Summary

  • Twenty-four trunk fibre pairs at "over 26 Tbps" each reproduce ASN's "over 624 Tbps of total design capacity" exactly — which is the arithmetic that proves the headline total is a trunk-only product. The same release then adds four Sardinia–Rome pairs it never values and never adds.
  • The buyer's key-features line repeats 624 Tbps with "24 fiber pairs" as the system figure, two years and eight months after that same buyer printed 20 Tbps per pair and both route totals separately. The optical basis — band, channel count, bandwidth — appears in neither page.

Four pairs the total never counts

Alcatel Submarine Networks published the ready-for-service release for the Unitirreno system on 16 October 2025 with all three terms of the arithmetic within a few lines of each other. The 1,156-kilometre network connecting Sicily, Sardinia, Rome and Genoa is, in the release's own words, "delivering over 624 Tbps of total design capacity". A few sentences later: "The main Genoa–Mazara trunk provides 24 fiber pairs, each capable of over 26 Tbps." And immediately after that: "An additional 4 fiber pairs enhance the Sardinia–Rome link" (ASN release, 16 October 2025).

Twenty-four times twenty-six is 624 (ARITH-1, my multiplication, not theirs). The per-pair figure and the pair count reproduce the printed system total to the last terabit, which is a strong sign that the total is a trunk product and nothing more. The release never prints a per-pair value for the four Sardinia–Rome pairs, and never says whether the 624 includes them or excludes them. That is the absence of a reconciling sentence. It is not evidence that the total is wrong, that the branch is unbuilt, or that anyone misrepresented anything.

If those four pairs carried the same value as the trunk, the system would be twenty-eight pairs and over 728 Tbps (ARITH-2, mine). Nothing printed entitles a reader to that number, and I am not adopting it. It exists here for one purpose: to show what the unvalued pairs would add if a reader silently treated them as identical to the trunk. That silent substitution is exactly what a headline capacity figure invites.

The buyer's bullet, and the release it replaced

Unitirreno Submarine Network's own media release, same date, repeats the same three sentences, including the total and the trunk pair count (Unitirreno ready-for-service release, 16 October 2025). Then, in the key-features block a reader actually skims, it prints: "Over 624 Tbps capacity with 24 fiber pairs and cutting-edge XBU 24FP technology". The four additional pairs do not appear in that line. The line that sells the system names the pairs that fit the total and drops the pairs that do not.

Two years and eight months before that, in February 2023, the same buyer had done the opposite. Its contract-in-force release printed the basis first and the totals second: "At 20Tbps per fiber pair, the system's total capacity will be 480 Tbps on the trunk route and 320 Tbps on the branch to Rome-Fiumicino" (Unitirreno contract-in-force release, 1 February 2023). Divide the printed totals by the printed per-pair value and the pair counts fall out — 480 divided by 20 gives 24 pairs on the trunk, 320 divided by 20 gives 16 on the branch (ARITH-3 and ARITH-4, both mine; that release prints no pair counts).

Set the two printings side by side and the shape of the disclosure changes: forty pairs and 800 Tbps described route by route in 2023, against twenty-eight pairs and a single unqualified total in 2025 (ARITH-5). A design that moves between a pre-construction announcement and an as-built release is ordinary engineering. What moved in the printing is different. In 2023 the reader could check the arithmetic because both terms were on the page. In 2025 the reader can check it too — but only for the trunk, because the trunk is all the total counts. The per-pair figure itself rose from 20 Tbps to over 26 Tbps, a factor of 1.3 (ARITH-6, mine).

A second cable where the numerator is missing

The supplier does not print the same subset of terms every time. For the E2A system, in a release dated 24 March 2025, it prints the pair count and a total and no per-pair value at all: "12 fiber pairs (FP) delivering capacity of more than 192Tbps" (ASN release, 24 March 2025). Dividing 192 by 12 gives 16 Tbps per pair, which is my arithmetic and not the company's. On one cable the supplier prints both terms; on another it prints one. A reader who wants the per-pair figure has to supply the division herself, and a reader who wants to know what that figure is qualified by gets nothing on either page.

The basis is published in a third document, and it does not meet

The number a per-pair capacity figure is built from is not a capacity at all. It is bandwidth multiplied by spectral efficiency, with a great deal else folded in. ASN publishes those ingredients on separate pages, for separate purposes, in forms that cannot be multiplied together.

