A single fiber carries 24 parallel streams of data
Chinese researchers have achieved real time bidirectional communication at 2.5 petabits per second through a single optical fiber, using 24 cores to carry information along parallel paths. The breakthrough was announced by the State Key Laboratory of Optical Communication Technologies and Networks under China Information and Communication Technologies Group Corporation (CICT).
Contents
- A single fiber carries 24 parallel streams of data
- How 24 cores change the structure of a fiber
- Why artificial intelligence needs more capacity
- Combining more cores with more wavelengths
- Keeping neighboring signals apart
- Three partners connect research with manufacturing
- China's position in optical communications
- What faster optical networks could support
- The Bottom Line
The project brings together CICT's State Key Laboratory, Pengcheng Laboratory and Fiberhome Fujikura Optic Technology Company. Their work combines a new fiber structure with techniques that transmit information using several wavelength regions of light at the same time.
The capacity is described as approximately 290,000 gigabytes of data every second, or enough to download 14,000 high definition 4K movies, each requiring 20 gigabytes of storage, in one second. Those comparisons give a sense of the volume involved: thousands of large video files moving through one fiber in the time it takes to say a short word.
CICT research fellow Yang Chao, who contributed to the project, explained the change by comparing communication infrastructure to transport routes:
In the past, we were building highways, but now, it's more like constructing a three-dimensional transportation network.
The achievement establishes a new benchmark for optical communication as artificial intelligence, large language models and massive data centers demand greater bandwidth. Bandwidth is the amount of information a connection can carry in a given period. The reported capacity represents an increase of roughly an order of magnitude over current commercial optical networks, meaning a jump on the scale of ten times.
Traditional fiber systems have carried the internet's traffic for decades. The challenge now is to keep expanding their capacity while the amount of information generated, stored and exchanged continues to grow rapidly.
How 24 cores change the structure of a fiber
Conventional optical communication can be pictured as a road with a single lane. Light carries information through a core, the central region of the fiber that guides the signal. The new system uses 24-core single-mode fiber, providing 24 parallel channels within a fiber described as no thicker than a human hair.
Each core acts as an independent waveguide, a structure that keeps light traveling along a defined path. Each can carry traffic at the scale of terabits per second while preserving the quality of the signal over very long distances.
Researchers call this approach space-division multiplexing. Multiplexing means carrying several streams of information through the same communication system. Here, the streams are separated by their physical positions inside the fiber, rather than all being crowded into one core.
Earlier methods increased capacity by separating signals in time or by using different wavelengths, often described as different colors of light. Time-division multiplexing gives different streams their own time slots. Wavelength-division multiplexing allows different streams to travel on different wavelengths.
Those methods face physical constraints, including nonlinear optical effects and Shannon capacity limits. Nonlinear effects occur when stronger light signals change how they behave inside the fiber, potentially disturbing transmission. Shannon capacity describes the limit on how much information a channel can carry reliably under given conditions.
The CICT team also demonstrated bidirectional transmission: data travels simultaneously in both directions across all 24 cores. Compared with transmission in only one direction, this effectively doubles the practical throughput described for the system.
Why artificial intelligence needs more capacity
Artificial intelligence training clusters and data centers running large language models generate and consume volumes of information that were difficult to imagine five years ago. Their processors need a continuing supply of data, and the connections between computing equipment must keep pace.
AI accelerators are processors designed to perform the calculations used in artificial intelligence. High performance computing nodes are individual computing units working together on demanding tasks. Both depend on moving information quickly enough to keep their processing resources occupied.
Photonic input and output chips use light to move data into and out of computing equipment. Research into the requirements of future computing centers points toward chips capable of handling traffic at the scale of petabits per second, so processors are not left waiting for information.
The technical obstacles include integrating lasers, increasing modulation speed and improving photodetector efficiency. Modulation is the process of changing a light signal to encode information. A photodetector converts the arriving light into an electrical signal that computing equipment can use.
Projections described for the sector suggest that optical interconnects operating at petabits per second will become essential laboratory achievements within the next decade, with commercial deployment following shortly afterward. No specific date is given for commercial deployment of the CICT system.
Traditional single-mode fiber systems carry most global internet traffic and are moving toward their capacity ceilings. The growing demands of cloud computing, streaming services and AI applications make additional transmission paths increasingly valuable.
Combining more cores with more wavelengths
CICT's approach expands capacity in two directions at once. The 24 cores provide more physical paths, while transmission across multiple wavelength bands provides more usable spectrum within those paths. Spectrum, in this setting, means the range of light wavelengths available to carry signals.
The combination creates a two dimensional scaling strategy. Instead of relying entirely on making one channel carry more information, researchers can increase both the number of paths and the range of wavelengths traveling through them.
The project presents this combination as a practical route toward networks operating at the scale of exabits per second, a further level of capacity beyond petabit systems. Such networks are intended to meet the information demands of future generations of computing and communications.
Reaching that scale requires progress across several parts of the system. Alongside lasers, modulation and photodetectors, engineers must expand multiplexing and demultiplexing, the processes that combine streams for transmission and separate them again at their destination. Packaging density also matters because more components and connections must fit into the equipment.
These demands become harder as data centers support AI training runs involving models with trillions of parameters. Parameters are the numerical values a model adjusts during training. Moving the information needed to calculate those adjustments can become a constraint on how quickly training finishes.
The engineering challenge includes delays measured in microseconds, or millionths of a second. Across the repeated exchanges of a very large training run, such delays can accumulate and extend the work by days or weeks.
