Japan’s NICT and Sumitomo Electric shatter every known limit in fiber optic science — transmitting one petabit of data per second across 1,808 kilometres — while a second, quieter revolution redefines what standard cables can carry
On the morning of April 3, 2025, inside a conference hall in San Francisco, a post-deadline paper quietly upended decades of assumptions about how much information a glass fibre the width of a human hair can carry. Researchers from Japan’s National Institute of Information and Communications Technology and their partner Sumitomo Electric Industries announced they had transmitted data at one petabit per second — over a distance of 1,808 kilometres. The scientific community, accustomed to speed records that arrive in breathless bursts only to be cloistered in short-distance laboratory demonstrations, sat up.
This was different. This was distance.
One petabit is one million gigabits. The transmission sustained that rate across a span comparable to the rail distance from Sapporo to Fukuoka — or from Berlin to Naples. In terms of sheer simultaneous information throughput, the experiment could theoretically carry over ten million 4K video streams without dropping a single frame. It represented, according to NICT, the highest capacity-distance product ever recorded in an optical fibre with a standard cladding diameter: 1.86 exabits per second per kilometre.
A 19-lane superhighway, threaded through a tube no thicker than a standard telecommunications cable — and it fits into every trench already dug.
THE ARCHITECTURE OF A RECORD
At the heart of the achievement lies a deceptively modest-looking component: a 19-core randomly-coupled multicore optical fibre developed by Sumitomo Electric, engineered to carry 19 independent streams of light simultaneously — yet compressed into the industry-standard cladding diameter of 0.125 millimetres. This outer measurement is critical. It means the new fibre is physically compatible with existing global cable infrastructure, from urban backbone networks to transoceanic submarine systems.
Previous milestones in multi-core fibre transmission had achieved comparable or higher raw speeds — an earlier NICT 19-core fibre, for instance, demonstrated 1.7 petabits per second in 2023 — but over distances barely exceeding 63 kilometres. Beyond that, signal loss devoured the data. The challenge, in the laconic language of fibre optics engineering, was loss management at scale.
Sumitomo’s solution was structural: by meticulously optimising the arrangement and geometry of all 19 cores, engineers reduced optical loss simultaneously across two critical spectral regions — the C-band and L-band — the wavelength ranges commercially deployed across the global fibre network. NICT’s parallel contribution was the development of optical amplification relay systems capable of boosting all 19 cores simultaneously, every 86.1 kilometres, across 21 loops of fibre — simulating a real-world long-haul transmission route.
Signal processing completed the equation. A MIMO (Multiple-Input Multiple-Output) digital subsystem eliminated interference between cores — the optical crosstalk that had long been the ghost haunting multi-core ambitions — and used 180 wavelength channels modulated at 16QAM to extract the maximum information density from every core. The result, confirmed independently and presented as the best hot-topic paper at the 48th Optical Fibre Communication Conference (OFC 2025), was 1.02 petabits per second, sustained, across 1,808.1 kilometres.
WHY DISTANCE CHANGES EVERYTHING
It is easy to be seduced by speed alone. But the benchmark that experts in the field scrutinise most carefully is the capacity-distance product — the multiplication of how much data a system can transmit and how far it can carry it. Raw speed records set over a few dozen kilometres may dazzle in a press release, yet translate poorly into the physical geography of global telecommunications infrastructure. Submarine cables span oceans. Continental backbones must traverse thousands of kilometres without regeneration.
The 1.86 exabit per second-kilometre capacity-distance product achieved by NICT and Sumitomo eclipses every prior record for standard-cladding fibres. The previous best had been set using a 4-core uncoupled fibre transmitting 0.138 petabits per second across 12,345 kilometres — a longer distance but a fraction of the capacity. The new record reconfigures the trade-off frontier entirely.
The implications for undersea cables — the arteries of the world’s internet, carrying over 95 per cent of all transcontinental data — are immediate. Any next-generation submarine cable programme that incorporates this 19-core standard-cladding technology could, in principle, handle data volumes equivalent to the combined capacity of several existing major cable systems, without requiring the widening of cable conduits on the ocean floor.
