#China #Shanghai #Huawei Cloud #AI Pioneers #CloudMatrix

Huawei Cloud Innovations at HUAWEI CONNECT 2025

At the recent HUAWEI CONNECT 2025 event in Shanghai, Zhang Ping'an, a board member and CEO of Huawei Cloud, delivered an impactful keynote presentation titled "All Intelligence Empowering AI Pioneers for Industries." This presentation highlighted Huawei’s latest advancements in artificial intelligence (AI) computing services and the transformative changes they promise for various industries.

The Power of AI Computing Services

One of the key announcements was the introduction of the AI computing service based on the Huawei CloudMatrix384 architecture. This new service dramatically expands the capabilities of Huawei’s supernodes, increasing their specifications from 384 to an incredible 8,192 cards. These supernodes can now support hyperscale clusters with capacity ranging from 500,000 to 1 million cards, offering unparalleled computing power essential for AI applications in the intelligence era.

Huawei also launched an Elastic Memory Service (EMS), providing a cutting-edge storage solution that enhances video RAM capabilities. This innovation significantly reduces latency in multi-turn conversation models, thus improving user experience.

To further support AI operations, Huawei has established fully liquid-cooled AI data centers in China's Guizhou, Inner Mongolia, and Anhui regions. These facilities can dissipate up to 80 kW per rack and achieve a Power Usage Effectiveness (PUE) of 1.1, optimizing energy efficiency while providing robust AI-driven operations and maintenance capabilities.

Simplifying AI Integration

Zhang Ping'an emphasized that Huawei's AI Token Service abstracts the underlying technical complexity, allowing users to directly access final AI computation results. This approach maximizes efficiency in leveraging inference computing capabilities. The CloudMatrix384 architecture allows seamless pooling of computation, storage, and resource systems, decoupling tasks and enhancing system performance, particularly in scenarios with varying latency requirements such as online, nearline, and offline inference tasks.

During the conference, the official launch of the AI Token Service powered by CloudMatrix384 was announced, promising superior performance and service quality to customers.

Supporting Businesses with Custom Models

Huawei Cloud has further refined its Pangu models by exploring industry-specific scenarios, collaborating closely with clients to tackle pressing challenges and redefine opportunities within their sectors. Utilizing openPangu, Huawei provides best practices for AI training and inference, making it easier for developers to harness AI computing power effectively. Additionally, the company is committed to continuously enhancing its Pangu models, increasing investments while studying industry scenarios to better meet client needs in their unique environments.

Pangu models have already been successfully deployed in over 500 scenarios across more than 30 industries, including key areas such as government services, finance, manufacturing, healthcare, and more, playing a crucial role in advancing intelligent transformations.

Innovations in Embodied AI

This year, Huawei Cloud introduced the CloudRobo Embodied AI Platform, which utilizes complex algorithms and intelligent logic in the cloud to create smarter robots. By leveraging the immense computing power and advanced AI models available in the cloud, the platform makes robotic operations more intelligent and adaptable to a wider range of scenarios. Moreover, to establish a secure and unified communication channel between robots and the cloud, the R2C protocol (Robot to Cloud) was introduced, with the first 20 partners already on board.

Kunpeng Cloud Services and Industry Innovations

One of Huawei Cloud's strategic priorities is the development of Kunpeng-based ARM cloud services, which ensure performance, security, and reliability. The number of Kunpeng computing cores in Huawei Cloud has increased significantly from 9 million to 15 million, demonstrating a 67% growth over the past year. Continuous enhancements have expanded compatibility with mainstream software, allowing adaptation for over 25,000 applications across diverse computing scenarios.

The GaussDB databases in Huawei Cloud utilize general-purpose computing supernodes, enabling layered pooling of computing, storage, and resource capabilities. This approach drastically improves performance for parallel transaction processing, surpassing traditional architectures by allowing concurrent read and write operations across nodes.

Towards a Universal and Optimized Cloud

Huawei Cloud has also rolled out a distributed cloud solution covering every scenario, from central regions to hotspot areas and edge locations. This ensures a consistent experience across various environments where customers operate. Moreover, Huawei has launched the Versatile platform, designed as an easy-to-use and effective space for developing and operating AI agents, significantly expediting the agent generation process for users.

As a follow-up to the keynote, Huawei Cloud hosted numerous additional events, including summit forums and round-table discussions, showcasing a wide array of innovative technologies across cloud infrastructure, large models, databases, AI agents, and embodied AI. These sessions highlighted how technology can facilitate the digital and intelligent transformation of industries, marking a significant step forward in Huawei Cloud's mission to empower AI pioneers and drive industry innovation.