No Fiber, No AI: All-Optical Networks Accelerate Industries with Accessible AI

As the tide of digital and intelligent transformation surges forward, industries across the board are rushing to embrace AI. Over the next decade, AI will become the main engine of growth in the digital and intelligent economy, creating a market space worth trillions of dollars. Industries that acted early to adopt AI are already seeing the benefits. A survey shows that industries with the highest AI penetration have seen 4.8 times higher productivity. Lower-cost open-source foundation models like DeepSeek have driven down inference costs to 1% of the original costs, paving the way for industries' intelligent transformation and making AI more accessible across industries.

The popularization of AI has accompanied the evolution of the optical industry, which is encapsulated as "Three In's, Three Out's": "Fiber In, Copper Out", "fgOTN In, SDH Out", and "Optical-sensing In, Hard-work Out". This evolution has witnessed the rapidly expanded adoption of optical technology from intra-DC and inter-DC interconnection deployment to a wider range of scenarios, such as industry-specific communication networks, campus networks, and sensing devices. In turn, optical network technology can boost intelligent computing in four key ways: Greater computing efficiency, deterministic instant access to computing networks, wider adoption of AI applications, and intelligent remote operations for industries. Optical networks are making AI more accessible to numerous industries in multiple ways.

All-optical data centers: Boosting intelligent computing efficiency

Intelligent computing is known to involve oceans of data processing and bandwidth-intensive communication, among other demanding characteristics, something legacy networks are struggling to keep up with. Optical technology can help effectively address such issues.

There are multiple bottlenecks holding back traditional networks from supporting intelligent computing. To support training at the DCN layer with more than 4,000 cards, traditional switch networks face scalability challenges. There are statistics showing that intelligent computing centers require a cluster update every 15 days on average, but it takes one week to rebuild a DCN that is built on switches. Meanwhile, optical module failures account for 50% of DCN faults, making the infrastructure more vulnerable. The port rate of computing cards doubles every two years on average, which means switches frequently need to be replaced by those with larger capacities.

Huawei has released the DC-OXC all-optical switches with the particular aim of addressing these bottlenecks. Positioned for data centers, they are designed with ultra-dense ports and ultra-low power consumption. These switches can be scaled to support thousands to millions of cards within minutes. In addition, they do away with optical modules, slashing fault rates by 20%. They adopt port-level optical switching to become theoretically rate-irrelevant, meaning they support signals of different rates without changing the hardware and thereby support evolution from 400G to 1.6T and beyond.

Intelligent computing facilities are often constructed in phases, and decoupled storage and compute is being increasingly adopted for intelligent computing. These trends make cross-DC coordinated training a necessity, and demand no drops in compute power. However, computing power is highly sensitive to network interruption and packet loss. A packet loss of just 0.1% can result in a 50% loss of computing power, an issue that cannot be solved with traditional data center interconnect (DCI).

This year, Huawei launched its all-optical lossless DCI solution, which can achieve zero packet loss in remote direct memory access (RDMA), and zero bit errors upon intermittent fiber cuts and protection switching. The solution is able to meet the requirements of cross-DC remote AI training.

All-optical DCN and all-optical lossless DCI combined can improve the efficiency of collaboration for intelligent computing.

fgOTN: Boosting deterministic instant access to computing networks

Industries like electric power and transportation are actively adopting AI inference applications and launching AI foundation models. They are using AI for intelligent grid checks, precise load prediction, road traffic optimization, and more. These dynamics again put traditional industry-specific communication networks under pressure. For instance, intelligent power device checks require real-time data collection from thousands of terminals in substations. This results in an over tenfold increase in the volume of data and number of connections, which exceeds the capabilities of synchronous digital hierarchy (SDH) technology on the live network.

In its fine-grained Optical Transport Network (fgOTN) standard released in late 2023, the ITU-T identified fgOTN as the next-generation technology for replacing SDH. Both SDH and MPLS-TP will evolve to fgOTN. Huawei has released the OptiXtrans E6600 series, the industry's first products which meet the fgOTN standard and have been independently verified by China Electric Power Research Institute and State Grid Liaoning Electric Power. OptiXtrans E6600 has been enabling the transmission of teleprotection services for over a year now. This year, the International Council on Large Electric Systems (CIGRE) has set up an optical workgroup to research fgOTN and introduce it into the electric power industry standards. In 2025, fgOTN will enter large-scale commercial use worldwide. This will facilitate deterministic instant access to compute and bolster AI applications for electric power, transportation, and beyond.