TIMELINE OF MAJOR OPTICAL FIBRE TRANSMISSION RECORDS
| Year | Institution(s) | Speed | Distance | Fiber Type |
| 2020 | NICT, Japan | ~1 Pbps | ~2,000 km | Multi-mode |
| 2022 | NICT, Japan | 1.02 Pbps | 51.7 km | 4-core std. |
| 2023 | NICT + partners | 1.7 Pbps | 63.5 km | 19-core (earlier gen) |
| 2024 | NICT + Aston Univ. | 402 Tbps | Standard range | Standard single-mode |
| May 2025 | NICT + Sumitomo Electric | 1.02 Pbps | 1,808 km | 19-core std. cladding |
| Late 2025 | NICT + Aston Univ. + 10 partners | 430 Tbps | Standard range | ITU-T G.654 std. |
Sources: NICT (Japan), Sumitomo Electric Industries, Aston University (UK), OFC 2025, ECOC 2025. Highlighted rows represent 2025 breakthroughs.
A PARALLEL REVOLUTION IN STANDARD FIBRE
While the petabit-class story commanded the headlines, a second NICT-led achievement in late 2025 deserves equal scrutiny from a near-term engineering perspective. In a paper presented at the 51st European Conference on Optical Communication (ECOC 2025) in Denmark in October, a twelve-institution international consortium — led again by NICT and including the UK’s Aston University, Germany’s Fraunhofer Heinrich-Hertz-Institut, Nokia Bell Labs, and Sumitomo Electric among others — set a new world record for transmission over commercially standard ITU-T G.654 optical fibre: 430 terabits per second.
The 430 Tbps result, surpassing the team’s own previous record of 402 Tbps set in mid-2024, carries a twist that engineers find as interesting as the number itself: it was achieved using nearly 20 per cent less total bandwidth. The technique exploited O-band wavelengths — shorter wavelengths not conventionally used in this fibre type — to achieve three-mode transmission, multiplying the spectral capacity of specific wavelength regions by up to three times. The practical consequence is that existing fibre, buried under millions of kilometres of streets and seabeds, can be coaxed to carry far more data than its designers ever envisaged, without excavation or replacement.
These are not incremental refinements — they are proofs of principle that the physics of glass is more capacious than the economics of deployment has so far dared to exploit.
THE DEMANDS THAT DRIVE THE SCIENCE
The urgency behind these experiments is not academic. Global internet traffic, driven by the mass deployment of generative AI services, cloud computing expansion, high-resolution video streaming, IoT sensor networks, and the infrastructure requirements of 5G and emerging 6G systems, is growing at a pace that existing commercial optical deployments are straining to match. Industry analysts at Dell’Oro Group raised their forecasts for network infrastructure buildout in 2026, citing AI data-centre demand as the primary driver. Submarine cable capacity through existing routes is growing at roughly 30 per cent per year, while new data centre buildouts are adding to cumulative demand at 27 per cent annually.
Against this backdrop, both NICT records speak directly to a cost-of-infrastructure problem. Laying new fibre is expensive — measured in millions of dollars per route — and geopolitically sensitive in the case of submarine systems. Technologies that radically expand the capacity of fibres already in the ground, or that make next-generation cables with dramatic capacity advantages deployable within existing cable architecture, directly reduce the capital burden on the telecoms industry at precisely the moment that burden is rising fastest.
Real-time AI model training across geographically distributed data centres — a scenario that today requires careful bandwidth rationing across intercontinental links — would, on a petabit-class backbone, become a routine operation. The latency implications for global financial systems, scientific collaborations requiring petabyte-scale dataset transfers, and telemedicine infrastructure are equally tangible.
COLLABORATION ACROSS CONTINENTS
Both records underscore a pattern that has become characteristic of the most ambitious photonic science: they are not the products of single institutions or nations. The 1.02 petabit record combined the fibre manufacturing expertise of Sumitomo Electric with the transmission system architecture of NICT and the experimental participation of Eindhoven University of Technology in the Netherlands, Politecnico di Milano in Italy, and the University of Stuttgart in Germany. The formal paper — presented at OFC 2025 as a post-deadline contribution, the conference’s highest distinction — carries the authorship of researchers across four countries.
