Dr. Peter Zhou, Director of the Board at Huawei and CEO of Huawei Cloud, pointed out that the era of agentic AI is driving a fundamental shift in computing paradigms.
At this event, Huawei Cloud officially introduced Agentic Infra, a new paradigm featuring an efficient token factory, continuous learning, unified general & AI compute scheduling, and secure autonomy. Four new Agentic Infra products were also released:
- AI Cluster Service (AICS): Built on the ultra-high bandwidth UnifiedBus (UB) network, it supports clusters with over 100,000 cards, delivering a total computing power of up to 200 EFLOPS. The latency for token generation is reduced to less than 10 milliseconds, with a throughput of 5 million tokens per second across 1,000 cards. The online service availability reaches 99.95%, making it a token factory with the ultimate efficiency.
- Agentic Memory Storage (AMS): The solution leverages NPU passthrough to Context Memory Storage (CMS) hardware, creating a PB-scale memory space. It also supports tiered KV-cache pooling, which reduces inference costs while enabling multi-day long-running tasks, breaking the memory bottleneck of agents and facilitating continuous learning for agents.
- CCE Volcano Next unified general & AI scheduling engine: It achieves innovation in unified general-purpose and AI workload scheduling through "shared training-inference pooling + fragmentation consolidation", improving resource utilization by over 30%.
- AgentSphere: It delivers a secure and autonomous agent runtime environment, providing a secure autonomous runtime foundation with ultimate elasticity and proactive intent protection. Leveraging ultra-lightweight sandbox technology, it achieves fast startup within 100 milliseconds and the ability to batch-create hundreds of thousands of instances per minute, helping agents scale securely and efficiently on the cloud.
ModelArtsNext: The new-generation model platform to bring foundation models into enterprise scenarios
ModelArts Next is a model training and inference platform that provides four core capabilities: Reinforcement Learning as a Service (RLaaS), confidential inference, model routing, and model matrix. MaaS model routing supports three policies: experience-first, efficiency-first, and balanced mode, and dynamically routes each request to the optimal model based on its characteristics. To date, more than 15 state-of-the-art (SOTA) model services have been provided, with a model scheduling accuracy of over 95% and an average reduction of 20% in calling costs. The enterprise-grade RLaaS service enables reinforcement learning as a core capability that can be invoked by every enterprise. It allows users to create tasks in just one minute, achieve end-to-end visualization, and ensure consistency between training and inference. This enables large models to be applied to more specific scenarios and become smarter with each use.
AgentArts: Enterprise-grade agent platform now available for OBT and committed to open source
Huawei Cloud AgentArts enterprise-grade agent platform fully implements harness engineering and builds four core capabilities: production-grade long-running tasks, enterprise-grade security, in-depth industry know-how, and end-to-end observability, accelerating the large-scale deployment of industry-specific agents. The open-source edition of AgentArts, openJiuwen, has also been launched, sharing over 90% of its kernel with the AgentArts Enterprise Edition. Huawei Cloud also released a brand-new portal: AgentArts Orchard, which integrates full-stack agentic cloud services, diverse agents, and a vast range of models and applications. This portal provides users with a wide range of skill- and command-line interface (CLI)-based functions, enabling the entire process — from intent understanding, function development, resource provisioning, to application deployment — to be automated by agents. This allows AgentArts Orchard to deliver on-demand, efficient token services and bring users a brand-new interactive experience.
Unveiling a full-stage security solution for secure agentic AI
Security is the cornerstone of Huawei Cloud's digital and intelligent services. Huawei Cloud has built a security solution that covers the entire AI lifecycle, providing end-to-end protection for agents, models, and Agentic Infra. At the conference, Huawei Cloud released the data security zone, which ensures end-to-end data security from cloud migration to cloud operation through three innovative technologies: dedicated hardware encryption with Hold Your Own Key (HYOK) technology, data capsules, and multi-dimensional isolation for agentic infrastructure, allowing enterprises to maintain full control over their data sovereignty.
In addition, Huawei Cloud released the AI confidential computing solution, which provides five core capabilities: confidential virtual machines (VMs), remote attestation on the cloud, confidential computing key management, confidential inference gateway, and PCIPC-based NPU passthrough. This solution supports three core scenarios: confidential inference, confidential pre-training, and confidential federated learning, making high-value data and models truly trustworthy.
To date, Huawei Cloud has been running stably for 1,037 days without any major incidents, making it the most reliable cloud service provider for customers.
Releasing the agent-oriented hybrid cloud white paper, accelerating digital intelligence for critical industries
Huawei Cloud's hybrid cloud has become the core foundation for digital and intelligent transformation of governments and enterprises. It has maintained the No. 1 market share in key industries such as government services, finance, and state-owned enterprises for multiple years, serving more than 5,500 customers worldwide. At the event, the White Paper: Building Agent-Oriented Hybrid Cloud for Enterprises was officially released. It provides reference guidance for the evolution of hybrid cloud architecture and enterprise practices in the agentic era. The white paper covers areas such as building an AI data lake to break down data silos, realizing seamless coordination between stable online models and agile offline iterations, and establishing a secure and reliable environment for agent development and runtime. The white paper helps government and enterprise customers securely and efficiently achieve private deployment of enterprise agents, unlocking the value of their data.
Launching four zones on the Industry AI Foundry, accelerating AI adoption across industries
The Smart Healthcare Zone is continuously upgraded. Huawei Cloud's healthcare AI enablement platform will be available for open beta test (OBT) on June 30. Huawei Cloud's smart pathology solution has been widely replicated across China, covering top-tier hospitals, prefecture-level hospitals, and county-level hospitals, accelerating the adoption of AI in healthcare. At the event, more than 20 hospitals — including Ruijin Hospital (of Shanghai Jiao Tong University School of Medicine), Handan Central Hospital, Affiliated Hospital of Hebei University of Engineering, Ruian People's Hospital, Xingyi People's Hospital of Qianxinan Prefecture, and Wu'an First People's Hospital — officially joined the Smart Healthcare Zone, marking the nationwide large-scale deployment of the smart pathology solution and making AI accessible to more doctors and patients.
The Embodied AI Zone provides a one-stop platform for data synthesis, model development, and simulation verification, helping enterprises accelerate the implementation of embodied AI in real-world scenarios. Huawei Cloud also released CloudRobo, the world's first end-to-end intelligent development platform for robots. It provides a secure and trustworthy PB-scale data foundation and development pipelines, the industry's first cloud-native robot model production engine, and the first fully homegrown Real-Sim data production and model evaluation system. This enables robot migration to the cloud in just hours and model deployment in minutes. The platform will be available for OBT on June 30.
At the event, Huawei Cloud also launched the Smart Manufacturing Zone, which provides a one-stop innovation and entrepreneurship environment for industrial AI agents, enabling the implementation of innovative industrial AI agents. The Scientific Computing Zone was also launched, offering AI4S industry customers a unified and rich experience with models and agents, and facilitating agile scientific research and innovation.
Collaborating with top model providers to launch the AI Model Partner Program
At the event, Huawei Cloud, together with more than 20 top model providers, including Zhipu AI, DeepSeek, MiniMax, Kimi, StepFun, Baidu, iFLYTEK Spark, Meituan, AIsphere, and Shengshu Technology, released the AI Model Partner Program. This plan aims to build a systematic business ecosystem and create a new model of industry development that benefits all parties with diverse models and shared success.
Moving forward, Huawei Cloud will drive software-hardware-chip synergistic innovation to build the foundation for enterprise-grade AI innovation, working alongside global customers, partners, and developers to usher in a brand-new era of Agentic AI.
