What is Huawei Cloud's Agentic AI Supernode Infrastructure?

Huawei Cloud has engineered a new infrastructure and suite of tools designed to manage the specific computational and workflow demands of enterprise-level agentic AI.

Huawei Cloud's strategy is built on a supernode architecture, industry-specific model training and an agent platform that integrates with existing business systems, placing new demands on the underlying data centre architecture.

Zhang Yuxin, CTO of Huawei Cloud, outlined the resulting change in computing architecture at Huawei Connect 2025.

“Unlike traditional systems with their fixed processes and resources, agentic AI makes decisions independently”, Zhang says.

He adds: “It adapts dynamically, reshaping how computing systems interact and allocate resources.”

This evolution prompted Huawei Cloud to create a framework addressing the computational power and practical business requirements for implementing these advanced AI systems.

Supernode architecture for AI compute demands

The intense computational needs for training and inference in foundation models have highlighted limitations in some existing data centre architectures. Huawei Cloud’s AI Compute Service utilises CloudMatrix384 supernodes connected via a MatrixLink network.

This configuration creates a hybrid system that merges general-purpose and intelligent compute resources.

The supernode structure is targeted specifically at Mixture of Experts (MoE) models. By facilitating expert parallelism inference, this design could reduce the time NPUs are idle during data transfers.

According to Huawei, Huawei has recorded single-PU inference speed increases of four to five times compared to other models.

This hardware is paired with a memory-centric AI-Native Storage system designed for the access patterns typical in AI training and inference workloads. ModelBest has integrated its MiniCPM series with this infrastructure.

When running on Huawei’s systems, training energy efficiency reportedly improved by 20% with performance exceeding industry standards by10%.

Incremental training for industry-specific models

Creating effective AI models for specific sectors requires more than just access to powerful compute.

Huawei Cloud has established processes for data preparation, incremental training and evaluation to help companies build models tailored to their domains.

The incremental training workflow modifies parameters according to core model characteristics and industry goals, which Huawei states can boost model performance by 20% to 30%. This method has been applied by Shaanxi Cultural Industry Investment Group for cultural tourism.

Working with Huawei, the group consolidated datasets covering historical film and intangible heritage.

Huang Yong, Chairman of Shaanxi Cultural Industry Investment Group, described the project's scope.

“Using Huawei Cloud’s data-AI convergence platform, the group combined diverse cultural tourism data to create comprehensive datasets across areas like history, film and intangible heritage, strengthening Shaanxi’s cultural tourism foundation”, Huang says.

The collaboration produced the Boguan cultural tourism model, which is now underpins tools like a cultural tourism intelligent brain and a smart management assistant.

The versatile platform for enterprise deployment

Enterprise AI agents must integrate with established workflows, manage complex situations spanning multiple systems and adhere to corporate standards for reliability and auditability.

To meet these needs, Huawei Cloud developed Versatile, a platform covering the application cycle from development to deployment, usage and management. It integrates with Huawei Cloud’s AI compute models, data platforms and tools.

Conch Group has used this approach to create agents for the cement industry, building models for production operations and management.

The system predicts clinker strength at three and 28 days with deviations of less than 1MPa from actual results, achieving over 90% accuracy.

For cement calcination, the model can generate process parameters that reduce standard coal usage by 1% compared to class A energy efficiency standards. Xu Yue, Assistant to Conch Cement's General Manager, discussed the industry-wide implications.

“The model’s success with quality control production optimisation, equipment management and safety lays the groundwork for end-to-end collaboration and decision-making for cement agents”, Xu says.

He adds: These advanced agents are moving the cement industry from relying on traditional expertise to being fully guided by data across all processes.”