A single strand of glass, no thicker than the fiber already buried under city streets and ocean floors, just carried more than one million gigabits of data per second. Two independent research teams crossed the petabit-per-second barrier using fibers built to the same 125-micrometer cladding standard that telecom carriers have deployed for decades. The results, published in peer-reviewed journals and presented at the field’s top conference, suggest that the next massive leap in internet capacity may not require tearing up the ground at all.
Two experiments, two paths, one threshold
The first result appeared in Nature Communications in 2021, when a team demonstrated 1.01 petabits per second through a 15-mode fiber. Rather than sending light down a single pathway, the researchers used mode-division multiplexing to pack separate data streams into distinct spatial patterns of light within one core, somewhat like transmitting multiple radio stations through the same antenna simultaneously. A comb-based transmitter operating across more than 80 nanometers of optical spectrum in the C+L band provided the raw wavelength count, while digital signal processing at the receiver untangled the overlapping modes.
The second, and more recent, result pushed the threshold further. In April 2025, Japan’s National Institute of Information and Communications Technology (NICT) and Sumitomo Electric Industries announced they had transmitted 1.02 petabits per second through a 19-core fiber over 1,808 kilometers. That combination of speed and distance set a new record for standard-cladding fibers. The team presented the work as a post-deadline paper at the Optical Fiber Communication Conference (OFC 2025), one of the field’s most selective venues, according to materials released by NICT and Sumitomo Electric. A detailed description of the 19-core fiber design, including its coupling regimes and dispersion characteristics, was subsequently published in Nature Communications in 2025.
“We have shown that it is possible to transmit more than one petabit per second over transoceanic distances using a fiber that conforms to the existing cladding standard,” the NICT and Sumitomo Electric team stated in their press materials. That claim is significant because it implies carriers could upgrade capacity without replacing the physical plant they already own.
To put the speeds in perspective: Netflix’s entire streaming library, estimated at roughly 36,000 hours of content, amounts to somewhere around 10 to 12 petabytes at 4K bitrates. At one petabit per second, transferring that volume would take roughly 80 to 100 seconds. The most advanced commercial single-mode fibers in service today top out at an estimated 100 terabits per second or so, though exact figures vary by vendor and configuration. These experimental fibers carried about ten times more data through the same physical footprint.
Both approaches achieved their results without changing the outer diameter of the glass. Standard telecom fiber uses a 125-micrometer cladding, and both the 15-mode and 19-core designs maintained that dimension. That detail matters enormously for cost. If a new fiber can slide into existing cable ducts, connectors, and splice trays, operators avoid the staggering expense of digging new trenches or laying new undersea routes.
Why the timing matters
Global internet traffic has been climbing steeply, and the pressure is intensifying. AI training clusters now shuttle enormous datasets between GPU racks. Cloud providers are expanding regions faster than ever. Submarine cable operators, who connect continents with fiber pairs that currently carry on the order of 250 terabits per second per pair, are already planning next-generation systems to keep pace. As of mid-2026, no commercial fiber in production comes close to a petabit per second, but the fact that two separate lab demonstrations have crossed that line gives network planners a credible target.
NICT has been steadily pushing capacity records for years, previously demonstrating multi-hundred-terabit speeds in 2020 and 2022. The jump to 1.02 petabits per second over nearly 2,000 kilometers represents a significant step, because long-haul performance is far harder to achieve than short-distance bursts. Signal degradation, amplifier noise, and crosstalk between cores all compound over distance, making the 1,808-kilometer result particularly relevant for real-world submarine and terrestrial backbone routes.
Competing approaches on the horizon
Multi-core and multi-mode fibers are not the only technologies vying to solve the bandwidth crunch. Hollow-core fiber, which guides light through air rather than solid glass, promises lower latency and reduced nonlinear distortion, though it has not yet matched the capacity figures achieved by the NICT or 15-mode teams. Several companies are actively developing hollow-core cables for data-center and short-haul links, and the technology could eventually complement rather than compete with solid-glass multi-core designs.