Keeping neighboring signals apart
Fiberhome Fujikura manufactured the 24-core fiber used in the demonstration. Its task included precisely controlling the refractive index profiles of the cores, which determine how light is guided through each path.
The refractive index describes how light travels through a material. By controlling how it changes across a core and its surroundings, manufacturers can help keep the light in the intended channel.
Signal leakage between neighboring cores, known as crosstalk, would reduce transmission quality. A fiber with many paths must keep those paths sufficiently separate for each stream of data to reach its destination reliably, especially over long distances.
The successful demonstration indicates that the manufacturing techniques needed for these advanced fibers have reached a level capable of supporting the transmission result. The project also points toward the commercial production skills required to turn such fibers into communication infrastructure.
Three partners connect research with manufacturing
CICT's State Key Laboratory, Pengcheng Laboratory and Fiberhome Fujikura bring different capabilities to the project. Their collaboration joins theoretical research, computing resources and experience in making optical fiber.
Fiberhome Fujikura contributes knowledge of fiber drawing, coating and cabling. Drawing produces the thin fiber from its starting material, coating protects it, and cabling prepares it for use in communication networks. These processes are necessary for producing fibers with 24 cores at industrial scales.
Pengcheng Laboratory provides high performance computing infrastructure needed to generate and verify streams of data at the scale of petabits per second. That computing capacity helps prevent the equipment producing or checking the traffic from becoming the bottleneck in a transmission demonstration.
The achievement validates the engineering feasibility of combining space-division multiplexing with transmission across multiple wavelength bands. It is presented as a step toward networks that can operate reliably beyond controlled experimental environments and toward infrastructure that telecommunications operators can install and maintain.
CICT, a core enterprise in China's optical communication sector, has sustained its focus on three goals: very large capacity, very high speed and very long transmission distance. The latest result brings those priorities together in one system.
The work builds on years of independent research into advanced modulation formats, digital signal processing algorithms and specialized optical amplifiers. Modulation formats determine how information is encoded onto light. Digital signal processing helps recover that information, while optical amplifiers strengthen signals along their journey.
China's position in optical communications
Xiang Ligang, director-general of the Zhongguancun Modern Information Consumer Application Industry Technology Alliance, described the result as an advance for China's position in the international optical communication sector:
The breakthrough further consolidates China's leading international position in the field of optical communications and will promote the upgrading and development of the optical communication industry.
The project also strengthens CICT's position in developing the next generation of networks through domestic research. Its work on core optical technologies supports an effort to reduce dependence on foreign suppliers of optical components.
Optical communications have become an arena of competition over technological sovereignty, meaning the ability to develop and control essential technology. The infrastructure carrying global information has economic and strategic value, affecting cloud service costs and communications used for national security.
Countries and companies that help establish transmission standards can influence how future networks are built. Standards set common technical requirements so equipment and connections can work together. A role in setting them can also shape demand for components and manufacturing expertise.
Demonstrating transmission at the scale of petabits per second before it becomes a commercial necessity gives Chinese researchers time to work on standardization, international patents and manufacturing readiness. Domestic companies could supply the specialized cables, optical components and transmission equipment needed for later network upgrades.
The potential advantage resembles the lasting commercial position associated with single-mode fiber standards during the early expansion of the internet, when Western companies benefited from their role in the technology. The current project places fiber with multiple cores at the center of a new stage of competition.
What faster optical networks could support
Advanced mobile networks, cloud computing platforms and big data analysis depend on stable connections capable of carrying large volumes of information. The new transmission approach is intended to support those services and the increasingly digital economies envisaged by policymakers and technology leaders across Asia and beyond.
Possible consumer uses include 8K live streaming without buffering, AI interactions with responses that feel instantaneous, and cloud gaming with visuals comparable to playing on a local console. Each involves moving substantial amounts of information between remote computing systems and users.
Medical applications include transmitting images with high resolution in real time for consultations involving remote surgery. Financial markets could use communication systems seeking advantages measured in microseconds for algorithmic trading, where software executes trades according to programmed rules.
Manufacturers could support more sophisticated automation through centralized control systems communicating with distant facilities. Such uses depend on connecting machines, monitoring systems and computing resources across locations.
For artificial intelligence, a central application is moving large datasets between storage systems and processing units spread across computing clusters. More transmission capacity could allow those clusters to grow beyond current constraints and help accelerate work in scientific research, drug discovery, climate modeling and autonomous vehicle development.
The broader ambition is to move massive flows of information across cities, countries and continents with fewer capacity bottlenecks. The 24-core demonstration addresses the underlying communication infrastructure on which those services would depend.
The Bottom Line
- CICT's State Key Laboratory announced real time bidirectional transmission at 2.5 petabits per second through a single optical fiber.
- The system uses 24-core single-mode fiber, combining parallel physical paths with transmission across multiple wavelength bands.
- The capacity is described as approximately 290,000 gigabytes per second, or 14,000 4K movies of 20 gigabytes each in one second.
- Pengcheng Laboratory and Fiberhome Fujikura Optic Technology Company partnered with CICT on computing verification and fiber manufacturing.
- The work targets rising communication demands from AI training, large language models, data centers and cloud services.
- Potential applications range from 8K streaming and cloud gaming to medical consultations, industrial automation and scientific computing.
- The project supports China's ambitions in optical technology, manufacturing and future network standards. No specific commercial deployment date is given.






