Huawei Cloud Drives AI Advancement with Innovative Compute Solutions at HUAWEI CONNECT 2025
On the second day of HUAWEI CONNECT 2025, held in Shanghai, Huawei's Executive Director of the Board and CEO of Huawei Cloud, Zhang Ping'an, took center stage to unveil transformative advancements in AI and computing technology. His keynote, titled "All Intelligence Empowering AI Pioneers for Industries", outlined Huawei Cloud's pivotal role in enabling companies to harness advanced AI capabilities.
This year marked a significant leap for Huawei Cloud as it introduced its AI Compute Service, driven by the powerful CloudMatrix384 infrastructure. The service is a game-changer, with plans to upgrade its supernodes from 384 cards to an impressive 8,192 cards. This monumental shift facilitates the establishment of hyperscale clusters capable of running on a staggering 500,000 to 1 million cards, thereby delivering unprecedented processing power essential for the intelligent era. Additionally, Huawei Cloud's Elastic Memory Service (EMS) achieved a groundbreaking milestone by expanding video RAM, resulting in remarkable reductions in latency during multi-round conversations on foundational AI models. This innovation significantly enhances user experience, showcasing Huawei's commitment to seamless AI interactions.
Huawei Cloud is also making strides in sustainability with its fully liquid-cooled AI data centers located across China, including provinces like Guizhou, Inner Mongolia, and Anhui. These advanced data centers support up to 80 kW heat dissipation per cabinet and have reduced power usage effectiveness (PUE) to an impressive 1.1. This means companies can access powerful AI compute capabilities without the hassle of reconstructing traditional data centers. Instead, they only need a pair of optical fibers to connect to Huawei's data centers, ensuring efficient AI cloud services.
One of the highlights of Zhang Ping'an's speech was the introduction of the AI Token Service. This service abstracts the underlying technical complexities, allowing users to obtain AI computing results directly and efficiently. The CloudMatrix384 effectively pools compute, memory, and storage resources while decoupling various tasks, which greatly enhances system performance. It caters to different latency needs in inference scenarios, providing a performance boost averaging three to four times that of previous iterations.
Huawei Cloud has been dedicated to refining its Pangu Models by engaging with industry-specific challenges. Collaborating closely with customers, Huawei aims to redefine possibilities within diverse sectors. The openPangu initiative has been particularly notable, providing best practices for AI training and making it accessible for developers. Alongside open-source progressions, the closed-source Pangu Model is also under development, ensuring a continuous investment in these models to meet real-world demands. As a testament to their effectiveness, Pangu Models have been integrated into over 500 different scenarios spanning 30 industries, including finance, healthcare, manufacturing, and more.
This year's conference also saw the launch of the CloudRobo Embodied AI Platform, a revolutionary step in robotic intelligence. By leveraging extensive cloud computational resources, the platform enables more sophisticated and versatile robots. Furthermore, the introduction of the Robot to Cloud (R2C) Protocol aims to create a secure communication channel between robots and the cloud, culminating in smarter, more capable robotic solutions.
Part of Huawei's strategic initiatives includes developing Kunpeng-powered ARM cloud services focused on performance and security. In the past year alone, the count of Kunpeng compute cores on Huawei Cloud surged from 9 million to 15 million, showcasing an impressive increase of 67%. This agility also extends to the adaptability of the Kunpeng platform, ensuring compatibility with versatile applications and enhancing efficiency across cloud services.
Huawei Cloud's GaussDB databases exemplify how innovative architecture can redefine data management by facilitating dynamic load scheduling and enhanced performance in transaction processing. The clustered configuration of GaussDB promises a transaction processing capacity reaching 5.4 million transactions per minute—a remarkable improvement compared to standard setups.
Lastly, Huawei's distributed cloud solutions, which consist of CloudOcean, CloudSea, CloudLake, and CloudPond, cover numerous operational environments. This enables a consistent and efficient cloud experience irrespective of user location.
Huawei Cloud continues to position itself at the forefront of AI and cloud services innovation. As outlined by Zhang Ping'an, the company's commitment to providing robust, scalable, and effective solutions is paving the way for countless industries to undergo digital and intelligent transformations.
With a dedicated focus on empowering enterprises through AI development and compute enhancements, Huawei Cloud is not only reshaping technological landscapes but also creating fertile ground for the next wave of industrial innovation.
