#China #Shanghai #Cloud Computing #AI Technologies #Huawei Cloud

Huawei Cloud Innovations at HUAWEI CONNECT 2025

On September 21, 2025, at the second day of HUAWEI CONNECT 2025, Zhang Ping'an, the Executive Director of Huawei's Board and CEO of Huawei Cloud, delivered a keynote speech titled "All Intelligence Empowering AI Pioneers for Industries." He detailed the company's breakthroughs in AI computing services, core models, embodied AI technologies, and more. These advancements mark a significant leap in Huawei's commitment to innovate and enhance computational capabilities for industries in the intelligent era.

AI Compute Service Enhancements

This year, Huawei Cloud introduced the AI Compute Service built on the CloudMatrix384 technology. The specifications of the Huawei CloudMatrix supernode have expanded, increasing from 384 to 8,192 cards. This allows for hyper-scaled clusters capable of supporting between 500,000 to 1 million cards, creating a robust AI computing infrastructure crucial to the smart age. Furthermore, Huawei Cloud launched Elastic Memory Service (EMS), the industry's first service to extend video RAM memory, significantly reducing latency for multi-turn conversations on base models, which enhances user experience.

In attempts to push boundaries further, Huawei Cloud has deployed liquid-cooled AI data centers in several Chinese provinces, including Guizhou, Inner Mongolia, and Anhui. These data centers can dissipate up to 80 kW of heat per cabinet, lowering the Power Usage Effectiveness (PUE) to 1.1. This innovative approach reduces the need for companies to reconstruct traditional data centers, enabling them to connect to these AI-efficient services using merely optical fibers.

AI Token Service Launch

Zhang emphasized the launch of Huawei Cloud's AI Token Service that simplifies technical complexities and delivers direct computational outputs to users. It enables the optimal use of inference computing power, with the CloudMatrix384 achieving a performance boost that allows for greater efficiency in handling inferencing tasks. In scenarios with varying latency requirements such as online, nearline, and offline inference, CloudMatrix384 manages to deliver inference performance 3 to 4 times higher than previous models like H20.

Advancing with Pangu Models

Huawei Cloud is refining its Pangu models by delving deep into specific industry scenarios to collaborate with clients on resolving their most pressing challenges, redefining possibilities within their sectors. The openPangu initiative is providing proven practices for AI training and inference, allowing developers to maximize AI computational efficiency. Alongside the open-source model, there are plans to develop a closed-source version to continuously enhance investment in Pangu models, better comprehend customer needs, and aid them in creating tailored models for specific industries, thus accelerating smart transformation.

Pangu models have already been adopted in over 500 scenarios across 30 industries, prominently featuring in public administration, finance, manufacturing, healthcare, and more.

CloudRobo: Embodied AI Innovations

In 2025, Huawei Cloud also launched the CloudRobo Embodied AI Platform, designed to enable the development of lighter robots by leveraging comprehensive algorithms and intelligent cloud logic. This platform allows for more intelligent robotic actions, overcoming previous limitations and facilitating deployment across various scenarios. Huawei introduced the R2C (Robot to Cloud) protocol to establish a secure communication channel between robots and the cloud, announcing official partnerships with 20 initial partners in this initiative.

Kunpeng Cloud Services: Driving Industrial Innovation

A core strategy behind Huawei Cloud's growth centers on developing Kunpeng-powered cloud services that ensure strong performance, security, and reliability. Over the past year, the number of Kunpeng computing cores within Huawei Cloud has tremendously increased from 9 to 15 million, representing a 67% growth. The Kunpeng platform aims to improve compatibility with common software, having been tailored for over 25,000 applications, thus offering robust support for various computational scenarios.

Building Next-Gen Databases with GaussDB

Huawei Cloud's GaussDB databases leverage universal computing supernodes, enabling layered sharing of computing, memory, and storage resources, allowing simultaneous reads and writes on any node. This architecture enhances the capacity for concurrent transaction processing significantly, as evidenced by GaussDB's ability to manage 5.4 million transactions per minute, a marked increase over conventional clusters.

Comprehensive Distributed Cloud Solutions

Huawei Cloud's distributed cloud solution encompasses various scenarios, including CloudOcean, CloudSea, CloudLake, and CloudPond. This vast infrastructure assures that Huawei Cloud services are accessible wherever customer business activities are conducted.

Versatile AI Agent Platform

Launching the Versatile AI agents platform, Huawei Cloud aims to provide companies an efficient and open framework for developing AI-driven agents. Users can simplify agent creation by preparing descriptions and flowcharts, significantly improving development efficiency.

In addition to these major announcements, Huawei Cloud engaged in numerous forums to discuss a wide range of innovative technologies aiding digital transformation across industries, including cloud infrastructure, large models, databases, AI agents, and embodied AI technologies.