Request an intervention

Operational notes Engineering

Hollow-core fibre: 90 km instead of 60 between two data centres, and what changes in the specification

7 min read

A bundle of round metal tubes seen end-on, stacked and slightly staggered, in black and white
The advantage is the emptiness in the middle: that is where light runs faster than it does in glass.

A new line is starting to appear in data centre interconnect specifications: hollow-core fibre, distance up to 90 km. The number is real, it has a precise source and it is quoted correctly. It is simply not a field measurement: it is a multiplication. And once it lands in a specification written in the indicative, everything around it — acceptance testing, instruments, budgets — is one step behind.

The fact, with its particulars.

On 22 July 2026 Prysmian and Relativity Networks announced the highest-density hollow-core cable the pair has built to date: 24 ChronoCore hollow-core fibres in a 10 mm diameter cable, Sirocco architecture, manufactured at the Claremont plant in North Carolina. The tests took place at Dura-Line’s flagship test facility in Clinton, Tennessee, into microducts, at jetting speeds of up to 350 ft/min — around 107 m/min. The release adds that fibre production is being scaled at the Eindhoven facility as well.

Two sentences deserve a slow read. First: the release describes installation tests at a test facility, not a route in service. Second, on attenuation: the cable built at Claremont, it says, meets the customer’s attenuation requirements. That is a private contractual requirement. No dB/km figure is published, and that is the number you need to close an optical budget over a long span.

Where the “90 km instead of 60” comes from.

The release says it itself, in the descriptive note at the end. In a hollow-core fibre a glass waveguide guides light through an air-filled core; because the speed of light in air is greater than in glass, hollow-core fibre transmits “nearly 50% faster” than conventional fibre. Then the application: the transmission speed of glass fibres limits the distance between data centres in a synchronised cluster to 60 km, and hollow core could allow that to be extended to 90 km (56 miles).

Put the two numbers side by side. Nearly 50% more, applied to 60, gives 90. It is not the outcome of a route that was built and tested: it is 60 multiplied by the ratio of the two propagation speeds, and the verb is conditional. That is not a flaw in the release — it is its honesty. The flaw comes later, when the conditional quietly disappears.

The 60 km limit is a time budget, not an optical budget.

This is the part most often lost. The release does not say light fails to arrive at 60 km: it says synchronisation between two data centres in the same cluster stops at 60 km. That is a delay constraint, not a received-power one. Hollow core acts on it, because it cuts propagation time for the same length. Two checks follow, before anything gets signed.

Which constraint actually binds your design? If it is the time budget, the reasoning holds. If the optical budget or the reach of the transponders binds first, hollow core does not move that limit — and the attenuation of the fibre, as noted, is not published. On coherent spans the bottleneck moved years ago to OSNR and PMD, and on existing fibre that is settled by measuring, not by inference.

Which length are you doing the arithmetic on? Delay is proportional to the real length of the fibre, not to the distance on the drawing. Indirect routes, slack in chambers, loops at joints, entries into the room: those are real kilometres, and they are paid for in time like every other one.

Every instrument that measures a distance has to be reconfigured.

Here the specification touches acceptance testing, and the consequence is concrete. ITU-T G.650.1 (01/2024), in the backscattering method, gives the formula for deriving the length of a fibre from the trace: L = c × T / N, where T is the time interval between the two ends of the backscattering curve, c is the free-space speed of light and N is the group index of the fibre. That is the arithmetic inside every OTDR: the instrument measures times and converts them into metres using a constant that you supplied.

Change the medium and the constant changes. By how much? By the same ratio that earns you the latency. If the declared ratio is around 1.5, an event genuinely 30 km out reads at roughly 20 km on an instrument still set for glass. This is not the few metres argued over when choosing a pulse width: it is a third of the route, and it gets discovered with a digger.

The right value is not in a recommendation. ITU-T G.652 (08/2024) lists the attributes of single-mode glass fibre — mode field diameter, cladding diameter and non-circularity, cut-off wavelength, macrobending loss, proof stress, chromatic dispersion, attenuation, PMD — and the group index is not among them; it states, in fact, that the refractive index profile of the fibre does not generally need to be known. For glass the figure is so stable that it sits pre-loaded in the instruments and nobody argues about it. For hollow core it has to be requested from the manufacturer, written into the specification and printed on every report — otherwise you get a set of distances that agree with each other and are all wrong by the same factor.

What to ask for, at tender and at acceptance.

  1. The group index declared by the manufacturer, for each operating wavelength, and an obligation to set it on the instruments; the value used goes on the test report, not left at the default.
  2. Attenuation in dB/km as a supply figure, at the wavelengths you will actually use. “Meets the customer’s requirements” is not a figure: it is a pointer to a contract you do not hold.
  3. An explicit time budget: how many microseconds, one-way or round trip, how much belongs to the cable and how much to the equipment, and over which fibre length — the routed one, not the one on the map.
  4. Latency measured on the delivered route, with the method and reference stated, not calculated from the kilometre count. It is the only evidence that separates a press-release figure from a test figure.
  5. The transition to glass: where it happens, with which component, to which acceptance value and by whom. Transponders terminate in conventional fibre: that crossing point is a design item, not a site detail.
  6. Test methods named in full. ITU-T G.650.3 collects the methods for installed single-mode links and assumes the fibre families of the G.65x recommendations: which method applies to hollow core, and with which parameters, belongs in writing.
  7. Installation conditions verified on your route. The 350 ft/min came from a test facility, with those microducts; on a real route what counts is bends, joints and the cable’s pulling limits.

The point.

Hollow-core fibre is leaving the laboratory: a 24-fibre cable manufactured in series and installed at industrial speeds is no small thing. But the latency gain is only banked if the route is designed for it and the acceptance test is rewritten to match — group index, supply attributes, end-to-end time measurement. That is the work: route design and certification defined together, with traces and native data kept wherever suits you, on-premise in your own environment or on a dedicated cloud — an environment reserved to the single client, dedicated VPN, data centres in Italy, in premises staffed directly. If the prior question is who owns the infrastructure those kilometres connect, we have written about that separately.

Have an interconnect between data halls to specify, or a quotation with “hollow core” in it to assess? Talk to an engineer: agreeing up front which numbers are declared and which will be measured costs far less than a route that meets the specification and misses the time budget.

Sources