Revolutionizing AI Compute Services
This year marked a significant leap for Huawei Cloud as it introduced its AI Compute Service, driven by the powerful CloudMatrix384 infrastructure. The service is a game-changer, with plans to upgrade its supernodes from 384 cards to an impressive 8,192 cards. This monumental shift facilitates the establishment of hyperscale clusters capable of running on a staggering 500,000 to 1 million cards, thereby delivering unprecedented processing power essential for the intelligent era. Additionally, Huawei Cloud's Elastic Memory Service (EMS) achieved a groundbreaking milestone by expanding video RAM, resulting in remarkable reductions in latency during multi-round conversations on foundational AI models. This innovation significantly enhances user experience, showcasing Huawei's commitment to seamless AI interactions.
Sustainable and Efficient Infrastructure
Huawei Cloud is also making strides in sustainability with its fully liquid-cooled AI data centers located across China, including provinces like Guizhou, Inner Mongolia, and Anhui. These advanced data centers support up to 80 kW heat dissipation per cabinet and have reduced power usage effectiveness (PUE) to an impressive 1.1. This means companies can access powerful AI compute capabilities without the hassle of reconstructing traditional data centers. Instead, they only need a pair of optical fibers to connect to Huawei's data centers, ensuring efficient AI cloud services.
Simplifying AI Access with Token Services
One of the highlights of Zhang Ping'an's speech was the introduction of the AI Token Service. This service abstracts the underlying technical complexities, allowing users to obtain AI computing results directly and efficiently. The CloudMatrix384 effectively pools compute, memory, and storage resources while decoupling various tasks, which greatly enhances system performance. It caters to different latency needs in inference scenarios, providing a performance boost averaging three to four times that of previous iterations.
Industry-Specific Model Development
Huawei Cloud has been dedicated to refining its Pangu Models by engaging with industry-specific challenges. Collaborating closely with customers, Huawei aims to redefine possibilities within diverse sectors. The openPangu initiative has been particularly notable, providing best practices for AI training and making it accessible for developers. Alongside open-source progressions, the closed-source Pangu Model is also under development, ensuring a continuous investment in these models to meet real-world demands. As a testament to their effectiveness, Pangu Models have been integrated into over 500 different scenarios spanning 30 industries, including finance, healthcare, manufacturing, and more.
Advancements in Robotics with Cloud Intelligence
This year's conference also saw the launch of the CloudRobo Embodied AI Platform, a revolutionary step in robotic intelligence. By leveraging extensive cloud computational resources, the platform enables more sophisticated and versatile robots. Furthermore, the introduction of the Robot to Cloud (R2C) Protocol aims to create a secure communication channel between robots and the cloud, culminating in smarter, more capable robotic solutions.
Hardware-Software Synergy with Kunpeng Services
Part of Huawei's strategic initiatives includes developing Kunpeng-powered ARM cloud services focused on performance and security. In the past year alone, the count of Kunpeng compute cores on Huawei Cloud surged from 9 million to 15 million, showcasing an impressive increase of 67%. This agility also extends to the adaptability of the Kunpeng platform, ensuring compatibility with versatile applications and enhancing efficiency across cloud services.
GaussDB: Redefining Data Management
Huawei Cloud's GaussDB databases exemplify how innovative architecture can redefine data management by facilitating dynamic load scheduling and enhanced performance in transaction processing. The clustered configuration of GaussDB promises a transaction processing capacity reaching 5.4 million transactions per minute—a remarkable improvement compared to standard setups.
Distributed Cloud Solutions for All Scenarios
Lastly, Huawei's distributed cloud solutions, which consist of CloudOcean, CloudSea, CloudLake, and CloudPond, cover numerous operational environments. This enables a consistent and efficient cloud experience irrespective of user location.
Conclusion
Huawei Cloud continues to position itself at the forefront of AI and cloud services innovation. As outlined by Zhang Ping'an, the company's commitment to providing robust, scalable, and effective solutions is paving the way for countless industries to undergo digital and intelligent transformations.
With a dedicated focus on empowering enterprises through AI development and compute enhancements, Huawei Cloud is not only reshaping technological landscapes but also creating fertile ground for the next wave of industrial innovation.
Topics Consumer Technology）
【About Using Articles】
You can freely use the title and article content by linking to the page where the article is posted.
※ Images cannot be used.
【About Links】
Links are free to use.