BRENNEN

Industries / Technology / Telecommunications / Fiber optics

Making fiber

Making fiber is the work of turning purified chemicals into large glass rods called preforms, drawing those rods into coated strands thinner than a hair, and selling the fiber to standards that cable makers can buy against.

7

Milestones achieved, from 1974 to 2026.

2

To watch for: announced, not yet confirmed outside.

3

Open deliverables. Nobody has shipped them.

The road

AchievedWatch forOpen

Achieved

  1. 1974

    Glass grown from vapor

    At Bell Labs, John MacChesney and P. B. O'Connor invented modified chemical vapor deposition (MCVD), in which very pure glass forms from chemical vapors and is then collapsed into a solid rod, the preform, from which fiber is drawn. A 2000 Lightwave account of MacChesney's 1999 Draper Prize says the process "made mass production of high-quality optical fiber practical."

    lightwaveonline.comoptica.org

    Credit

    John B. MacChesney and P. B. O'Connor, Bell Laboratories

  2. 1977

    Preforms grown end-on

    Tatsuo Izawa of NTT invented vapor-phase axial deposition (VAD), which builds the preform up along its length and skips the collapsing step that MCVD and outside vapor deposition (OVD) need. NTT developed it with Furukawa Electric, Sumitomo Electric and Fujikura from 1975 to 1983. The IEEE Milestone for VAD, dedicated on 21 May 2015, says it is now the most used method and makes more than 50% of the fiber used in telecommunications.

    Lessona process that one telephone company worked out with three cable makers became the world's main way to make fiber.

    ethw.org

    Credit

    Tatsuo Izawa (NTT), with NTT, Furukawa Electric, Sumitomo Electric and Fujikura

  3. 1979

    Loss reaches the floor

    Four NTT researchers made single-mode fiber, whose core is so narrow that light takes one path, with a loss of 0.20 dB/km at a wavelength of 1.55 µm. Decibels per kilometer (dB/km) measure how much light is lost in each kilometer; at 0.20 dB/km about 95% is left after one. Their paper called it "ultimate low-loss" fiber, and the record has since fallen only to 0.1400 dB/km, reported by a Corning team in 2024. Sumitomo Electric mass-produced pure-silica-core fiber at 0.17 dB/km in 1988 and has sold fiber with a typical loss of 0.144 dB/km since January 2021.

    Lessonthe floor set by the glass itself was reached early, and later gains came in hundredths of a dB/km.

    doi.orgmanuscript.isc.acglobal-sei.comdoi.org

    Credit

    T. Miya, Y. Terunuma, T. Hosaka and T. Miyashita, NTT Ibaraki Electrical Communication Laboratory

  4. 1993

    Zero dispersion backfires

    Dispersion is the spreading of light pulses as they travel. Dispersion-shifted fiber put zero dispersion at 1,550 nm, where loss is lowest, but when systems began to send many wavelengths down one fiber, called dense wavelength-division multiplexing (DWDM), the channels mixed and made noise. A 2001 Lightwave review calls this effect, four-wave mixing, "a serious problem" on that fiber. Makers answered with non-zero dispersion-shifted fiber: Lucent's TrueWave, made from 1993 by the same review's account, and Corning's LEAF from 1998. ITU-T first standardized the type as G.655 in October 1996.

    Lessona fiber tuned for one kind of system can work against the next.

    lightwaveonline.comitu.int

    Credit

    Lucent Technologies (TrueWave); Corning (LEAF); ITU-T

  5. 2001

    The fiber glut

    In October 2001 Corning said it would idle optical fiber production, might close its plant in Deeside, Wales, and would raise its planned job cuts to 12,000, after a third-quarter net loss of $220 million. KMI Research counted fiber plants falling from a peak of 65 in 2000 to 46 in 2002, and worldwide fiber cable sales falling from $8.7 billion in 2001 to $3.5 billion in 2002, partly from "price erosion due to excess capacity."

