⚛️量子位•Stalecollected in 46m
Huawei Cloud Announces New Agentic AI Strategy
💡Understand how Huawei is integrating autonomous agents into enterprise cloud infrastructure.
⚡ 30-Second TL;DR
What Changed
Focus on Agentic AI architecture and deployment
Why It Matters
This signals a major shift for Huawei Cloud towards agent-based workflows, potentially impacting how enterprises build and deploy automated internal systems.
What To Do Next
Register for the Huawei Cloud Creators Conference to review the technical documentation on their new Agentic AI framework.
Who should care:Enterprise & Security Teams
Key Points
- •Focus on Agentic AI architecture and deployment
- •Integration of autonomous agents into cloud infrastructure
- •Strategic roadmap for enterprise-level AI agents
🧠 Deep Insight
Web-grounded analysis with 25 cited sources.
🔑 Enhanced Key Takeaways
- •Huawei Cloud's Agentic AI strategy is underpinned by a specialized "supernode architecture" featuring CloudMatrix384 supernodes and a MatrixLink network, optimized for the intensive computational and workflow demands of enterprise-level agentic AI, particularly for Mixture of Experts (MoE) models.
- •The strategy incorporates a platform named "Versatile" that manages the entire application lifecycle for AI agents, from development to deployment and ongoing management, seamlessly integrating with Huawei Cloud's existing AI compute models and data platforms.
- •Huawei is actively converging Agentic AI with digital twin technology, especially within the telecommunications sector, to establish a novel operational model where AI agents collaborate with human experts in a virtual, real-time network replica for advanced experimentation and closed-loop automation.
- •A significant component of Huawei's future vision includes the "Agentic-AI Core" (A-Core) for 6G networks, aiming to create self-programming networks that can autonomously generate, update, and execute control procedures, powered by a specialized large-scale network AI model called NetGPT.
- •The foundation for Huawei's autonomous agents is built upon its Pangu model family, which employs a three-layer architecture (L0 foundation models, L1 industry-specific models, and customer-specific models) offering capabilities in natural language processing, computer vision, multi-modal understanding, prediction, and scientific computing tailored for diverse industries.
📊 Competitor Analysis▸ Show
| Feature/Provider | Huawei Cloud Agentic AI | AWS (e.g., Bedrock Agents) | Microsoft Azure (e.g., AI Agent Service, Copilot Studio) | Google Cloud (e.g., Vertex AI Agent Builder) | Salesforce (Agentforce) | ServiceNow (AI Agents) |
|---|---|---|---|---|---|---|
| Core Offering | Full-stack AI-native cloud architecture, supernode infrastructure, Versatile platform for agent lifecycle, Pangu models. Focus on enterprise and industry-specific solutions. | Platform for building and deploying agents with foundation models (e.g., Claude, Llama 2), emphasizing security and data privacy. | Agentic cloud operations, AI-powered agents for workflow acceleration, Azure Copilot as agentic interface, governance features. | Cloud-native platform for building agents powered by Gemini models, emphasizing grounding in verified data sources. | Agentic AI layer of Salesforce platform, powered by Atlas Reasoning Engine, grounded in Data 360, CRM-native automation. | Agentic layer of ServiceNow Now Platform, targeting IT operations, HR, and customer service workflows. |
| Infrastructure | CloudMatrix384 supernodes, MatrixLink network, memory-centric AI-Native Storage, optimized for MoE models and NPU inference. | Leverages AWS's scalable cloud infrastructure. | Leverages Azure's global cloud infrastructure, with focus on hybrid solutions and edge. | Leverages Google Cloud's global infrastructure. | Integrated with Salesforce's existing cloud infrastructure. | Integrated with ServiceNow's Now Platform. |
| Model Integration | Pangu models (L0, L1, Pangu E/P/U/S series) for NLP, CV, multi-modal, prediction, scientific computing; industry-specific incremental training. | Supports various foundation models, allowing choice and flexibility. | Integrates with Azure AI models and services. | Powered by Gemini models. | Powered by Atlas Reasoning Engine. | Utilizes AI capabilities within the ServiceNow platform. |
| Target Use Cases | Enterprise-level agentic AI, industrial applications (cement, cultural tourism, manufacturing, energy, telecom networks), 6G. | General enterprise use cases, productivity, efficiency. | Accelerating development, migration, optimization in cloud operations, IT service management. | Solving deeply entrenched pain points across various domains for shared customers. | CRM-native agentic automation for sales and service workflows. | IT operations, HR service delivery, customer service management. |
| Key Differentiator | Full-stack AI-native approach with specialized hardware (supernodes, NPUs), deep industry integration, and focus on 6G network evolution. | Multi-model flexibility, deep integration with existing AWS cloud infrastructure. | Agentic cloud operations, governance at every layer, unified immersive experience via Azure Copilot. | Grounding agent responses in verified data sources, extensive partner ecosystem. | CRM-native agentic automation, reduced hallucination risk for sales/service. | Strong in ITSM, pre-built capabilities for asset management, knowledge base, self-service. |
🛠️ Technical Deep Dive
- Supernode Architecture: Huawei Cloud's strategy is built on a "supernode architecture" utilizing CloudMatrix384 supernodes connected via a MatrixLink network, creating a hybrid system that merges general-purpose and intelligent compute resources.
