A strand of glass thinner than a human hair just carried more than one million gigabits of data in a single second. Two separate research teams have independently broken the petabit-per-second barrier on a single optical fiber, and they did it using fibers that match the outer diameter of cables already buried under streets and oceans worldwide. The results, one published in Nature Communications in June 2021 and another announced by Japan’s NICT in November 2022, suggest that the world’s existing fiber infrastructure is nowhere near its physical limits.
To put the speed in perspective: today’s most advanced commercial fibers typically carry somewhere between 10 and 40 terabits per second using wavelength-division multiplexing, the technique that splits light into many colors traveling simultaneously. One petabit per second is roughly 25 to 100 times faster than that. At 1.53 petabits per second, the faster of the two results, you could transfer the equivalent of about 20 million simultaneous 4K video streams.
Two experiments, two approaches
The Nature Communications paper, led by researchers working with a 15-mode fiber, reported a throughput of 1.01 petabits per second. Instead of sending data on a single beam of light, the team used a technique called space-division multiplexing, or SDM. Think of it as turning one lane of highway into 15 parallel lanes inside the same road. Each “mode” is a distinct spatial pattern of light that can carry its own independent data stream. At the receiving end, sophisticated algorithms called MIMO (multiple-input, multiple-output) digital signal processing sort out which data belonged to which mode, much the way noise-canceling headphones separate a voice from background chatter.
The critical detail: the fiber’s cladding, the outer glass layer that determines whether it fits into existing connectors and ducts, measured 125 micrometers across, the same as standard telecom fiber. That compatibility matters enormously because the most expensive part of building a fiber network is not the glass. It is the trenching, the permits, the rights-of-way, and the undersea cable ships. If a new fiber slots into old infrastructure, operators can upgrade by swapping equipment at each end of a cable run rather than digging up roads.
The second experiment, conducted by Japan’s National Institute of Information and Communications Technology (NICT), pushed even further. Using a 55-mode fiber with the same standard cladding diameter, NICT reported 1.53 petabits per second. More modes meant more parallel channels and higher aggregate throughput, but also far greater signal-processing complexity at the receiver. NICT is a Japanese government research institute with decades of published work in optical communications, lending institutional credibility to the result, though as of early 2023 the 55-mode figure had been announced via press release rather than a peer-reviewed journal article.
Why this matters for the internet’s plumbing
Global internet traffic has been roughly doubling every two to three years, driven by streaming video, cloud computing, and the steady migration of enterprise workloads online. Network operators face a familiar bind: demand grows exponentially, but laying new fiber routes is slow, expensive, and tangled in regulatory approvals. Industry analyses from groups like TeleGeography and the ITU consistently show that civil engineering, not the fiber itself, accounts for the majority of network construction costs.
That economic reality is what makes petabit-class transmission on standard-diameter fiber so significant. An alternative approach, multicore fiber, bundles several separate glass cores inside one cable but often requires a larger cladding diameter. Larger cladding means new connectors, new splicing tools, and potentially wider ducts. The SDM approach demonstrated in both experiments avoids that problem entirely. If the signal-processing and manufacturing challenges can be solved at commercial scale, carriers could multiply the capacity of routes they already own without replacing physical plant.
What still stands between the lab and your router
Neither experiment was conducted over the kind of distances that real networks demand. Undersea cables stretch thousands of kilometers; metropolitan links run tens to hundreds of kilometers through splices, bends, and temperature swings. How a 55-mode fiber performs after crossing an ocean, or even after threading through a city’s aging conduit, is an open engineering question that neither result addresses.
Signal processing is another hurdle. As mode counts climb, the MIMO algorithms needed to untangle overlapping signals grow larger and hungrier for power. Neither team has published full details of their equalizer structures, training sequences, or error-correction overhead in publicly accessible datasets, which makes independent apples-to-apples comparison difficult. The 55-mode approach delivers more raw throughput, but if its receiver consumes significantly more electricity per bit, the economics could favor the 15-mode design instead.
There is also the question of the supply chain. Manufacturing multi-mode fibers with tight tolerances, building compatible amplifiers and connectors, and training field technicians to install them all take time. The gap between a laboratory demonstration and a product sitting in a carrier’s equipment rack typically spans five to ten years, shaped as much by economics and standardization politics as by physics.
Both results date from 2021 and 2022, and as of mid-2025 no public follow-up from either team has surfaced announcing a major leap beyond these numbers. Whether subsequent work has quietly advanced or hit unforeseen obstacles is unknown from available sources. Readers should treat these milestones as validated proof-of-concept results, not previews of an imminent product launch.
About that headline number
The claim that one petabit per second could “beam every movie ever made around the world in a minute” depends on how you count. Estimates of the total number of feature films ever produced range widely, from roughly 500,000 to over half a million titles. Compressed to streaming quality (around 1.5 to 4 gigabytes per film), the entire catalog would total somewhere in the low exabyte range. At 1 Pbit/s, transferring two exabytes would take on the order of four to five hours, not one minute. The one-minute figure works only if you assume a much smaller slice of that catalog or a very aggressive compression ratio. The core achievement, a single fiber carrying a million gigabits per second, is real and verified. The movie analogy, while vivid, should be taken as illustrative rather than precise.
Where the glass ceiling actually is
What these experiments prove beyond doubt is that the fiber already threaded beneath city streets and across ocean floors is not close to its ultimate information-carrying limit. The Shannon limit, the theoretical maximum data rate for a given channel and noise level, still sits well above what even these record-breaking demonstrations achieved when you account for all available wavelengths, modes, and polarization states. The glass is not the bottleneck. The electronics, the economics, and the engineering of deployment are.
For telecom planners weighing their next decade of capital spending, the takeaway is concrete: standard-diameter multi-mode fiber offers a credible path to massive capacity gains without ripping out existing physical infrastructure. For everyone else, the message is simpler. The internet’s pipes have room to grow, possibly by a factor of 100 or more, before anyone needs to dig a new trench.
More from Morning Overview
*This article was researched with the help of AI, with human editors creating the final content.
