Huawei Cloud has officially launched its new AI Cluster Service (AICS) globally, aiming to reinforce the foundation for agentic AI with enhanced fault recovery and token throughput. This update coincides with broader expansions in memory storage, open-source agent development, and specialized industry zones designed to support long-horizon tasks and secure autonomy.
Key Takeaways
- Huawei Cloud’s new AI Cluster Service (AICS) delivers 20% higher token throughput compared to previous generations and supports stable training for over 40 days.
- The service features a five-level fast recovery mechanism capable of fault recovery within 10 minutes, supported by full-chain observability.
- Context Memory Storage (CMS) offers petabyte-scale memory space with twice the capacity of industry peers and 50% higher read performance.
- The Agentic Infra paradigm has served over 3,500 customers since its launch in June 2026, focusing on efficient tokens and unified scheduling.
- Enterprise-grade agent platform AgentArts will be commercially available outside China on December 30, while its open-source version, openJiuwen, has reached 3.29 million downloads.
- Two new zones have been added to the Industry AI Foundry: Smart Government and AI Hardware, bringing the total number of specialized zones to seven.
Accelerating Agentic Infrastructure with AICS
The newly announced AI Cluster Service (AICS) incorporates a five-level fast recovery mechanism combined with full-chain observability. According to Huawei Cloud, this architecture allows for fault recovery within 10 minutes and supports stable training on the cloud for more than 40 days. By coordinating optimization across scheduling, cache, and algorithms, the company states that AICS delivers 20% higher token throughput when compared against the previous generation of compute services.
Commercial availability for AICS is scheduled for September 30 in China, followed by markets outside China on November 30. This service is part of a broader "Agentic Infra" paradigm introduced in June 2026. This paradigm focuses on efficient tokens, enhanced memory, unified general and AI scheduling, and secure autonomy. To date, this infrastructure approach has served more than 3,500 customers.
Dr Peter Zhou emphasized the shift in infrastructure purpose during the event: “In the agentic era, infrastructure no longer simply exists to provide compute. Instead, its purpose has expanded to making every token more efficient, coordinating general and AI compute, enabling models to keep learning, and, more importantly, running agents securely and reliably within real production environments.”
Memory Storage and Model Services
To address the memory capacity and access efficiency requirements of long-horizon agent tasks, Huawei Cloud introduced its Context Memory Storage (CMS) solution. This system features a petabyte-scale memory space. The company reports that CMS delivers twice the storage capacity of comparable industry products and supports high-speed terabyte-scale memory reads with 50% higher performance than industry peers.
Additionally, Huawei Cloud highlighted its Agentic Model as a Service (MaaS) offering. This allows developers to use state-of-the-art models from leading providers with a single click, eliminating the need for complex deployment processes. During the event, MiniMax, a Shanghai-based company developing proprietary large language models across various modalities, demonstrated an advanced multimodal model combined with Huawei Cloud to deliver open multimodal capabilities. Rhea He, Head of Market and Commercial Operations at MiniMax, showcased an AI-generated professional quality video during the presentation, noting that it required zero manual editing.
Open Source Agent Development and Enterprise Adoption
Huawei Cloud is advancing its enterprise-grade agent offering through both commercial and open-source approaches. The platform includes AgentArts, an enterprise-grade agent platform with full harness engineering capabilities, and its open-source edition, openJiuwen. This offering provides access to more than 5,000 general Model Context Protocol (MCP) assets and more than 1,000 industry-specific MCP assets to help enterprises develop, use, and manage agents.
The platform has served more than 100 customers to date, including China Southern Power Grid, Guangzhou Laboratory, KingSoft Office, and KingMed Diagnostics. At the HUAWEI CONNECT event, Wu Qingyun, Vice President of KingSoft Office and General Manager of Enterprise Business Division, discussed how KingSoft is working with Huawei Cloud to drive agent adoption across industries. KingSoft Office integrated Huawei Cloud’s agent platform with WPS 365 Document Center and WPS Comate to build vertical office agents for multiple industries, including the finance and government sectors.
Commercial availability for AgentArts in markets outside China is planned for December 30. The openJiuwen open-source community has accumulated more than 50,000 stars and 3.29 million downloads. Huawei Cloud has also launched partner commercial editions alongside Chinasoft International, iSoftStone Group, and Beiming Software.
Expanding Industry AI Zones
The Industry AI Foundry brings together industry scenarios, models, data, knowledge, and agents to turn learnings from isolated industry projects into reusable capabilities. The foundry has accumulated more than 1,000 industry assets across more than 1,000 deployed projects. Huawei Cloud has added two new industry-specific zones: the Smart Government Zone and the AI Hardware Zone, bringing the total number of zones to seven.
The Smart Government Zone features 24 founding partners and covers public services, urban governance, and government office operations. The AI Hardware Zone has onboarded 15 core partners covering more than 20 device types—including AI glasses and AI recording cards—alongside more than 10 scenario templates and more than 110 scenario skills. Existing zones already established by the business unit cover smart healthcare, embodied AI, AI for science, smart manufacturing, and smart finance.