- Mixture of Experts (MoE) Models: This supernode structure is specifically optimized for Mixture of Experts (MoE) models, facilitating expert parallelism inference to reduce NPU (Neural Processing Unit) idle time during data transfers.
- Performance Enhancements: Huawei reports single-PU (Processing Unit) inference speed increases of four to five times compared to other models, and its infrastructure has shown a 20% improvement in training energy efficiency, exceeding industry standards by 10%.
- Memory-Centric AI-Native Storage: The hardware is complemented by a memory-centric AI-Native Storage system, designed to optimize for the typical access patterns of AI training and inference workloads.
- Pangu Models: The foundation for Huawei's agentic AI is the Pangu model family, which features a three-layer architecture:
- L0 (Foundation Models): Comprises five basic large models for Natural Language Processing (NLP), Visual, Multimodal, Prediction, and Scientific Computing.
- L1 (Industry-Specific Models): Consists of numerous industry-specific models trained using public and customer data from various sectors like government, finance, manufacturing, mining, and weather.
- Pangu Series: Includes specialized versions such as Pangu E (Embedded, 1 billion parameters for devices), Pangu P (Professional, 10 billion parameters for low-latency/cost reasoning), Pangu U (Ultra, 135-230 billion parameters for complex tasks), and Pangu S (Super, trillion-level parameters for cross-domain/multi-tasking).
- Incremental Training Workflows: Huawei Cloud has established processes for data preparation, incremental training, and evaluation to help companies build models tailored to their domains, which can boost model performance by 20% to 30%.
- Versatile Platform: This platform covers the entire application lifecycle for AI agents, from development to deployment, usage, and management, integrating with Huawei Cloud's AI compute models, data platforms, and tools.
- Agentic-AI Core (A-Core) for 6G: A blueprint for a 6G core network that utilizes specialized AI agents and NetGPT (a large-scale network AI model fine-tuned on telecommunication knowledge) to autonomously generate, update, and execute its own control procedures, involving mission-planning, mission-execution, and resource-management agents.
🔮 Future ImplicationsAI analysis grounded in cited sources
Huawei Cloud's Agentic AI strategy will significantly accelerate the adoption of autonomous operations in industrial and telecommunications sectors.
The focus on industry-specific models, supernode architecture for demanding AI workloads, and integration with digital twins and 6G networks directly addresses the complex needs for real-time, self-optimizing systems in these verticals.
The emphasis on full-stack AI-native cloud infrastructure will position Huawei as a key enabler for enterprises seeking to build and deploy highly efficient, resource-optimized AI agents.
By providing specialized hardware (NPUs, supernodes) and software frameworks (Pangu models, Versatile platform) optimized for agentic AI, Huawei aims to reduce the computational burden and complexity for enterprises.
Huawei's Agentic AI initiatives, particularly in 6G, will drive a paradigm shift in network management towards self-programming and intent-driven networks.
The Agentic-AI Core (A-Core) and AgenticCore solutions aim to enable networks to autonomously plan, execute, and adapt control procedures, moving beyond predefined rules to dynamic resource allocation based on AI agent needs.
⏳ Timeline
2018-10
Huawei launched its Full-Stack, All-Scenario AI Solution and formal AI strategy at HUAWEI CONNECT.
2021-09
Huawei introduced its Pangu models, designed to help industries go intelligent.
2023-07
Huawei introduced PanGu 3.0, a large language model tailored for enterprise sectors, utilizing Huawei Cloud solutions.
2025-06
Huawei Cloud announced Pangu Models 5.5 and an all-new AI Cloud Service, built on CloudMatrix 384 supernodes, at HDC 2025.
2025-10
Huawei Cloud showcased its AI-native cloud capabilities and supernode architecture for enterprise agentic AI at Huawei Connect 2025 and GITEX GLOBAL 2025.
2026-03
Huawei released the AgenticCore solution at MWC Barcelona 2026, integrating AI into mobile Internet, voice, O&M, and Telco Cloud infrastructure, and launched 115 industrial intelligence showcases.
📎 Sources (25)
Factual claims are grounded in the sources below. Forward-looking analysis is AI-generated interpretation.
- Google Search Source
- Google Search Source
- Google Search Source
- Google Search Source
- Google Search Source
- Google Search Source
- Google Search Source
- Google Search Source
- Google Search Source
- Google Search Source
- Google Search Source
- Google Search Source
- Google Search Source
- Google Search Source
- Google Search Source
- Google Search Source
- Google Search Source
- Google Search Source
- Google Search Source
- Google Search Source
- Google Search Source
- Google Search Source
- Google Search Source
- Google Search Source
- Google Search Source
📰
Weekly AI Recap
Read this week's curated digest of top AI events →
👉Related Updates
AI-curated news aggregator. All content rights belong to original publishers.
Original source: 量子位 ↗