A team of researchers at Japan’s National Institute of Information and Communications Technology (NICT) has transmitted 1.7 petabits of data per second through a single strand of optical fiber, shattering the previous record for throughput in a cable that could slot directly into existing telecommunications infrastructure. The result, published in Nature Communications in May 2025, was achieved over a distance of 63.5 kilometers using a fiber with 19 light-carrying cores packed inside the same 125-micrometer cladding diameter that telecom companies have been burying underground for decades.
To put that number in perspective: 1.7 petabits per second is roughly 1.7 million gigabits per second. At that rate, you could transfer every film ever produced, assuming a catalog of around 500,000 titles encoded in high definition, in well under a minute. The comparison is approximate, but the scale is real. Global average internet traffic hovers around 1 petabit per second according to industry estimates from groups like TeleGeography. A single fiber operating at the rate NICT demonstrated could, in theory, shoulder more than the entire planet’s average data flow on its own.
Why the cladding diameter matters more than the speed
Record-breaking fiber experiments are not new. Research labs have been pushing multi-core and multi-mode fibers to extraordinary throughput for years. What distinguishes this result is the form factor. Previous high-capacity multi-core fibers often required wider cladding, sometimes 200 micrometers or more, which meant they could not fit into the ducts, connectors, and splicing machines that carriers already own. Deploying them would have required tearing out and replacing physical infrastructure, a cost so steep that no operator would seriously consider it.
The NICT team designed their 19-core fiber to match the 125-micrometer standard. That decision was deliberate: it means the fiber can, in principle, be pulled through existing conduits and terminated with compatible hardware. For telecom operators who have spent decades and billions of dollars building out their cable plants, this compatibility is the difference between a lab curiosity and a technology worth evaluating for real networks.
How 19 cores share a single strand
Inside the fiber, 19 individual cores each carry a portion of the total data load. Rather than trying to isolate each core from its neighbors, which becomes impractical at this density, the researchers used a “randomly coupled” design. Light signals in adjacent cores are allowed to interact, and the resulting tangle of overlapping signals is unscrambled at the receiving end using multiple-input, multiple-output (MIMO) digital signal processing, the same mathematical framework that lets modern Wi-Fi routers handle multiple antennas.
The tradeoff is computational complexity at the endpoints in exchange for a dramatic increase in how much data a single fiber strand can carry. Advanced modulation formats squeezed additional information into each optical signal, and the combined system achieved stable performance across all 19 cores over the full 63.5-kilometer span.
The long road from lab bench to buried cable
The 63.5-kilometer transmission distance falls within metro-network range, but the experiment took place under controlled laboratory conditions. Real-world fiber routes contend with mechanical stress from installation, temperature swings, moisture, and the accumulated losses from dozens or hundreds of splice points where fiber segments are permanently joined together.
Splicing is one of the biggest practical hurdles for multi-core fiber. Joining two standard single-core fibers is routine; field technicians do it thousands of times a day worldwide. Aligning 19 cores simultaneously with the precision needed to avoid signal loss or crosstalk is a fundamentally harder problem. Specialized multi-core splicers would be required, and the NICT paper does not include splicing-yield data or cost-per-bit projections for the 19-core design.
Amplification presents another challenge. Today’s long-haul fiber networks rely on erbium-doped fiber amplifiers spaced at regular intervals to boost signals over hundreds or thousands of kilometers. Those amplifiers are engineered for single-core fiber. Boosting all 19 cores simultaneously without introducing excessive noise or crosstalk would require new amplifier architectures. The paper describes stable core-by-core performance but does not address whether existing amplifier technology can be adapted economically for multi-core deployment. Without that piece, extending these links to regional or transoceanic distances remains an open engineering problem.
Manufacturing scalability is similarly unresolved. Producing a few kilometers of experimental fiber under tightly controlled conditions is a different challenge from mass-manufacturing thousands of kilometers with consistent core geometry and refractive index profiles. Tiny variations could degrade the carefully balanced coupling that the MIMO decoding algorithms depend on. The paper does not provide yield statistics across multiple production runs.
Where this fits in the bigger picture
The result builds on a progression of multi-core fiber experiments that have pushed capacity higher over the past decade, each iteration using more cores, more spatial modes, or more sophisticated signal processing. NICT itself has been at the forefront of several previous records. Within that trajectory, this work represents a milestone that combines extreme throughput with a form factor compatible with today’s networks, a combination no prior experiment had achieved at this scale.
No competing research group has publicly challenged the 1.7-petabit figure, and the peer review process at Nature Communications requires rigorous methodology disclosure. Independent replication by other labs has not yet been reported, however, so the result remains a strong but singular data point. The gap between a successful lab demonstration and a product that field technicians can install and maintain over decades has historically been wide in fiber optics.
For now, the clearest takeaway is that the physical and technological ceiling for data transmission through standard-diameter fiber is far higher than what commercial networks currently use. Whether that ceiling translates into faster internet for anyone depends on answers that have not yet been published: manufacturing yield, splicing practicality, amplifier design, and total system cost. Until those pieces fall into place, the NICT result stands as a landmark in optical research and a signal that the next generation of fiber infrastructure may not require ripping out the old one to arrive.
More from Morning Overview
*This article was researched with the help of AI, with human editors creating the final content.
