A team led by Japan’s National Institute of Information and Communications Technology, known as NICT, has transmitted 1.7 petabits of data per second through a single optical fiber stretching 63.5 kilometers. That is roughly 1.7 million gigabits per second, enough raw throughput to transfer every film ever produced, an estimated 500,000 titles totaling about two petabytes, in under 10 seconds. The result, published in Nature Communications in May 2025, sets a new record for data capacity through a fiber that matches the standard 125-micrometer cladding used in telecom networks worldwide.
To put the number in perspective, the fastest commercial links connecting major data centers today top out around 400 gigabits per second per wavelength channel. The NICT experiment moved more than 4,000 times that aggregate capacity through a single strand of glass thinner than a human hair.
How 19 cores fit inside one fiber
Conventional telecom fiber carries light through a single core. The NICT team’s fiber contains 19 cores arranged inside the same glass strand, a design known as randomly coupled multi-core fiber. Each core acts as a parallel highway for data, and because the cores are allowed to exchange some light between them (a controlled form of crosstalk), the digital signal processing needed to untangle the signals at the receiving end is simpler than it would be if every core had to be kept perfectly isolated.
This approach belongs to a broader family of techniques called space-division multiplexing, or SDM. Rather than cramming more data into a single light path by adding wavelengths or speeding up the clock, SDM opens additional spatial channels inside the fiber. The critical advantage of the NICT design is that it does this without making the fiber physically larger. The 125-micrometer cladding diameter means the fiber could, in theory, slot into the ducts, connectors, and splicing equipment already deployed across millions of kilometers of routes globally.
The NICT result is not the first peta-bit-class demonstration in a standard-diameter fiber. A separate peer-reviewed study led by researchers at NICT and collaborators at several Japanese institutions, published in Nature Communications in 2021, achieved peta-scale throughput using a 15-mode fiber, which packs channels into different light propagation patterns within a single core rather than splitting them across multiple cores. Both experiments prove the same essential point: the familiar glass strands running under streets and oceans have far more capacity left than their current use suggests, provided engineers rethink what happens inside the cladding.
Why the gap between lab record and live network is wide
Record-breaking fiber demonstrations have a long history of arriving years, sometimes decades, before commercial deployment. Wavelength-division multiplexing, the technique that underpins virtually all modern long-haul fiber links, was proven in laboratories in the 1980s but did not reach widespread commercial use until the mid-1990s. SDM faces a similar gauntlet of engineering, manufacturing, and standardization challenges.
The 63.5-kilometer test distance is meaningful for metro-scale connections between data centers or city-center hubs, but it falls well short of the hundreds or thousands of kilometers typical of submarine cables and terrestrial backbone routes. At longer distances, signal degradation, crosstalk between cores, and the need for inline optical amplifiers every 80 to 100 kilometers all introduce complications the published study does not address.
Manufacturing is another open question. Drawing a glass preform with 19 precisely positioned cores is significantly more demanding than producing conventional single-mode fiber. Neither the NICT paper nor any other primary source has published cost models or yield statistics for volume production. Without that data, it is difficult to predict whether multi-core fiber would initially serve only premium, high-margin links or could scale quickly enough to become a mass-market product.
Energy consumption matters too. The researchers note that their coupled-core design reduces the computational load of signal recovery, which should lower power consumption per bit. But the paper does not report measured energy-per-bit figures for the full system, including lasers, amplifiers, and receivers. For network operators whose electricity bills increasingly rival their capital spending, that gap in the data is significant, especially as AI training clusters and cloud services continue to drive explosive traffic growth.
Two architectures, one cladding, no clear winner yet
The 19-core and 15-mode approaches represent fundamentally different bets on how to pack more data into a standard fiber. The coupled-core design accepts some inter-core crosstalk and compensates with efficient digital processing. The multimode design keeps a single core but exploits the different spatial patterns light can take as it bounces down the glass. Each has trade-offs in connector loss, splice reliability, amplifier compatibility, and ease of integration with the vast installed base of single-mode fiber.
Which architecture wins in commercial deployment may hinge on factors the lab results do not yet resolve. Telecom ecosystems move cautiously: introducing a new fiber type requires consensus across standards bodies, equipment vendors, cable manufacturers, and operators. Neither multi-core nor multimode SDM fibers are part of widely deployed standards as of June 2026. Even proven physics must clear the bar of interoperability, certification, and operational monitoring in networks where downtime is measured in lost revenue per minute.
Where peta-bit fiber is most likely to land first
The most plausible near-term application is inside and between data centers. Link distances there are short, often under two kilometers, and the pressure to move massive volumes of data between GPU clusters, storage arrays, and network switches is intensifying as AI workloads scale. The DSP savings from coupled-core designs deliver immediate value in environments where power density per rack is already a binding constraint.
Long-haul networks present a harder target. They must coexist with decades of installed single-mode fiber, and any change to the physical layer ripples through amplifier chains, monitoring systems, and maintenance procedures. Operators are unlikely to rip out working infrastructure on the strength of a lab record alone. More probable is a gradual introduction: SDM fibers deployed on new routes or capacity upgrades where the cost of fresh construction makes the incremental expense of a new fiber type easier to justify.
For now, the NICT experiment and its multimode predecessor confirm that the theoretical ceiling for a single optical fiber is far higher than what today’s networks use. The engineering required to close the gap between 1.7 petabits in a lab and 1.7 petabits under the ocean floor is substantial, but the physics no longer stands in the way. As global internet traffic continues to double roughly every three years, the ideas proven in these experiments are moving from journal pages toward pilot deployments, and the question is shifting from “can it be done” to “how fast can it be built.”
More from Morning Overview
*This article was researched with the help of AI, with human editors creating the final content.
