Huawei Galaxy AI Data Center Network: Unleashing the Computing Potential in the AI Era
With the rapid development of artificial intelligence technology, the demand for computing power for applications such as large model training and real-time reasoning has exploded. Global technology companies have increased their investment in the construction of intelligent computing centers, trying to seize the technological high ground with large-scale hardware investment. However, there are many practical problems hidden behind the high investment: high energy consumption of data centers, insufficient computing power utilization, low efficiency of cross-regional resource coordination, and network communication performance restricting training speed... These problems not only push up the cost of AI research and development, but also become the key bottleneck for the large-scale implementation of technology.
As the core engine driving the intelligent transformation of thousands of industries, the release of computing power is inseparable from an efficient and reliable data center network. Faced with the many problems existing in current data centers, how to fully release computing power through data center networks has become a key issue that the industry needs to solve urgently.
Recently, at the Huawei China Partner Conference 2025, Zhao Zhipeng, vice president of Huawei's data communication product line, accepted an exclusive interview with the media, analyzed the current status and challenges of data center networks, and shared the breakthrough of Huawei's Galaxy AI data center network.
New challenges for data center networks in the AI era
Traditional data center network architectures have many deficiencies in terms of energy consumption, computing power utilization, cross-regional collaboration, and network communication performance.
The first is the problem of insufficient computing power utilization. Due to the limited performance of devices and algorithm optimization, the actual computing power utilization of many data centers is far lower than the theoretical value, and a large amount of computing power resources are wasted.
The second is the difficulty of cross-data center collaboration. The scale of a single data center is restricted by factors such as electricity, and the decentralized regional computing nodes face the problem of low collaboration efficiency. Network delays and communication losses lead to a significant reduction in overall training efficiency.
Another problem is that the threshold for enterprise universal access is high. Even if the intelligent computing center is successfully built, how to ensure that enterprises can call computing power at low cost and high efficiency is still a difficult problem to be overcome.
Two-way empowerment, reshaping network value
Facing the various challenges of data center networks, Zhao Zhipeng introduced that Huawei has upgraded the Galaxy AI data center network solution, which provides a solid foundation for the industry's intelligent transformation through two-way empowerment of "Netwrok for AI" and "AI for Network".
In terms of releasing computing efficiency, the current demand for inference has exploded, and the network communication time for the same task is generally longer than the computing time, resulting in nearly 20% of computing power wasted. Huawei has launched the CloudEngine XH9000 full series of switches, which supports 350ns ultra-low latency forwarding. Through the Star Intelligence AI inference scheduling algorithm, the communication time of inference traffic is reduced, and the inference performance is improved by 20%. In the general computing scenario, Huawei's latest network security integrated simulation solution ensures accuracy through an exclusive CMOS simulation algorithm, and realizes automatic generation of security policies and 100% accuracy. In addition, Huawei has continuously enhanced the network-level load balancing algorithm NSLB, effectively improving the training efficiency of AI and making limited computing power play its maximum potential. At the same time, by building a "computing power basic network", the scattered computing power nodes are virtualized into a unified resource pool, which not only improves the energy efficiency of AI, but also realizes the effective integration and improvement of computing power.
In terms of "AI for Network", Huawei focuses on using AI technology to improve network quality. AI algorithms monitor the status of optical modules, switches and other equipment in real time to achieve predictive maintenance. In addition, customers can use AI to analyze user needs, provide differentiated computing services and pricing strategies, and thus increase revenue.
In addition, Huawei has also launched the Starlink optical module for data center networks, which has three capabilities: ultra-long transmission, ultra-high reliability, and ultra-high security, creating a 3S high-quality network experience for enterprises and meeting their high requirements for network performance.
Traditional optical modules have the problems of high price, low quality and high failure rate, which seriously restricts the development of data center networks. Huawei has achieved a breakthrough through three major innovations. In terms of high-integration device design, by reducing the number of components, the reliability of optical modules has been effectively improved and the probability of failure has been reduced. At the same time, the use of multi-channel redundancy technology can automatically reduce the speed of operation without disconnection when a single channel fails, greatly reducing the risk of training interruption and ensuring the stability of computing power transmission. In addition, under the same performance, the coverage distance and compatibility of optical modules are improved to better adapt to different network environments, providing strong support for the efficient operation of data center networks.
Conclusion: Reshaping network value and opening up a new future for AI
To solve the AI computing power dilemma, we cannot rely solely on hardware stacking, but also need to break the resource island through network architecture innovation to achieve efficient flow and intelligent scheduling of computing power. Huawei Galaxy AI Data Center Network is redefining the network value in the intelligent era, and providing a solid foundation for the industry's intelligent transformation with a two-way empowerment model of "network + AI".