Beyond the Chip Wars:
Huawei is Weaving a Larger AI Computing Network
AI New Engine
“Computing power has been, and will continue to be, the key to artificial intelligence, and it is crucial for China’s artificial intelligence as well.” On September 18, Xu Zhijun, Vice Chairman and Rotating Chairman of Huawei, emphasized this during his keynote speech at the Huawei Connect 2025 conference.
In this speech titled “Leading a New Paradigm of AI Infrastructure with Innovative Supernode Interconnection Technology,” Huawei elaborated on its strategic layout in the field of AI computing power. At the same time, Huawei Cloud clarified its positioning at the AI Summit held concurrently—reshaping the intelligence leap across various industries with the “AI New Engine.”
Xu Zhijun’s speech in Shanghai conveyed a clear message: computing power is the key to AI development, and Huawei is solidifying China’s AI computing foundation through its “supernode + cluster” strategy. Under Huawei’s comprehensive bet on AI computing power, Huawei Cloud has not only avoided strategic adjustments but has taken on an even more critical role.
01Computing Power Strategy Upgrade: Supernode Architecture Breaks Through Computing Bottlenecks
The most eye-catching release at this Huawei Connect conference was undoubtedly its“supernode + cluster” computing solution. This architectural innovation is redefining the competitive rules of AI computing power.
Xu Zhijun candidly pointed out the industry’s misconceptions: “AI computing power is not equal to chip performance; the shortcomings of a single chip’s performance can be addressed through the ‘supernode + cluster’ approach, achieving even stronger performance.” This statement directly responds to long-standing concerns in the industry about China’s limitations in chip manufacturing processes.
The Atlas 900 supernode launched by Huawei supports a full configuration of384 Ascend 910C chips, with a maximum computing power of 300 PFLOPS, making it the largest supernode in the world to date. Even more impressively, Huawei plans to upgrade the specifications of the cloud supernode from 384 cards to 8192 cards, ultimately achieving a super-large cluster of 500,000 to 1,000,000 cards.
The essence of supernode technology is to highly integrate and interconnect multiple servers, forming a “computing giant” that can operate like a single computer. Huawei’s self-developed interconnection protocol “Lingqu” achieves2.1 microseconds ultra-low latency and TB-level ultra-large bandwidth transmission, enabling thousands of chips to work efficiently together.
Huawei has not limited this technology to AI computing but has also expanded it to the general computing field. The world’s first general computing supernode, TaiShan 950, combined with the GaussDB distributed database, can replace traditional mainframes, achieving a performance improvement of2.9 times without requiring distributed transformation of the database.
02Huawei Cloud’s New Positioning: From Cloud Service Provider to AI New Engine
At the Huawei Cloud AI Summit held concurrently with the Huawei Connect conference, Huawei Cloud CTO Zhang Yuxin clearly stated: “Huawei Cloud is planning a new layout for the core needs of Agentic AI on the cloud platform, forming a full-stack layout coveringAI infrastructure, large models and toolchains, and Agent platforms.”
This statement clearly reveals Huawei Cloud’s new role in Huawei’s overall AI strategy. Huawei Cloud is no longer just a platform providing basic cloud services but is taking on the role ofthe core engine driving the implementation of AI technology in industries.
The strategy proposed by Huawei Cloud, “A Global Network, Dual Engines of Intelligence and Data,” is highly coordinated with Huawei’s overall AI computing strategy. The four core capabilities articulated by Huawei Senior Vice President and President of Global Marketing and Sales Services Yang Yougui—global storage and computing network, data enabling services, one-stop AI development platform, and global intelligent practice accumulation—correspond to the complete chain of enterprise intelligent transformation.
Huawei Executive Director and Huawei Cloud Computing CEO Zhang Pingan emphasized that Huawei Cloud must “steadfastly develop the commercial version of the Pangu large model, continue to increase investment in the Pangu large model, and deepen industry engagement.” This indicates that Huawei Cloud is rewriting the competitive landscape through the “commercial version of the Pangu large model + all-scenario computing power” dual-wheel drive.
03Data and Intelligence Integration: Deep Collaboration Between Data and AI
During the conference, Huawei Cloud launched anew data and intelligence integration product and solution system, further strengthening its infrastructure position in the AI era.
Huawei Cloud Global Ecosystem President Kang Ning pointed out at the Data Intelligence sub-forum: “We are entering a new development stage driven by both data and intelligence. The core logic of ‘data and intelligence integration’ based on ‘infrastructure + data governance + artificial intelligence’ is a key advantage for enterprises and even countries in the competition of the digital economy.”
The DataArts Fabric data and intelligence integration computing service and DataArts Studio knowledge lake service launched by Huawei Cloud have achieved a leap from “data lake” to “knowledge lake“. This transformation greatly enhances the efficiency of converting data into knowledge, compressing the data preparation cycle from “months” to “days.”