    Lessoncapacity built ahead of orders pushed prices down for every maker.

    lightreading.comcablinginstall.com

    Credit

    Corning (its own figures); KMI Research

  6. 2006

    Fiber that bends tightly

    In December 2006 ITU-T issued G.657, a standard for single-mode fiber that loses little light when bent, for the access network, the links that reach homes and buildings. On 23 July 2007 Corning announced a fiber for home installs that it said was "over 100 times more bendable than standard fibers." By 2026 CRU, an analyst firm, reports that demand for the G.657.A2 grade from data centers and fiber-guided drones has led many makers to move drawing capacity away from standard fiber.

    itu.intlightwaveonline.comcrugroup.com

    Credit

    ITU-T; Corning

  7. 2026

    Prices swing on preforms

    CRU reports that a two-year fiber shortage ended by mid-2018, after a record amount of new preform and drawing capacity came online. It then measured world cable demand down almost 7% in fiber-km in 2019 and about 24% in value, and in September 2020 China Mobile's tender set a record low of $5.61 per fiber-km for standard cable. CRU's price index for bare G.652.D fiber, the ordinary single-mode type, rose from a record low of 78.6 in March 2025 to 263.0 in March 2026, "the sharpest rise since CRU began monitoring in 2017." TrendForce says new preform capacity takes 18 to 24 months to build, and expects shortages to last until at least 2028.

    Lessonbecause preform plants are slow to build, a jump in demand becomes a shortage, and the capacity ordered during it can arrive into a glut.

    crugroup.comcrugroup.comcrugroup.comcrugroup.comtrendforce.com

    Credit

    CRU and TrendForce, independent analysts

Watch for

  1. Two cores across the Atlantic

    Meta says Petal, a cable of about 7,000 km between the United States and France due in service in 2029, will carry 1 petabit per second (one million Gbit/s) on 24 fiber pairs in which each fiber has two cores, two separate light paths in one strand. Sumitomo Electric makes the fiber. Meta calls Petal "the first to deploy multi-core fiber at scale" and gives no measured loss or crosstalk, the light that leaks between cores. ITU-T agreed only in October 2025 to write standards for this kind of fiber.

    Achieved whenthe cable carries traffic and someone outside Meta, NEC and Sumitomo publishes measured loss and crosstalk for the laid fiber, or ITU-T approves a multicore fiber standard and the fiber is shown to meet it.

    engineering.fb.comieee802.org

    Claimed by

    Meta, with NEC and Sumitomo Electric, September 2026

  2. Hollow-core fiber in volume

    Hollow-core fiber carries light through air inside a fine glass structure instead of through solid glass. In September 2025 Microsoft said Corning would make Microsoft's hollow-core design at Corning's North Carolina plants, and it signed an agreement with Heraeus Covantics to scale up production. Lightwave reported in May 2026 that Microsoft has used hollow-core cables in its Azure data centers since 2023, but gave no total made or laid, and 18% of respondents to a survey in the same report said they were waiting for cost and manufacturability to improve.

    Achieved whenan outside party reports hollow-core fiber made on the Corning or Heraeus lines in a stated volume, in km per year, or a buyer other than Microsoft installs fiber from those lines.

    lightwaveonline.com

    Claimed by

    Microsoft, with Corning and Heraeus Covantics, September 2025

Open

  1. 01

    One table for fiber specs

    An open, machine-readable table of the published specifications of at least 40 single-mode fiber products from at least 8 makers, in the grades G.652.D, G.657.A1, G.657.A2 and G.654.E, with each value linked to the datasheet page it came from and each product checked against the limits of the ITU-T standard it claims to meet.

    The brief
    Credit · unclaimed
  2. 02

    A public fiber price record

    An open series of fiber and cable prices per fiber-km from 2015 to 2026, built only from public tender award notices and company filings, with each point linked to an archived copy of its notice and the arithmetic shown: bid value divided by awarded volume, with tax noted.

    The brief
    Credit · unclaimed
  3. 03

    A tested bend-loss calculator

    An open script that predicts a fiber's macrobend loss, the light lost when it is coiled, at the bend radii used in G.657, from the fiber's refractive-index profile (how the glass's index changes from center to edge, including the low-index "trench" used in bend-insensitive fiber), with a validation set of published designs whose bend loss was measured.

    The brief
    Credit · unclaimed

Open 01 · the brief

One table for fiber specs

Credit · unclaimed

The first to publish is credited on this line, and the deliverable moves from Open to Achieved.