A product-overview page states the efficiency basis as a range: "Work in the optimum spectral efficiency of SLTE : 2-3 b/s/Hz and lower chromatic dispersion compensation", and prints the same range a second time further down (ASN SDM page). That page carries no capacity value, no optical band, no channel count, no symbol rate and no bandwidth figure.

A second page supplies a symbol rate and a per-wavelength capacity: the SMAT3 transponder, "achieving an operating speed of 130 Gbaud, facilitating next-generation coherent transport with capacities of up to 1.2 Tb/s per wavelength" (ASN upgrade page). The same two figures appear in a dated release from 15 March 2023, which also describes planned trials and prints no optical band and no channel count (ASN coherent-optics release, 15 March 2023).

A third document reports a field result on one named transatlantic cable: "achieving a remarkable 7.52 bit/s/Hz and a spectral efficiency of 6.17 bit/s/Hz if we take into account only the usable ethernet payload" (ASN Amitié release, 4 September 2025). That is two and a half times the top of the range printed on the product page, on a different cable, in a trial, and nothing I read reconciles the two. They are not two measurements of one thing and I do not average them.

The only document in the whole set that prints an optical band and a channel count together is legacy material published by Alcatel-Lucent in June 2014, mirrored on a third-party filing site rather than hosted by ASN: the 1620LM submarine line terminal equipment supports "up to 96 coherent channels on a 50 GHz grid, in the Extended C-band" (1620LM product information and planning guide, June 2014). Ninety-six channels on a 50 GHz grid is 4,800 GHz of optical spectrum, or 4.8 THz (ARITH-7, mine). Apply the supplier's own printed optimum range to that bandwidth and the result is 9.6 Tbps at 2 b/s/Hz and 14.4 Tbps at 3 b/s/Hz (ARITH-8, mine) — below the 26 Tbps per pair the Unitirreno release claims for a pair in 2025.

That shortfall is a fact about which terms are unpublished, not a finding that any figure is inflated. More spectrum, a better efficiency than the printed optimum, a higher per-wavelength capacity, or some combination of the three would close it. What it does show is the size of the hole a reader has to fill in: 26 Tbps divided by 1.2 Tbps per wavelength is roughly 21.7 wavelengths on a single pair (ARITH-9, mine), and no document I read says how many the equipment actually carries.

Hawaiki: 43 Tbps, and a denominator that cannot sit still

Unitirreno is a cable where the denominator is printed and the branch is not. Hawaiki is the case where the headline figure is easy to find and the denominator moves depending on who is talking.

Trade press reported the July 2018 ready-for-service with a capacity and no pair count: "The 15,000-km subsea cable links Australia, New Zealand, Pacific island nations, and the United States with 43 Tbps of capacity" (Lightwave Online, 23 July 2018). An architecture firm's landing-station page pairs the same headline with a count, scoped to one direction and one segment: "The Hawaiki submarine cable provides 43 terabits of carrier-neutral data capacity to Hawaii through six fiber pairs connecting the cable landing station in Kapolei to the U.S. west coast, Australia, New Zealand and American Samoa" (Bowers and Kubota). Forty-three divided by six is about 7.17 Tbps per pair (ARITH-10, mine).

An encyclopedia entry gives a different count and a different total in the same paragraph: "The cable has three fibre pairs across its main segments, and had an initial design capacity of 43.8 terabits per second", with a later increase to 67 Tbps recorded in the same entry (Hawaiki Cable entry). Forty-three point eight divided by three is 14.6 Tbps per pair (ARITH-11, mine).

Two third-party statements, three pairs and six pairs, and implied per-pair values that differ by a factor of two. Neither is adopted here. The operator's own current project page, in the text I read, prints routes, distances and latency and no design capacity and no pair count at all (BW Digital Hawaiki page). An absence from the passage I read is not proof the page never prints one. It does mean that on 19 November 2026 the two numbers a reader needs in order to check a Hawaiki headline were not both on the operator's page.

What a capacity figure has to print

A cable capacity figure becomes checkable when three things stand in the same document. First, the per-unit value with its scope stated: per fibre pair, per direction, trunk only, or system-wide. Second, the count over which it is multiplied, on the same scope. Third, the basis: the optical band, the channel count, the per-wavelength capacity and the efficiency figure, so that the per-unit value can be re-derived rather than taken on trust.