Huawei Cloud’s Chief Architect of Data and Intelligence Integration, Zeng Kai, pointed out the key issue: “The key to breaking the deadlock lies in promoting the leap from ‘data lake’ to ‘knowledge lake’, deeply integrating AI into data governance and business processes, and achieving a leap from ‘usable’ to ‘easy to use.'”
This capability of data and intelligence integration has already demonstrated its value in practical scenarios. The Shaanxi Cultural Investment Group, based on Huawei Cloud’s data and intelligence integration platform, has integrated multi-source data from cultural tourism, constructing high-quality datasets covering seven categories of distinctive cultural tourism, including historical culture, film and television works, and intangible cultural heritage, and developed the “Boguang Cultural Tourism” large model.
04Ecological Co-construction: From Technological Monopoly to Open Win-Win
The open attitude demonstrated by Huawei at this conference is noteworthy. Xu Zhijun announced that Huawei willopen the Lingqu 2.0 technical specifications, hoping that the industry chain will actively develop related products and components based on Lingqu, co-constructing an open ecosystem for Lingqu.
This decision has far-reaching significance. Xie Tao, Director of the Software Science and Engineering Department at Peking University, believes: “The interconnection protocol is the ‘nerve center’ of the supernode cluster, and Huawei’s choice to open Lingqu is expected to define the interconnection standards for the next generation of computing clusters.”
Huawei Cloud’s ecological openness strategy is also clear. Zhang Pingan stated that Huawei Cloud will open the fine-tuning toolchain of the Pangu large model, supportingthird-party models like DeepSeek and Kimi to run efficiently on Ascend AI cloud services. Additionally, through the Ascend CANN open-source community, partners are allowed to customize operator optimization performance.
In the HarmonyOS ecosystem, Huawei announced an investment of1 billion yuan in the “Tiangong Plan”, supporting AI innovation ecosystems and incentivizing the incubation of over 10,000 AI-native meta-services and more than 5,000 intelligent agents. This initiative aims to create a scale effect on both the supply and demand sides through a combination of “funding + tools.”
Huawei Cloud Stack, as Huawei Cloud’s hybrid cloud solution for government and enterprise customers, has served over5,200 government and enterprise customers worldwide, including over 800 government clouds and more than 300 financial institutions. This extensive customer base provides rich application scenarios for Huawei’s AI ecosystem.
05Industry Intelligence: AI Implementation Across Various Industries
The theme of the Huawei Cloud AI Summit, “AI New Engine, Reshaping Intelligent Leap Across Various Industries,” accurately summarizes the ultimate goal of Huawei’s AI strategy—promoting the practical implementation of AI technology across various industries.
In the building materials sector, Conch Group, in collaboration with Huawei, has created the first “AI + Cement Building Materials Large Model” in the cement industry, achieving an accuracy rate of over90% in predicting clinker strength, with standard coal consumption decreasing by 1% based on first-level energy efficiency.
In the corporate travel management sector, Huitong Travel, relying on Huawei Cloud’s enterprise-level intelligent agent platform Versatile, has developed the travel intelligent agent “Tongbao,” achieving the industry’s first full-process embedded intelligent service, with an intelligent recommendation adoption rate exceeding50%, allowing bookings to be completed within 2 minutes.
The design concept of Huawei Cloud’s enterprise-level intelligent agent platform Versatile highlights Huawei’s deep understanding of the implementation of AI in industries: unlike agents aimed at individual users, enterprise-level agents need to be deeply embedded in the entire business process, facing more complex scenario challenges and stricter capability requirements.
Huawei Cloud has implemented the Pangu large model inover 30 industries and more than 500 scenarios, providing value in fields such as government, finance, manufacturing, healthcare, coal mining, steel, railways, autonomous driving, and meteorology. This extensive industry coverage provides rich implementation scenarios for Huawei’s AI strategy.
Finally
In the global wave of industrial intelligence, Huawei has completed its strategic layout from self-developed technology to ecological co-construction. Huawei Cloud, as akey bearer of Huawei’s AI strategy, is driving the intelligent transformation of various industries through the dual-wheel drive of “commercial version of the Pangu large model + all-scenario computing power.”
With the accelerated formation of the national integrated computing power network, Huawei’s “supernode + cluster” computing solution is expected to completely resolve the computing power bottleneck of China’s AI industry. As the output window for these advanced computing powers, Huawei Cloud will take on an even more important role.
As Xu Zhijun stated, Huawei’s goal is to “eliminate computing power concerns in China“. Under this grand goal, Huawei Cloud is becoming a key bridge for Huawei to connect AI technology with industry applications, making it easier and better for applications across various industries to thrive on Huawei Cloud’s fertile ground.
What do you think about this? Feel free to leave your comments for discussion!
-END-