What to build

An open, machine-readable table of the published specifications of at least 40 single-mode fiber products from at least 8 makers, in the grades G.652.D, G.657.A1, G.657.A2 and G.654.E, with each value linked to the datasheet page it came from and each product checked against the limits of the ITU-T standard it claims to meet.

Why it is still open

Makers publish these values in their own datasheets and formats, and buyers and splicers compare them product by product. The differences matter: the Fiber Optic Association says that shrinking the mode field diameter, the width of the light spot in the core, for better bend performance "leads to mismatch losses when splicing," and that some G.657.A2 and B3 fibers show false readings in one-way OTDR tests. GitHub searches for "G.657 fiber" and "optical fiber specifications datasheet" found no such table.

foa.org

Built with what exists

The ITU-T G.652, G.654 and G.657 recommendations in their 08/24 editions; Python and pandas; a Claude-based document reviewer that pulls values out of PDFs and flags those that do not match the source or the standard (see the Note "Seven more flagged"). Language models that read tables in PDFs are what make this a one-person job.

  1. First step

    Write the G.657.A2 limits from the 08/24 edition as a JSON file, then load three makers' G.657.A2 datasheets into it.

  2. Done looks like

    At least 40 products from at least 8 makers, every value cited, marked as maximum or typical, and recorded with its test conditions (wavelength, bend radius, number of turns); a second person checks a random 10% of values by hand and finds at least 98% correct; each product gets a pass or fail for each ITU limit, with the reason.

Open 02 · the brief

A public fiber price record

Credit · unclaimed

The first to publish is credited on this line, and the deliverable moves from Open to Achieved.

What to build

An open series of fiber and cable prices per fiber-km from 2015 to 2026, built only from public tender award notices and company filings, with each point linked to an archived copy of its notice and the arithmetic shown: bid value divided by awarded volume, with tax noted.

Why it is still open

The most-cited price index is CRU's, which goes to subscribers, and CRU's articles sit behind registration. Some of the raw inputs are public: CRU's own 2020 analysis draws on China Mobile's published 2020/2021 award, in which YOFC's lowest bid was RMB 39.23 per fiber-km. A GitHub search found no open series.

crugroup.comcrugroup.com

Built with what exists

A web agent that collects and archives award notices from operators' procurement portals (see the Note "Deleting 80% to start over"); the Internet Archive's Save Page Now; a language model to translate notices and pull out bid values, volumes and shares; a SQLite file with one row per award.

  1. First step

    Find China Mobile's 2020/2021 loose-tube cable award notice and recompute YOFC's price from it.

  2. Done looks like

    At least 30 awards from at least 3 countries between 2015 and 2026, each with a source link and an archived copy; the series reproduces YOFC's RMB 39.23 per fiber-km from the 2020/2021 China Mobile notice exactly; anyone can recompute every point from the archived files.

Open 03 · the brief

A tested bend-loss calculator

Credit · unclaimed

The first to publish is credited on this line, and the deliverable moves from Open to Achieved.

What to build

An open script that predicts a fiber's macrobend loss, the light lost when it is coiled, at the bend radii used in G.657, from the fiber's refractive-index profile (how the glass's index changes from center to edge, including the low-index "trench" used in bend-insensitive fiber), with a validation set of published designs whose bend loss was measured.

Why it is still open

The open tools stop short. PyFiberModes, an MIT-licensed solver, computes cutoff, dispersion and effective area, but its README says it is "not a general finite-element solver for... bends." The open finite-element solver femwell has a bend-loss example, but for a 0.5 µm silicon chip waveguide, not a fiber. A GitHub fiber bend-loss script uses a fixed formula that ignores the profile.

github.comgithub.comgithub.com

Built with what exists

femwell, with its absorbing outer layer and bend-radius input; PyFiberModes for checks on straight fiber; a language model to pull index profiles and measured bend losses out of patents and papers.

  1. First step

    Find five published trench designs that report both the index profile and measured bend loss, then model a plain step-index fiber in femwell with no bend and compare it with PyFiberModes to check the mesh and boundary settings.

  2. Done looks like

    At least 15 published trench designs with measured bend loss, and for at least 12 of them a predicted loss at 10 mm and 7.5 mm radius at 1,550 nm within a factor of 2 of the measured value; inputs, code and results public, rerunnable with one command.