Huawei Accelerates AI Empowerment in New Power Systems Through All-Optical Networks

Huawei: No Light, No AI – Full Optical Networks Accelerate AI Empowerment in New Power Systems

On April 24, 2025, during the Energy Network Communication Innovation Application Conference, Yang Xi, President of Huawei’s Government and Enterprise Optical Division, delivered a keynote speech titled “No Light, No AI – Full Optical Networks Accelerate AI Empowerment in New Power Systems.” He emphasized that the new power system, integrating source, grid, load, and storage, increasingly requires advanced control for energy flow, business flow, information flow, and carbon flow. Intelligent measures must be employed to enhance the efficiency of power production and operations.

Yang noted that the deployment of large models by power grid companies necessitates widespread use of full optical networks. Optical technology will rapidly extend from power data center interconnections to every power communication network, every power park, and every power sensing terminal, thereby empowering the new power system.

He explained that AI will drive the integration of optical technology into power data centers, power communication networks, and office parks, ensuring optimal collaborative efficiency between data centers. This will achieve “strong computing with light”; ensure deterministic connections from substations to large models for “computing enabled by light”; facilitate the easy deployment of large models to enhance office efficiency, making AI accessible to power office parks, achieving “computing benefitting from light”; and utilize optical fibers, spectral, and visual sensing technologies to generate a wealth of high-quality data for AI training and inference, embodying the concept of “empowerment through light.”

Strong Computing with Light: Full Optical Lossless DCI Enhances Intelligent Computing Efficiency

As a core infrastructure supporting the digital transformation of the power industry, power data centers are rapidly advancing intelligent computing applications to improve efficiency, reliability, and intelligence within the power system. Intelligent applications, such as enhancing the accuracy of defect recognition in transmission and transformation, utilizing deep reinforcement learning for inter-regional power balancing, reducing fault isolation times, and building 3D simulation models of the power grid combined with real-time data for AI training to predict grid fluctuations, require collaborative training across multiple data centers.

However, interconnections for intelligent computing centers are highly sensitive to network packet loss and fiber breaks, which can lead to up to 50% loss of computing power and severely affect training efficiency. Huawei’s full optical lossless DCI solution achieves ultra-wide lossless connections between intelligent computing centers, facilitating strong computing with light. By sensing congestion in real-time and managing traffic, it prevents packet loss during network congestion. Furthermore, its long-distance zero-error function ensures “0” errors during fiber interruptions or switching, and its zero-loss “0” interruption feature provides rapid 50ms protection during fiber outages, ensuring continuous business operations.

Promoting Computing with Light: Integrated Full Optical Power Communication Target Network

The proliferation and application of AI models, such as precise load forecasting and intelligent diagnostics of power equipment, greatly enhance power production and operations but also pose new challenges for power communication networks. For instance, intelligent diagnostics require real-time data collection from thousands of terminals within substations, resulting in a tenfold increase in data volume and a hundredfold increase in connections.

Currently, the power industry predominantly uses SDH technology for network construction, which is limited by a maximum bandwidth of 10G, with SDH standards having ceased development in 2007. Additionally, independently constructed network segments struggle to match the integrated collaboration of source, grid, load, and storage. These communication challenges hinder the application of large models in the power sector, necessitating a full optical power communication target network for deterministic connections.

Thus, the grid must establish a broader backbone network on the main grid side to support the decentralization of power control nodes to 35kV, enabling intelligent inspections and enhancing supply reliability and flexibility. On the distribution network side, the core area needs to be fully fiber-optic to improve supply reliability and precise load control. On the consumption side, panoramic real-time load detection must be established to enhance real-time interaction and regulation between the source and load, ensuring grid stability.

In light of the challenges faced by power communication networks in the AI era, the ITU-T plans to release the fgOTN standard by the end of 2023, marking the next-generation technology evolution of SDH. fgOTN continues to leverage the small granularity, hard isolation, high reliability, and deterministic low-latency characteristics of SDH under the OTN framework, while also providing significant bandwidth expansion capabilities. This will ensure high reliability, low latency, and efficient transmission channels for granular services like power, while also flexibly expanding network bandwidth to support future digital development, making it a hot technology for current power communication networks.

Empowering Computing with Light: AI Standard Configuration for Power Parks

Currently, in the realm of parks, with the popularization of AI models like DeepSeek, hospitals, universities, hotels, and manufacturing enterprises are extensively deploying large models to enhance production efficiency and user experience. Power office parks are also entering an era of universal AI application, which brings an upgrade demand for park networks.

Huawei’s full optical network solution for power parks features five key characteristics: ultra-10G, multiple connections, low latency, high reliability, and ultra-green technologies, accelerating the universal application of AI in power parks.

Empowering AI with Light: Providing Massive High-Quality Data

Due to severe electromagnetic interference around substation perimeters, traditional vibration cables and point sensors are highly susceptible to interference, leading to high false alarm rates and even failures. Optical fibers naturally resist electromagnetic interference; however, traditional optical sensing solutions in the industry often rely on single detection and judgment dimensions, resulting in numerous false alarms. Moreover, many sites have high installation space and power consumption requirements, making stacked solutions difficult for large-scale deployment.

Based on fiber-optic and visual sensing technologies, the optical-visual linked perimeter protection solution can rapidly identify and locate intrusion behaviors, quickly report alarms, and achieve dual verification through fiber and visual data, ensuring a false alarm rate of less than 1% and less than one false alarm per kilometer per day. This solution also supports online self-learning and continuous model optimization, providing a wealth of high-quality data for perimeter protection around substations, representing a truly reliable AI perimeter protection solution.

In conclusion, Yang Xi stated that looking ahead to the AI era, Huawei will continue to innovate and deepen its expertise in optical technologies, developing better full optical network solutions to accelerate the empowerment of new power systems by AI.