• Huawei Cloud launched its latest AICS.
• Huawei expanded its enterprise AI agent platform.

Huawei Cloud has introduced a new version of its AI Cluster Service (AICS) and expanded its enterprise agent platform as part of announcements made during the second day of HUAWEI CONNECT 2026.

Peter Zhou, director of the board at Huawei and CEO of Huawei Cloud, announced the services during his keynote. The announcements also covered context memory storage, model services, two new areas within Huawei’s Industry AI Foundry, and plans to make AgentArts available outside China.

AICS expands Huawei’s AI infrastructure

The latest AICS is designed for large-scale AI training and inference workloads. Huawei said the service uses a five-level recovery mechanism and full-chain observability to identify and respond to infrastructure faults.

According to Huawei, AICS has supported model training runs lasting more than 40 days and can recover from faults within 10 minutes. The company also said optimisations across scheduling, caching, and algorithms deliver 20% higher token throughput than the previous generation of the service.

Huawei did not provide independent benchmark results for the performance comparison.

Huawei first disclosed AICS in June, when it said the service uses its UnifiedBus interconnect and can support clusters containing more than 100,000 cards. UnifiedBus provides the high-bandwidth connections used to link computing resources across the cluster.

Huawei is increasingly connecting large numbers of processors into computing systems to compensate for limitations in individual chips. The approach forms part of the company’s broader Ascend AI computing roadmap.

AICS is scheduled to become commercially available in China on September 30, followed by markets outside China on November 30.

The international rollout comes as Huawei faces production constraints in its AI computing business. “Since we don’t have enough capacity to even satisfy the demand in China,” Huawei rotating chairman Eric Xu said, adding that the company does not plan a full-scale international expansion of its AI computing business for now.

Huawei also introduced Context Memory Storage, or CMS, for AI agents that need to retain and retrieve large volumes of contextual information during longer-running tasks. The company said CMS provides petabyte-scale storage and supports terabyte-scale memory reads.

Context memory systems can be used to store key-value, or KV, cache generated during model inference. KV cache retains previously computed information from a model’s context, reducing the amount of data that has to be recalculated as processing continues.

Separately, Huawei introduced its OceanStor M900 Context Memory Storage on the first day of HUAWEI CONNECT. The system extends KV cache from on-chip memory and DRAM to SSD storage, with Huawei saying UnifiedBus enables shared, tiered KV-cache storage of up to 64 PB in a single cluster.

Huawei claimed the CMS service offers twice the storage capacity and 50% higher performance than comparable products, although it did not disclose the products or benchmark methodology used for the comparison.

AICS and CMS sit within what Huawei calls Agentic Infra, which combines AI and general-purpose computing, storage, scheduling, and security capabilities for running AI agents. Huawei said Agentic Infra has served more than 3,500 customers.

AgentArts and openJiuwen broaden the agent stack

Huawei is also expanding AgentArts, its platform for developing and managing enterprise AI agents.

The platform provides access to more than 5,000 general Model Context Protocol assets and more than 1,000 industry-specific MCP assets, according to Huawei. MCP provides a standardised way for AI applications and agents to connect with external data sources and tools.

Huawei said its enterprise agent platform has served more than 100 customers, including government organisations, utilities, research institutions, software companies, and healthcare organisations in China. AgentArts is scheduled to become commercially available in markets outside China on December 30.

Huawei is developing the platform alongside openJiuwen, its open-source agent framework.

Public openJiuwen repositories show that the framework includes Python and Java development kits, a distributed runtime, visual development tools, long-term memory components, and support for protocols including MCP and agent-to-agent communication. Its Apache 2.0-licensed core SDK supports workflow orchestration, tool calling, state management, debugging, and observability, while AI models are connected separately.

Huawei said the openJiuwen community has surpassed 50,000 stars and recorded 3.29 million downloads, although it did not specify how those figures were calculated.

Kingsoft Office demonstrated its use of Huawei Cloud’s agent platform at the event. The company has integrated the technology with WPS 365 Document Center and WPS Comate to develop office agents deployed in industries including finance and government.

Huawei also outlined its Agentic Model as a Service platform, which provides access to models from multiple providers without requiring customers to deploy the models themselves. MiniMax demonstrated its multimodal model with Huawei Cloud at the event, although Huawei did not disclose further commercial details for the MaaS service.

Huawei extends industry and partner deployment programmes

Huawei Cloud has also expanded its Industry AI Foundry, a collection of industry-specific models, datasets, knowledge resources, agents, and implementation assets.

Huawei said the programme contains more than 1,000 industry assets and supports more than 1,000 deployed projects across healthcare, embodied AI, AI for science, manufacturing, and finance. The company did not provide a breakdown of the projects by deployment stage or scale.

Huawei added Smart Government and AI Hardware zones during the event.

The Smart Government Zone has 24 founding partners covering areas including government operations, public services, urban governance, and industry development. The AI Hardware Zone has 15 core partners and covers more than 20 device types, including AI glasses, AI toys, and AI recording cards.

Huawei said the AI Hardware Zone also includes more than 10 scenario templates and over 110 scenario skills.

Huawei separately introduced SCALE, a partner support programme covering solution development, testing, marketing, service delivery, and collaboration.

Leo Chen, Huawei senior vice president and president of Enterprise Sales, said the programme is intended to help partners take individual AI projects into wider deployments. “We want to help partners solve customer issues more efficiently and help customers achieve large-scale AI adoption,” Chen said.

Huawei is making reference architectures available for 48 scenarios, alongside testing and verification environments through its cloud-based OpenLab. The company said partners have used its products to develop more than 100 scenario-based solutions across eight industries.

Huawei is also providing access to its eFly and ePartner platforms for marketing, transactions, service operations, documentation, and technical tools. Its O3 Partner Service Enablement Platform provides partners with technical resources and service tools, while Huawei said its training and certification programmes have supported more than 50,000 AI professionals over the past three years.

Want to learn more about AI and big data from industry leaders? Check out AI & Big Data Expo taking place in Amsterdam, California, and London. The comprehensive event is part of TechEx and is co-located with other leading technology events, click here for more information.

Tech Wire Asia is powered by TechForge Media. Explore other upcoming enterprise technology events and webinars here.