返回首页
AI agents are reshaping the landscape of cloud computing, while Alibaba Cloud has completed the restructuring of its full-stack technology system. Beyond product upgrades, this marks a key strategic layout, as Alibaba Cloud is betting on its development path for the next decade in the agentic AI era.
On May 20, the 2026 Alibaba Cloud Summit kicked off in Hangzhou, unveiling a host of official updates showcasing the company’s latest product progress.
During the main summit forum, Liu Weiguang, Senior Vice President of Alibaba Cloud, Li Feifei, Chief Technology Officer of Alibaba Cloud, Zhou Jingren, Head of Alibaba Tongyi Large Model Division, Gao Hui, Vice President of T-Head Semiconductor, and Yu Wenyuan, Technical Lead of Alibaba Cloud Bailian Platform delivered keynote speeches respectively, covering core sectors ranging from self-developed chips and large models to ecological alliance cooperation in a progressive manner.
Starting with the chip sector, T-Head, an Alibaba affiliate, officially launched its high-end self-developed AI chip Zhenwu 810E on its official website last January. Four months later, T-Head unveiled its new-generation training-and-integration AI chip Zhenwu M890. Equipped with 144GB on-board video memory and achieving an inter-chip interconnect bandwidth of 800GB/s, it delivers three times the performance of Zhenwu 810E. Natively supporting diverse data precisions from FP32 down to FP4, it is applicable to full-scenario business including high-precision model training as well as low-precision and ultra-low-precision inference tasks.
In its recent 2026 financial report, Alibaba disclosed that T-Head’s self-developed GPU chips have achieved large-scale mass production and extensive industrial deployment, with over 60% of its computing power serving external commercial clients across core sectors such as the internet, finance and autonomous driving. According to the latest data released by Gao Hui, Vice President of T-Head Semiconductor, the cumulative shipment volume of the Zhenwu series chips has reached 560,000 units, a sharp rise from the previously announced 450,000 units. These chips now serve more than 400 clients covering over 20 industries including China Telecom, FAW Group and SPD Bank.
This breakthrough carries far-reaching strategic significance for Alibaba. During the 2026 earnings call, Wu Yongming, CEO of Alibaba Group, stated that the deployment ratio of T-Head self-developed chips within Alibaba Cloud remains relatively low for now, mainly constrained by domestic semiconductor production capacity, which is nevertheless in continuous expansion in recent years. The rising penetration rate of T-Head full-stack self-developed chips will bring a notable positive boost to corporate gross profit margins.
Meanwhile, Alibaba rolled out a 128-card super-node server built around the new-generation Zhenwu M890 AI chip, equipped with the ICN Switch 1.0 interconnection chip that cuts communication latency down to the hundred-nanosecond level. It enables 128 AI chips to operate as an integrated supercomputing cluster, fully meeting the demands for high-concurrency inference and large-scale model training in the agentic era.
At the large model level, Alibaba officially launched its new flagship Qwen model Qwen3.7-Max. In the overall global blind evaluation ranking conducted by third-party institution Arena, Qwen3.7-Max outperforms Kimi-K2.6, DeepSeek-v4-pro and GLM-5.1, ranking first among domestic Chinese large models and closing the performance gap with top-tier models including GPT, Claude and Gemini. As pointed out by Zhou Jingren, the most outstanding strength of Qwen3.7-Max lies in its capability to operate independently from scratch and accomplish continuous long-term tasks lasting up to 35 hours. "Qwen3.7-Max is originally designed to act as the intelligent core of AI agents, endowed with capabilities including autonomous planning, continuous iterative optimization and cross-tool collaborative execution."
If self-developed chips lay the solid foundation and large models serve as the core power engine, Alibaba’s large model development and service platform — Bailian Platform — acts as the core workshop enabling massive AI agents to land practical applications and deliver commercial services. At the 2026 Alibaba Cloud Summit, Alibaba announced the full opening of the Bailian Platform and established strategic partnerships with leading AI enterprises including Moonshot AI, Minimax, Zhipu AI, Stepfun, Aishi Technology and Shengshu Technology.
Alibaba has previously announced that it will invest over 380 billion yuan in AI infrastructure construction in the next five years. Industry analysts believe the bulk of the funds will be channeled into AI chip research and manufacturing to enhance independent R&D capabilities, reduce external reliance and further drive iterative upgrades of large models. This summit clearly signals Alibaba’s complete development roadmap: independent chip research and development, leading large model competitiveness and open ecological cooperation. Steady progress in all three sectors is expected to turn Alibaba Cloud’s profit growth expectations into tangible business results.
Towards the end of the summit, Alibaba Cloud launched its brand-new official AI product website named Qwen Cloud, positioned as a full-stack intelligent infrastructure built exclusively for AI agents. This marks the first standalone dedicated product website launched by Alibaba Cloud in its 17-year development history, separate from its official cloud service portal. Distinct from conventional web layouts, its homepage features no navigation menus but only a single prompt instruction line.
Qwen Cloud integrates API access to more than 150 mainstream large models including Qwen, GLM, Kimi, DeepSeek, Wan and HappyHorse. It encapsulates core model service capabilities into standardized Skills and CLI tools. AI agents can directly parse the homepage prompt, master all functional capabilities of the platform and call relevant services autonomously according to actual task demands.
By revamping its product entry logic, Alibaba Cloud is also placing a bold bet on future user evolution trends: when AI agents rather than human users become the primary consumers of cloud services, all user-oriented interfaces, operational workflows and human-computer interaction rules will be thoroughly reinvented.
