On the second day of HUAWEI CONNECT 2025, Zhang Ping’an, Huawei’s Executive Director of the Board and CEO of Huawei Cloud, delivered a keynote titled “All Intelligence: Empowering AI Pioneers for Industries.”
He outlined Huawei Cloud’s work in AI compute services, foundation models, embodied AI, and AI agents.
Huawei Cloud announced its AI Compute Service powered by CloudMatrix384. The CloudMatrix supernode will be upgraded from 384 cards to 8,192 cards and can support clusters running on 500,000 to one million cards.
The company also introduced Elastic Memory Service (EMS), which expands video RAM with memory to reduce latency in multi-round conversations with foundation models.
Huawei has deployed liquid-cooled AI data centres in Guizhou, Inner Mongolia, and Anhui.
These support 80 kW heat dissipation per cabinet, achieve a power usage effectiveness (PUE) of 1.1, and provide AI-enabled operations and maintenance.
According to Huawei, enterprises only require optical fibre connections to access these services, avoiding the need for new data centre construction.
Zhang said Huawei Cloud’s AI Token Service “abstracts away the underlying technical complexity and directly provides users with the final AI computing results.”
The CloudMatrix384 supernode pools compute, memory, and storage resources, and converts serial tasks into distributed parallel tasks.
For inference tasks with different latency requirements, the system delivers average inference performance per card three to four times higher than H20, he noted.
The AI Token Service powered by CloudMatrix384 was officially launched at the event.
Huawei Cloud continues to develop its Pangu Models, both open-source and closed-source, to address sector-specific use cases.
Pangu Models have been applied in more than 500 scenarios across over 30 industries, including finance, manufacturing, healthcare, mining, transport, autonomous driving, and meteorology.
The company also introduced the CloudRobo Embodied AI Platform, which runs algorithms and logic on the cloud to support more lightweight robots.
To enable communication between robots and the cloud, Huawei launched the Robot to Cloud (R2C) Protocol, with 20 initial partners onboarded.
Huawei Cloud is expanding its Kunpeng-powered ARM cloud services, with compute cores increasing from 9 million to 15 million over the past year.
In databases, Huawei Cloud reported that GaussDB clusters based on supernodes allow pooled compute, memory, and storage resources, and support multi-read and multi-write on all nodes.
A supernode-based GaussDB cluster can process 5.4 million transactions per minute, a 2.9-fold increase over non-supernode clusters.
Featured image credit: Huawei