Chinese Researchers Achieve 3.2 Petabit Optical Fiber Transmission Record

A Chinese research consortium has established a new world record for real-time optical fiber transmission, successfully pushing 3.2 petabits per second (Pb/s) through a single cable. The test, conducted over a 10.3-kilometer link, utilized a 24-core fiber to achieve simultaneous bidirectional data transfer.[1][2]

The milestone was reached by a joint team comprising the National Key Laboratory of Optical Communication Technology and Network at China Information and Communication Technologies Group (CICT), Pengcheng Laboratory, and Fiberhome Fujikura Fiber Optic Technology. Their system relies on commercially compliant hardware architectures rather than purely experimental laboratory equipment.[1][2]

To contextualize the 3.2 Pb/s figure—which equates to 3,200,000 gigabits per second—the capacity is theoretically sufficient to download more than 20,000 4K-resolution movies in a single second. Alternatively, the bandwidth could simultaneously support gigabit internet connections for 3.2 million households.[1][3]

The achievement was formally announced on the Chinese government website for Hubei Province, which highlighted the practical implications of the bandwidth. According to the provincial portal, the transmission speeds are sufficient for “3.2 million households simultaneously enjoying gigabit broadband.”[1]

The announcement arrives as global data infrastructure faces unprecedented strain from artificial intelligence workloads and cloud computing demands. Traditional communication standards are increasingly proving inadequate for the volume of traffic generated by modern data centers, prompting a race to maximize the throughput of physical fiber-optic networks.[3]

Multiplying Capacity Through Space and Spectrum

The researchers achieved the record by combining two advanced optical techniques: spatial division multiplexing and multi-band transmission. Instead of relying on a single glass core to guide light, the specialized fiber contains 24 independent light-guiding cores within a standard cable footprint.[1][2]

Within those 24 parallel cores, the team utilized the S, C, and L optical frequency bands to carry thousands of distinct data channels simultaneously. Conventional commercial single-core G.652 optical fiber systems typically operate only in the C and L bands, capping out at a total transmission capacity of approximately 100 terabits per second (Tb/s).[2]

By expanding into the S band and multiplying the physical pathways, the new system delivers 32 times the total transmission capacity of those standard commercial fibers. The researchers successfully managed the distinct transmission behaviors, operational tolerances, and technical specifications across the combined frequency bands.[1][2]

A major hurdle in spatial division multiplexing is crosstalk—the tendency for signals traveling down adjacent cores to bleed into and corrupt one another. The Chinese team utilized advanced modulation techniques and commercial coherent optical modules to maintain stable transmission while mitigating this interference over the 10.3-kilometer distance.[1][2]

Real-World Hardware vs. Laboratory Limits

A critical distinction of the 3.2 Pb/s benchmark is its reliance on real-time transmission using hardware that meets commercial specifications. While purely academic experiments have pushed similar 24-core telecommunications fibers to speeds as high as 5.29 Pb/s, those efforts depend heavily on experimental signal processing.[1][3]

Those higher theoretical thresholds often require non-standard laboratory equipment that cannot be easily integrated into existing network topologies. By contrast, the CICT and Pengcheng Laboratory team utilized optical modules developed in-house to maintain stable, real-time transmission that mirrors real-world deployment conditions.[2][3]

During the trial, the researchers demonstrated real-time transmission at 400 Gb/s per wavelength in the S band, and 1.2 Tb/s per wavelength in the C and L bands. This commercial compliance suggests the architecture could be practical and deployable across enterprise and public networks in the near future.[1][2]

A Rapid Pace of Optical Advancement

The 3.2 Pb/s achievement represents a nearly 30 percent performance improvement over the team's previous record. Earlier in 2026, the exact same research coalition demonstrated a 2.5 Pb/s transmission rate using similar 24-core fiber technology.[1][2]

Securing a 30 percent capacity increase within a single calendar year highlights the rapid acceleration of optical networking research. The results of the transmission test have been formally submitted to the Journal of Lightwave Technology for peer review and publication.[1][2]

While the 10.3-kilometer test proves the viability of the multi-core, multi-band approach for short-haul connections, scaling the technology for transoceanic subsea cables or long-haul terrestrial networks will introduce new physical challenges. Signal attenuation and the need for specialized multi-core optical amplifiers remain significant engineering hurdles for longer distances.[1][3]

For now, the immediate beneficiaries of this technology are likely to be hyperscale data center operators seeking to eliminate bandwidth bottlenecks between massive GPU clusters. The ability to route 3.2 petabits per second through a single fiber strand could fundamentally alter the physical footprint and efficiency of next-generation AI infrastructure.[3]

The breakthrough lands on a technology calendar already crowded with optical interconnect developments, as the industry races to integrate photonic solutions. As silicon photonics and co-packaged optics gain traction, the ability to move data at petabit scales over short distances becomes the critical enabler for scaling artificial intelligence models.[1]

The transition from 100 Tb/s single-core fibers to 3.2 Pb/s multi-core architectures represents a generational leap in networking capability. The next verifiable checkpoint will be the successful deployment of these 24-core cables in a live, commercial data center environment, moving the technology from a provincial government announcement to an active enterprise network.[1][3]