uncoupled multi-core fiber answers a third constraint entirely: it multiplies the number of spatial channels inside one 125 µm cladding, which matters when the binding limit is duct cross-section rather than decibels or microseconds.

The three media are therefore not ranked. Each relieves a different binding constraint, and a route whose limit is propagation delay reaches a different answer from a route whose limit is conduit occupancy, even when both routes run between the same two buildings. A high-frequency trading link between a matching engine and a colocation cage is bounded by microseconds; a metro span between two availability zones is bounded by the distance at which the round-trip budget still permits synchronous replication; a submarine cable is bounded by the electrical power that can be fed down the copper conductor; and a hyperscale campus interconnect is frequently bounded by the number of strands that fit through an existing 7/3.5 mm microduct. Four constraints, three media, and no universal ordering between them.

Scope and Method

This article compares anti-resonant hollow-core fiber, ITU-T G.652.D standard single-mode fiber and weakly coupled uncoupled multi-core fiber across eleven axes: effective group index, attenuation, propagation delay, chromatic and polarization dispersion, nonlinearity and launch power, capacity per fiber and per cladding, cable geometry and spatial density, splicing and interface loss, amplification and line-system architecture, monitoring and fault location, standards position, and installed cost. Coupled-core multi-core fiber and few-mode fiber appear only where the contrast clarifies an uncoupled result, because both require full multiple-input multiple-output digital signal processing at the receiver and belong to a different system class. Photonic bandgap hollow-core fiber appears in the historical section and then drops out, since anti-resonant designs carry every current telecom result.

Every figure in the article carries its evidence class in the same sentence: standard-specified, measured, vendor claim or theoretical limit. That discipline matters more here than in most comparisons, because the hollow-core literature quotes laboratory attenuations on short spooled samples alongside cabled field losses that are several times higher, and reading the two as one number produces span budgets that will not close. Where a laboratory value and a field value diverge, both appear.

2. Effective Group Index and Modal Power Overlap

Effective group index is the dimensionless factor by which a guided optical pulse travels slower than light in vacuum, written n_g and defined as the ratio of the vacuum speed of light to the group velocity of the mode. It fixes propagation delay directly, it is a property of the guided mode rather than of the core material alone, and it is measured on drawn fiber rather than calculated from a glass composition.

2.1 Distinctions From the Adjacent Quantities

Four quantities are routinely used as if they were one, and each substitution changes an answer.

Refractive index against effective index. The refractive index n of bulk silica near 1550 nm is about 1.444, and the effective index n_eff of the fundamental mode of a G.652.D fiber is about 1.447, because the mode samples core and cladding together weighted by its own transverse field. Neither of those is the number that belongs in a latency budget: the group index of the same mode is 1.4682, and substituting either index value for it understates delay by 1.4 to 1.7 %.

Effective index against group index. The effective index sets the phase velocity; the group index sets the velocity of the envelope, and therefore the delay of data. They differ by the material and waveguide dispersion terms, and only the group index belongs in a latency budget.

Group index against fraction of power in glass. In a hollow-core fiber the two are related but not interchangeable. A mode with less than 0.1 % of its power in the silica membranes carries a group index near 1.0027, which is what produces the delay advantage; the same overlap figure also sets the Kerr nonlinearity and the thermal coefficient of delay, which the group index alone does not describe. The overlap fraction η is the quantity that explains why one fiber wins on three separate axes at once.

Core diameter against mode field diameter. In hollow-core fiber the mode field diameter runs at roughly 70 % of the core diameter, so a 25 µm core presents an 18 µm mode field. That mismatch against the 10.4 µm mode field of standard fiber at 1550 nm is the whole of the interface problem treated in Section 9, and quoting the core diameter in its place understates the coupling loss.

2.2 Units and Conversion Arithmetic