ASN printed the first two for Unitirreno, and the arithmetic closes, which is more than most of this industry offers. It printed the third nowhere near them, and the four branch pairs sit outside the total with no sentence to say so. The buyer printed the trunk figure as its key feature and let the four pairs drop. That is how a trunk product becomes a system total in public circulation: not by anyone lying, but by a number moving one document at a time.

A total without its scope is not a smaller truth. It is a different object wearing the same units, and it will sit in a sales deck, a financing model and a regulator's file, unchanged, for as long as the cable lasts.

Boundaries

  1. Every capacity figure in this article is a printed specification, not a measurement. I read no optical time-domain reflectometer trace, no lit-capacity report, no traffic figure and no sales record. Nothing here describes what any system carries, has carried or has sold.
  2. The two Unitirreno documents are a supplier press release and a buyer media release, both dated 16 October 2025, and both reached as text extraction through public search retrieval. The buyer's PDF was never downloaded and opened as a document, and neither release was read as a page image, so any qualification printed beside these figures is outside what I read.
  3. The February 2023 Unitirreno release carries the same reading limit and a further one: it is a pre-construction contract-in-force announcement, not an as-built configuration. A design that changes between 2023 and 2025 is ordinary.
  4. I obtained no ASN technical data sheet or product specification. What I read from the supplier is press releases and product-overview pages. The absence of an optical band, a channel count or a repeater spacing is an absence from the material I read on 19 November 2026, not a statement that the company publishes none anywhere.
  5. The 1620LM document is legacy material published by Alcatel-Lucent in June 2014 and mirrored on a third-party filing site. It was not published by Alcatel Submarine Networks and is not evidence about current practice. It is used for one purpose only: it is the only document in the set that prints a band and a channel count. The arithmetic built on it is mine.
  6. The 2–3 b/s/Hz range and the 7.52 bit/s/Hz result are printed by the same company on different pages for different purposes — a product-overview optimum and a field-trial result on one named cable. Nothing I read reconciles them. They are not two measurements of one thing and they are not averaged here.
  7. Hawaiki's 43 Tbps figure is trade-press reporting of a July 2018 ready-for-service announcement, not an operator filing. No operator-published Hawaiki document I read on 19 November 2026 prints a design capacity beside a fibre-pair count.
  8. The six-pair statement is an undated architecture firm's page, scoped to capacity "to Hawaii". The three-pair statement is an encyclopedia entry. Both are third-party, they disagree, and neither is adopted. The implied values of about 7.17 and 14.6 Tbps per pair are my arithmetic on two third-party statements and nothing more.
  9. The 624 Tbps total and the four additional Sardinia–Rome pairs coexist in one document without a sentence reconciling them. That is the absence of a sentence. It is not a finding that the total is wrong, that the branch is unbuilt, or that anything was misrepresented.
  10. "Total design capacity", "capacity", "delivering" and "capable of" are the publishers' own words, reported as printed. No conversion is made anywhere between design capacity and lit, equipped, sellable or revenue-bearing capacity, and no equivalence between them is implied.
  11. Every Tbps value is reported at the printed precision, including its "over" and "more than" qualifiers. Nothing is rounded and no qualified figure is treated as exact.
  12. The trunk and branch pair counts of twenty-four and sixteen derived from the 2023 release are mine, produced by division from printed totals and a printed per-pair value. That release prints no pair counts.
  13. No Internet-number-registry claim is made here. This piece is about optical specification and publishes no address block, AS number, prefix or registry object.
  14. Prior coverage named rather than repeated: BTW's Research piece of 7 November 2026 on the French state's purchase of a majority of this business, and its briefing of 18 October 2026 on a cleared owner and an unmeasured fleet. Neither read a capacity specification and neither multiplied a per-pair figure by a pair count.
  15. Nothing here asserts that any capacity figure is inflated, that any system underperforms its design, that any buyer or investor was misled, or that any company, officer or employee acted improperly. The finding concerns which terms a public document prints and which it leaves the reader to supply.
  16. All documents were read on 19 November 2026. A later revision of any release, or a specification published since, is outside what I read.