At the Alibaba Cloud Summit on May 20, 2026, Alibaba Cloud officially announced the completion of a full-stack technology system upgrade oriented towards the "Agentic (Agent) Era." The core was restructured across the entire pipeline from the underlying chip, cloud platform, model, to the inference solution. This marks Alibaba Cloud's transition from a cloud that serves human users to an "AI factory" that supports the 24/7 continuous operation of massive agents.

1. Core Foundation: Tengxun Zhenwu M890 Chip and Super Node Server

The most fundamental power behind this upgrade comes from Tengxun's new generation of training and inference integrated AI chip - Zhenwu M890.

Performance Improvement: M890 is equipped with 144GB of large memory, three times the performance of the previous generation Zhenwu 810E, and natively supports multiple data precision levels from FP32 to FP4, perfectly matching the needs of high-precision training and ultra-low-precision concurrent inference in Agent scenarios.

Cluster Interconnection Breakthrough: Combined with the self-developed ICN Switch 1.0 interconnection chip, Alibaba Cloud has launched the Panjiu AL128 Super Node Server based on Zhenwu M890. By achieving system-level collaboration of "storage, computing, and networking" across 128 AI chips, it realizes ultra-low communication latency at the nanosecond level, significantly improving the efficiency and stability of large-scale intelligent computing clusters.

Future Plan: Tengxun publicly revealed the chip roadmap for the Zhenwu series for the first time, clearly stating that in the next two years, more powerful chips such as Zhenwu V900 and Zhenwu J900 will be released, establishing its long-term competitiveness in the data center computing market.

2. Core Access Point: Reconstructed "Qwen Cloud" and "Agentization Interaction"

Alibaba Cloud has made a disruptive transformation of the cloud interaction logic. Traditional cloud platforms were designed for "humans" (control panels, dashboards), while the cloud in the Agentic era must be designed for "agents."

AI-Native Website "Qwen Cloud": The homepage of the website is no longer a complicated product catalog, but a standardized Skills installation code. Agents can directly parse the code instructions, autonomously call the cloud platform's computing, storage, and model capabilities, without the need for manual configuration through the control panel.

Standardization of Capabilities: Alibaba Cloud has encapsulated over 150 mainstream models and cloud product capabilities into standardized Skills and CLI tools. Whether it's Claude Code or various mainstream Agent frameworks, only one line of instruction is needed to quickly install and call all of Alibaba Cloud's infrastructure capabilities.

3. Technical Strategy: Full-Stack Integration of "Chip-Cloud-Model-Inference"

This new technical system aims to address the unique challenges brought by Agent workloads: they exhibit characteristics such as "irregular elasticity, short life cycle, and extremely high instantaneous concurrency."

Deep Optimization: Alibaba Cloud not only provides models (such as the latest flagship model Qwen3.7-Max), but also achieves optimal scheduling of computing resources through deep integration between the underlying chip (Zhenwu series) and the upper-layer inference framework.

Shift in Objectives: As stated by Alibaba Cloud CTO Feifei Li and relevant experts, the focus of large models has shifted from merely "aligning with human preferences (saying well)" to "aligning with task objectives (getting things done)." The evolution of the entire system is aimed at ensuring that Agents can efficiently complete complex engineering tasks within milliseconds, thereby lowering the AI application threshold for thousands of industries.

Summary:

By combining the "Tengxun chip matrix + Qwen Cloud access point + full-stack model inference," Alibaba Cloud has become the first in the industry to complete the role shift from "computing power rental" to "AI factory." This system is not only an infrastructure solution to cope with the explosive growth of agents, but also demonstrates the ambition of Chinese tech giants to reshape the global productivity entry point through software and hardware collaboration in the Agentic era.