TLDR: At HUAWEI CONNECT 2025, Huawei Cloud’s CEO, Zhang Ping’an, announced significant advancements in AI compute services, including the upgraded CloudMatrix supernode, the new AI Token Service, and the CloudRobo Embodied AI Platform. The company is focusing on providing robust computing infrastructure and industry-specific Pangu Models to accelerate AI adoption across various sectors.
SHANGHAI – On September 23, 2025, at the HUAWEI CONNECT 2025 event, Zhang Ping’an, Huawei’s Executive Director of the Board and CEO of Huawei Cloud, delivered a pivotal keynote speech titled ‘All Intelligence: Empowering AI Pioneers for Industries’. His address highlighted Huawei Cloud’s latest innovations and strategic initiatives in AI compute services, foundational models, embodied AI, and AI agents, underscoring the company’s commitment to fostering intelligent transformation across diverse industries.
Revolutionizing AI Compute with CloudMatrix384 and AI Token Service
A cornerstone of Huawei Cloud’s announcements was the significant upgrade to its AI Compute Service, powered by CloudMatrix384. The specifications of the Huawei CloudMatrix supernode are set to expand dramatically, from 384 cards to an impressive 8,192 cards. These enhanced supernodes are designed to support hyperscale clusters capable of running between 500,000 to 1 million cards, thereby delivering unparalleled AI computing power essential for the intelligent era.
Further enhancing its compute offerings, Huawei Cloud introduced the Elastic Memory Service (EMS), an industry-first innovation that expands video RAM with memory. This breakthrough is expected to substantially reduce latency in multi-round conversations on foundation models, leading to a significantly improved user experience. Zhang Ping’an also officially launched the AI Token Service, powered by CloudMatrix384. This service aims to abstract away complex underlying technicalities, directly providing users with final AI computing results and enabling highly efficient utilization of inference computing power. The CloudMatrix384 supernode achieves full pooling of compute, memory, and storage resources, decoupling tasks and converting serial processes into distributed parallel tasks, which boosts inference performance by 3 to 4 times compared to H20 in various inference scenarios.
Sustainable and Efficient AI Data Centers
Huawei Cloud has strategically deployed fully liquid-cooled AI data centers across China in Guizhou, Inner Mongolia, and Anhui. These state-of-the-art facilities boast 80 kW heat dissipation per cabinet, achieve a low Power Usage Effectiveness (PUE) of 1.1, and incorporate AI-enabled Operations and Maintenance (O&M). This infrastructure allows enterprises to connect via optical fibers, bypassing the need for traditional data center reconstruction and providing immediate access to efficient AI compute and full-stack dedicated AI cloud services.
Empowering Industries with Pangu Models
Huawei Cloud continues to refine its Pangu Models, tailoring them for industry-specific applications. The company is actively collaborating with customers to address critical challenges and accelerate intelligent transformation. Huawei utilizes openPangu to offer best practices for AI training and inference, simplifying the efficient use of AI computing power for developers. Concurrently, Huawei is developing a closed-source Pangu Model, signaling ongoing investment to deepen understanding of industry scenarios and support customers in building their own specialized models. Pangu Models have already demonstrated significant impact, being applied in over 500 scenarios across more than 30 industries, including government services, finance, manufacturing, healthcare, coal mining, steel, railways, autonomous driving, and meteorology.
Advancing Robotics and Data Foundations
Moving beyond traditional terminals, Huawei Cloud launched the CloudRobo Embodied AI Platform. This platform deploys sophisticated algorithms and intelligent logic on the cloud, enabling more lightweight and intelligent robots by leveraging massive computing power and advanced AI models. To ensure seamless and secure communication between robots and the cloud, Huawei Cloud also introduced the Robot to Cloud (R2C) Protocol, with 20 partners already onboard.
In terms of data infrastructure, Huawei Cloud’s Kunpeng Cloud Services, powered by ARM, continue to expand, offering enhanced performance, security, and reliability. The number of Kunpeng compute cores on Huawei Cloud has surged by 67% in the past year, reaching 15 million. The Kunpeng platform’s compatibility with over 25,000 applications supports a wide range of general-computing scenarios. Furthermore, the GaussDB databases, built on general-purpose computing supernodes, feature layered pooling of resources and support multi-read and multi-write capabilities, processing 5.4 million transactions per minute—a 2.9-fold increase over non-supernode clusters.
Ubiquitous Distributed Cloud and Agent Development
Huawei Cloud’s comprehensive distributed cloud solution, encompassing CloudOcean, CloudSea, CloudLake, and CloudPond, ensures ubiquitous and optimized compute with local access across central regions, hotspot areas, and edge sites. The company also unveiled Versatile, an enterprise-grade agent platform designed to be an easy-to-use, effective, and open environment for developing and running AI agents. This platform streamlines agent generation, allowing users to create AI agents for specific application scenarios by simply providing business descriptions and flowcharts.
Also Read:
- Huawei and Partners Showcase Over 30 AI-Powered Digital Transformation Benchmarks at HUAWEI CONNECT 2025
- Oracle to Showcase Enterprise-Ready AI Innovations at GITEX 2025, Bolstering UAE’s $96 Billion AI Economy
These announcements at HUAWEI CONNECT 2025 underscore Huawei Cloud’s strategic vision to provide a robust, innovative, and open ecosystem for AI development, driving digital and intelligent transformation across industries.