Meanwhile, satellite constellations such as SpaceX’s Starlink and Amazon’s Project Kuiper aim to bypass terrestrial fiber entirely for certain use cases, particularly in remote or underserved regions. Satellites excel at coverage breadth but face fundamental latency and per-bit cost disadvantages compared with fiber for high-volume backbone traffic. For the foreseeable future, the bulk of global data will continue to travel through glass, making the petabit-per-second results directly relevant to how that glass evolves.
What stands between the lab and your internet connection
Neither experiment has been tested under field conditions. Lab setups control temperature, vibration, and bending stress in ways that buried conduits and ocean-floor cables cannot replicate. No published data from either team addresses long-term mechanical reliability, bend-loss statistics over years of thermal cycling, or performance after repeated splicing by field technicians. Those factors will determine whether carriers can actually deploy these fibers at scale.
Cost remains a blank spot. Manufacturing a fiber with 19 coupled cores or supporting 15 spatial modes requires specialized draw processes and, critically, new amplifier designs. The erbium-doped fiber amplifiers that power today’s long-haul networks were engineered for single-mode fiber. Adapting them to boost multiple cores or modes simultaneously is an open engineering challenge. Neither the peer-reviewed papers nor the NICT press materials provide cost-per-bit estimates or confirm compatibility with current amplifier chains.
Interconnection hardware is another gap. Splicing a 19-core cable to a conventional single-mode route would require fan-out devices that split traffic into separate fibers without introducing excessive loss. Mode-multiplexed systems would need connectors and patch panels that preserve the spatial structure of each mode across every junction. As of mid-2026, none of those components are standardized or widely available.
No network operator has publicly committed to a deployment timeline for either fiber type. Carrier-grade adoption of new optical technology has historically lagged lab demonstrations by five to ten years, depending on how quickly manufacturing scales and standards bodies certify new specifications. Even if the glass is ready, operators need interoperable transceivers, monitoring tools, and field-repair procedures before they can risk putting novel fibers into revenue-generating routes.
The two architectures also present an unresolved tension. The 15-mode fiber achieved its speed over a shorter, unspecified distance, while the 19-core fiber proved it could sustain petabit throughput across 1,808 kilometers. Whether mode-multiplexed fibers can match that reach, or whether core-multiplexed fibers can match the spectral efficiency of mode-based systems, will shape which design wins commercial favor. It is also possible that hybrid approaches, combining multiple cores with multiple modes per core, could eventually push capacity even higher.
Where petabit fiber is likely to land first
The practical path from lab to deployment almost certainly starts at short distances. AI data centers already face bandwidth bottlenecks between GPU clusters separated by tens or hundreds of meters. At those ranges, mode-coupling penalties and amplifier limitations matter far less than they do over transoceanic spans. A standard-cladding multi-core or multi-mode fiber could multiply the data flowing between server racks without requiring new conduit, making data-center interconnects a natural first commercial use case.
Metropolitan networks could follow. Operators might deploy petabit-class links between major exchange points or cloud regions spanning a few tens of kilometers, gaining operational experience while the technology matures. Only after those intermediate steps prove reliable would multi-core or multi-mode fibers become serious candidates for subsea cables, where repairs are extraordinarily costly and reliability standards are the most demanding in the industry.
For now, the significance is clear but bounded. Two independent teams have proven that a single fiber, built to the same physical dimensions already in the ground worldwide, can carry more than a petabit per second. The engineering questions that remain, from amplifier design to connector standards to manufacturing cost, are substantial. But the experiments establish something that network planners can build toward: as data demand accelerates, operators may not need to rip out and replace every conduit to keep up. Future cables threaded through familiar ducts could quietly multiply the capacity of the global internet, provided the research community can turn these laboratory feats into robust, affordable products.
More from Morning Overview
*This article was researched with the help of AI, with human editors creating the final content.