The 430 Tbps record extends that collaborative geometry further: twelve institutions across nine countries, spanning Japan, the United Kingdom, Germany, the Netherlands, Italy, Brazil, Australia, and the United States. The involvement of Nokia Bell Labs and Sumitomo Electric alongside academic research institutes signals something important — that industrial partners are engaged not merely as component suppliers but as co-investigators at the frontier.
THE ROAD TO COMMERCIALISATION
Both records remain, for now, laboratory demonstrations achieved under controlled experimental conditions. The gap between a world-record transmission and a commercially deployed network system is wide, and the engineering challenges that fill it are formidable. Mass production of 19-core coupled fibres at the quality and consistency required for global deployment has not yet been demonstrated at scale. Optical amplifier designs capable of simultaneously boosting 19-core signals over transoceanic distances require further refinement. Signal processing hardware capable of handling the computational demands of MIMO decoding at petabit data rates does not yet exist as a commercial product.
NICT has acknowledged these realities, framing the 1.02 petabit achievement as a proof-of-concept for the next generation of long-distance, high-capacity optical systems — a foundational demonstration that the physics works, not an announcement of imminent deployment. The institute has indicated its intention to continue refining amplifier efficiency and signal processing architectures toward real-world viability. Sumitomo Electric, which in September 2023 launched what it described as the world’s first mass-produced ultra-low-loss multi-core fibre, has already taken initial steps toward closing the production gap.
The 430 Tbps standard-fibre result sits closer to near-term deployment, given that it works within existing ITU-standard cable architecture. Researchers at Aston University’s Institute of Photonic Technologies have noted that the technology’s compatibility with installed fibre networks makes it a candidate for urban area network upgrades and data centre interconnection — markets where capacity demand is acute and the economics of fibre replacement are most punishing.
A BENCHMARK FOR THE POST-5G ERA
NICT’s repeated position at the vanguard of optical fibre science is not accidental. The institute has been systematically pushing the capacity-distance frontier for over a decade, progressing through multi-mode fibres, uncoupled multi-core fibres, and now the coupled 19-core architecture. Each iteration has targeted a different constraint — spectral bandwidth, fibre compatibility, amplification range, signal coherence — and each has informed the next.
The 2025 records, viewed together, constitute something more than the sum of their numbers. They represent a demonstration that the physics of glass and light contains substantially more headroom than existing commercial networks exploit. The global internet as currently constituted operates at a fraction of the theoretical capacity of the fibres through which it runs. The question is no longer whether that capacity can be reached; it is at what cost, and on what timeline, the engineering to reach it can be deployed at civilisational scale.
In a world where the volume of AI-generated and AI-processed data is growing faster than any prior wave of digital demand, those are questions with answers that cannot wait for the next decade’s researchers to find.
– Kuppuswamy S
KEY FACTS AT A GLANCE
| RECORD (MAY 2025) | 1.02 petabits per second over 1,808 km — highest capacity-distance product (1.86 Eb/s·km) ever recorded in standard-cladding optical fibre |
| INSTITUTIONS | NICT (Japan) · Sumitomo Electric Industries (Japan) · Eindhoven Univ. of Technology (Netherlands) · Politecnico di Milano (Italy) · Univ. of Stuttgart (Germany) |
| FIBRE TYPE | 19-core randomly-coupled multicore optical fibre; standard 0.125 mm cladding diameter — compatible with existing infrastructure |
| SECOND RECORD (OCT 2025) | 430 Tbps over ITU-T G.654 standard telecom fibre, using 20% less bandwidth than prior record — 12-institution global consortium led by NICT and Aston University |
| DEPLOYMENT STATUS | Laboratory demonstration; not yet commercially deployed. Commercialisation requires advances in multi-core fibre mass production, amplifier design, and signal processing hardware. |
