Huawei Cloud unveiled a new set of enterprise AI products at HUAWEI CONNECT 2026, bringing computing, memory, models, agent development and industry applications under what it calls “Agentic Infra.”
Dr. Peter Zhou, director of the board at Huawei and CEO of Huawei Cloud, said infrastructure in the agentic AI era must go beyond providing compute. The company says its approach focuses on improving token efficiency, expanding memory, coordinating general-purpose and AI computing, and operating agents securely in production.
For channel partners, the expanded stack could create implementation, customization, and managed-service opportunities as enterprises move AI agents from pilot projects into production. However, partners will need to weigh those opportunities against interoperability concerns and greater customer dependence on Huawei’s ecosystem.
Huawei said its Agentic Infra has already served more than 3,500 customers. At the center is the latest AI Cluster Service, or AICS. Huawei says its five-level recovery system and full-chain observability allow cloud training runs to remain stable for more than 40 days and recover from faults within 10 minutes. It also claims 20% higher token throughput than its previous-generation compute service.
The company, however, did not provide independent benchmark results for that performance comparison. Although Huawei described AICS as globally launched, the company said commercial availability will begin in China on Sept. 30 and in markets outside China on Nov. 30.
Memory becomes part of the agent stack
To support agents performing long-running tasks that demand the retention and retrieval of massive datasets, Huawei introduced Context Memory Storage (CMS).
The system features petabyte-scale storage along with terabyte-scale memory read capabilities. Huawei claims CMS provides twice the storage capacity of comparable industry products and 50% higher performance than what it calls “industry peers.” The company did not identify the products used for those comparisons. Huawei designed CMS to address the context and memory bottlenecks that can emerge as AI agents execute complex, extended workflows. While maintaining accessible, relevant context can minimize redundant processing, these performance claims require independent validation.
Agents move from experiments to enterprise systems
Huawei is advancing its enterprise agent platform through two editions: the commercial AgentArts platform and its open-source counterpart, openJiuwen.
The platform provides access to more than 5,000 general MCP assets and over 1,000 industry-specific assets. Huawei says more than 100 customers are using its enterprise agent platform, including Kingsoft Office, which has integrated it with WPS 365 Document Center and WPS Comate for industry-specific office agents.
Huawei is also bringing the platform to market through partners. The company said Chinasoft International, iSoftStone Group, and Beiming Software have launched commercial partner editions, giving integrators a potential route to package its infrastructure and agent tools into industry-specific offerings.
AgentArts is scheduled to become commercially available outside China on Dec. 30. Huawei says openJiuwen has passed 50,000 stars and 3.29 million downloads.
What this means for businesses and users
By consolidating infrastructure, context memory, model access, development tools, and industry assets, Huawei could give partners a more unified foundation for building and deploying enterprise agents.
That approach may reduce the work required to integrate systems from multiple vendors, but it could also deepen customer dependence on Huawei’s infrastructure and development tools. Partners evaluating the platform should consider interoperability, regional availability, data governance, independently verified performance, and whether workloads can be moved to another environment.
For end users, the benefits are likely to be indirect. Effective implementation could enable AI agents to execute extended, multi-step tasks without context degradation, while providing enterprise applications with increasingly specialized AI capabilities.
In parallel, Huawei is broadening its Industry AI Foundry with dedicated Smart Government and AI Hardware zones, complementing established offerings in healthcare, embodied AI, scientific computing, manufacturing, and finance. The foundry currently hosts over 1,000 industry assets and backs more than 1,000 active projects.
Huawei’s next test will be whether its international rollout generates meaningful adoption beyond China. For channel partners, the opportunity will depend on regional availability, customer demand, partner support, and whether Huawei can substantiate its performance claims through independent testing. Those factors will determine whether Agentic Infra becomes a practical foundation for partner-led AI services or remains primarily a Huawei-centered ecosystem.