Source: Cover Finance
Live Coverage of 2026 Alibaba Cloud Summit: T-Head Zhenwu M890 Chip Makes Debut & AI Product Official Website Qwen Cloud Launches
AI agents are reshaping the landscape of cloud computing, while Alibaba Cloud has completed the restructuring of its full-stack technology system. Beyond product upgrades, this marks a key strategic layout, as Alibaba Cloud is betting on its development path for the next decade in the agentic AI era.
On May 20, the 2026 Alibaba Cloud Summit kicked off in Hangzhou, unveiling a host of official updates showcasing the company’s latest product progress.
During the main summit forum, Liu Weiguang, Senior Vice President of Alibaba Cloud, Li Feifei, Chief Technology Officer of Alibaba Cloud, Zhou Jingren, Head of Alibaba Tongyi Large Model Division, Gao Hui, Vice President of T-Head Semiconductor, and Yu Wenyuan, Technical Lead of Alibaba Cloud Bailian Platform delivered keynote speeches respectively, covering core sectors ranging from self-developed chips and large models to ecological alliance cooperation in a progressive manner.
Starting with the chip sector, T-Head, an Alibaba affiliate, officially launched its high-end self-developed AI chip Zhenwu 810E on its official website last January. Four months later, T-Head unveiled its new-generation training-and-integration AI chip Zhenwu M890. Equipped with 144GB on-board video memory and achieving an inter-chip interconnect bandwidth of 800GB/s, it delivers three times the performance of Zhenwu 810E. Natively supporting diverse data precisions from FP32 down to FP4, it is applicable to full-scenario business including high-precision model training as well as low-precision and ultra-low-precision inference tasks.
In its recent 2026 financial report, Alibaba disclosed that T-Head’s self-developed GPU chips have achieved large-scale mass production and extensive industrial deployment, with over 60% of its computing power serving external commercial clients across core sectors such as the internet, finance and autonomous driving. According to the latest data released by Gao Hui, Vice President of T-Head Semiconductor, the cumulative shipment volume of the Zhenwu series chips has reached 560,000 units, a sharp rise from the previously announced 450,000 units. These chips now serve more than 400 clients covering over 20 industries including China Telecom, FAW Group and SPD Bank.
This breakthrough carries far-reaching strategic significance for Alibaba. During the 2026 earnings call, Wu Yongming, CEO of Alibaba Group, stated that the deployment ratio of T-Head self-developed chips within Alibaba Cloud remains relatively low for now, mainly constrained by domestic semiconductor production capacity, which is nevertheless in continuous expansion in recent years. The rising penetration rate of T-Head full-stack self-developed chips will bring a notable positive boost to corporate gross profit margins.
Meanwhile, Alibaba rolled out a 128-card super-node server built around the new-generation Zhenwu M890 AI chip, equipped with the ICN Switch 1.0 interconnection chip that cuts communication latency down to the hundred-nanosecond level. It enables 128 AI chips to operate as an integrated supercomputing cluster, fully meeting the demands for high-concurrency inference and large-scale model training in the agentic era.
At the large model level, Alibaba officially launched its new flagship Qwen model Qwen3.7-Max. In the overall global blind evaluation ranking conducted by third-party institution Arena, Qwen3.7-Max outperforms Kimi-K2.6, DeepSeek-v4-pro and GLM-5.1, ranking first among domestic Chinese large models and closing the performance gap with top-tier models including GPT, Claude and Gemini. As pointed out by Zhou Jingren, the most outstanding strength of Qwen3.7-Max lies in its capability to operate independently from scratch and accomplish continuous long-term tasks lasting up to 35 hours. "Qwen3.7-Max is originally designed to act as the intelligent core of AI agents, endowed with capabilities including autonomous planning, continuous iterative optimization and cross-tool collaborative execution."
If self-developed chips lay the solid foundation and large models serve as the core power engine, Alibaba’s large model development and service platform — Bailian Platform — acts as the core workshop enabling massive AI agents to land practical applications and deliver commercial services. At the 2026 Alibaba Cloud Summit, Alibaba announced the full opening of the Bailian Platform and established strategic partnerships with leading AI enterprises including Moonshot AI, Minimax, Zhipu AI, Stepfun, Aishi Technology and Shengshu Technology.
Alibaba has previously announced that it will invest over 380 billion yuan in AI infrastructure construction in the next five years. Industry analysts believe the bulk of the funds will be channeled into AI chip research and manufacturing to enhance independent R&D capabilities, reduce external reliance and further drive iterative upgrades of large models. This summit clearly signals Alibaba’s complete development roadmap: independent chip research and development, leading large model competitiveness and open ecological cooperation. Steady progress in all three sectors is expected to turn Alibaba Cloud’s profit growth expectations into tangible business results.
Towards the end of the summit, Alibaba Cloud launched its brand-new official AI product website named Qwen Cloud, positioned as a full-stack intelligent infrastructure built exclusively for AI agents. This marks the first standalone dedicated product website launched by Alibaba Cloud in its 17-year development history, separate from its official cloud service portal. Distinct from conventional web layouts, its homepage features no navigation menus but only a single prompt instruction line.
Qwen Cloud integrates API access to more than 150 mainstream large models including Qwen, GLM, Kimi, DeepSeek, Wan and HappyHorse. It encapsulates core model service capabilities into standardized Skills and CLI tools. AI agents can directly parse the homepage prompt, master all functional capabilities of the platform and call relevant services autonomously according to actual task demands.
By revamping its product entry logic, Alibaba Cloud is also placing a bold bet on future user evolution trends: when AI agents rather than human users become the primary consumers of cloud services, all user-oriented interfaces, operational workflows and human-computer interaction rules will be thoroughly reinvented.
Source: Cover Finance