Huawei Cloud's Innovations Propel AI Forward at HUAWEI CONNECT 2025

On Day Two of HUAWEI CONNECT 2025, Huawei Cloud’s CEO, Zhang Ping'an, took center stage to discuss his vision for the future of AI and cloud computing. Under the theme of "All Intelligence Empowering AI Pioneers for Industries," he outlined how Huawei Cloud is reshaping the AI landscape with its cutting-edge compute services and innovative solutions.

At the event, Huawei Cloud unveiled its AI Compute Service powered by the newly enhanced CloudMatrix384. This marks a significant leap forward in cloud computing capabilities, boosting the specifications from 384 cards to a staggering 8,192 cards. This configuration supports hyperscale clusters ranging from 500,000 to 1 million cards, thereby delivering unprecedented AI computing power. Alongside this, the introduction of the Elastic Memory Service (EMS) aims to minimize latency in multi-turn interactions, enhancing overall user experience.

Zhang highlighted that fully liquid-cooled AI data centers have been established in strategic locations such as Guizhou, Inner Mongolia, and Anhui. These facilities operate at an impressive heat dissipation capacity of 80 kW per cabinet while achieving a Power Usage Effectiveness (PUE) of only 1.1. The AI data centers provide efficient AI computation without necessitating extensive infrastructure changes for enterprises, requiring only a pair of optical fibers for connectivity.

Another ground-breaking initiative presented was the AI Token Service, designed to remove the complexities that often accompany technical processes. This service affords users direct access to the final AI computing results, allowing them to efficiently harness inference computing power. The CloudMatrix384 supernode optimally pools compute, memory, and storage resources and converts traditional serial tasks into distributed parallel tasks, vastly improving system performance.

Additionally, the AI Token Service enhances inference performance, outperforming previous generations by three to four times in various scenarios including online and offline inference tasks, a substantial upgrade over the capabilities of the earlier H20 model.

Tailoring AI Solutions for Diverse Industries

Huawei Cloud's commitment to addressing specific industry needs was evident in its development of the Pangu Models. By collaborating closely with clients, Huawei aims to dismantle existing barriers and reimagine possibilities within several sectors. The open-source Pangu framework provides industry best practices for rapid AI training and inference. As Zhang stated, the aim is to empower clients to create their own models tailored to their unique challenges, accelerating the pace of smart transformation across different industries.

These Pangu Models have been successfully implemented in over 500 scenarios across more than 30 distinct sectors, including essential services such as finance, healthcare, and autonomous driving, showcasing their versatility and potency.

Revolutionizing Robotics with Cloud Power

Another notable launch was the CloudRobo Embodied AI Platform, which places complex algorithms and enhanced intelligent logic within the cloud realm, enhancing the performance and applicability of robots. By leveraging robust cloud-based computing power and advanced AI models, this platform aims to transcend the limitations currently faced by robots in varied environments.

To facilitate seamless communication between robots and the cloud, Huawei Cloud introduced the Robot to Cloud (R2C) Protocol, marking a significant advancement in robotic technology that encourages collaboration among partners.

Kunpeng Cloud Services and Beyond

Huawei Cloud remains committed to creating Kunpeng-powered ARM cloud services that combine performance with security. The ambitious initiative has led to an increase in Kunpeng compute cores from 9 million to 15 million, an increase of 67% within a year. This performance boost supports myriad applications ranging from databases to transcoding.

In terms of data management, Huawei’s GaussDB databases utilize supernodes for pooled compute resources, enhancing data handling capabilities. The performance of GaussDB clusters was illustrated by their ability to process 5.4 million transactions per minute, achieving a remarkable 2.9-fold increase over traditional setups.

A Comprehensive Cloud Experience

Huawei Cloud has also implemented an all-encompassing distributed cloud solution, comprising diverse offerings such as CloudOcean, CloudSea, CloudLake, and CloudPond, designed for various operational environments. This ensures that clients enjoy a consistent cloud computing experience, irrespective of location.

In addition to these announcements, the event featured various sessions including summit forums and roundtables where Huawei Cloud and its partners shared insights into other cutting-edge technologies. As they explore the future of cloud infrastructure, large models, and embodied AI solutions, Huawei Cloud continues to illustrate its pivotal role in the ongoing digital transformation across all industry sectors.