FTTO: Boosting AI applications across campuses

Most campus customers choose to deploy foundation models such as DeepSeek on premises for concerns such as data privacy and a better user experience. AI is seeing wider application in campus settings like healthcare, education, and manufacturing. Hospitals are deploying AI foundation models for applications like AI-assisted pathological section analysis, which requires network bandwidth of over 10 Gbps. In the education sector, AI can generate teaching resources like 3D videos in real time and provide personalized learning suggestions for each student. This has resulted in a sharp increase in the number of terminals needed in classrooms, thus calling for high-density connectivity.

With this AI trend comes a steep spike in campus traffic, which is increasingly driving the "Fiber In, Copper Out" evolution. F5G-A 10 Gbps all-optical networks are set to become the mainstay campus network in the age of AI.

Since Huawei launched the industry's first FTTO (Fiber to the Office) solution, the industry has come to recognize the benefits brought by all-optical campus networks and see them as being ideal for 10 Gbps campus projects. So far, the solution has been commercially used across more than 10,000 campuses around the world. In China, the solution has become a preferred option for 10 Gbps campus deployment, including new campuses, renowned universities, over 500 grade-A tertiary hospitals, leading manufacturers, and over 1,000 large hotels. It is also seeing rapid growth outside China. The Oman e-School standard, for example, has included all-optical campus solutions in its campus network construction specifications.

This year, Huawei went on to release the FTTO 2.0 solution. Equipped with an industry-leading 50G PON OLT and the industry's first outdoor Wi-Fi 7 optical AP, the solution brings 50 Gbps to rooms and 10 Gbps to APs. In addition, the single-fiber 1:128 split ratio doubles the number of connections, enabling everyone on campus to use AI anytime, anywhere.

Innovative optical sensing: Boosting intelligent remote operations across industries

When it comes to applications like inspections of pipelines, optical cables, and perimeters, traditional remote inspection solutions such as radar, infrared, and vibration sensor cables have low precision, require high power supply, and are vulnerable to lightning strikes and electromagnetic interference. Furthermore, cameras used in such scenarios can be blocked by unexpected objects and have poor visibility in rain and fog. Better remote inspection technology is needed.

Next-generation optical fiber sensing resolves these issues with superior detection sensitivity, wider coverage, a longer lifecycle, and no need for power supply. Huawei has launched the industry's first Optical-Vision Linkage Solution – OVLS. Powered by AI, the solution uses optical fibers and videos for coordinated inspection, delivering 10 times higher accuracy than traditional solutions across diverse geological and perimeter settings.

By the end of 2024, the solution saw more than 90 applications worldwide. For example, PipeChina uses this solution for the intelligent remote inspection of more than 2,000 km of oil and gas pipelines. In Shandong province alone, this solution has rapidly identified and blocked more than 100 abnormalities, effectively averting threats that may have impeded construction. The Passenger Rail Agency of South Africa (PRASA) has prevented a large number of railway thefts with this solution.

This year, we expanded the solution to small perimeter inspection settings such as substations, oil and gas stations, and data centers. While maintaining a zero miss rate and a low false alarm rate, the solution reduced TCO by 30%, making it more affordable, especially for customers doing business on small campuses.

Optical sensing has huge potential of application, and more so in the age of AI. Huawei will keep advancing our research on optical fiber sensing, spectral sensing, and visual sensing. These foundational optical technologies can provide abundant high-quality data for AI, which will then enable intelligent remote inspection for even more industries.

As new trends and opportunities continue to emerge in the age of AI, we should lose no time in seizing upon them. Optical fiber plays a critical role by providing the high-speed, low-latency connectivity that is essential for AI's data processing needs. Huawei will continue to work with customers and partners worldwide to drive the optical industry evolution and accelerate the intelligent transformation of industries. Together, we can create new value and shared success from transformation.

Source: HuaweiTech

Disclaimer: The views and opinions expressed in this article are those of the author and do not necessarily reflect the official policy, position, products, and technologies of Huawei Technologies Co., Ltd. If you need to learn more about the products and technologies of Huawei Technologies Co., Ltd., please visit our product pages or contact us.