Key Highlights
- The Chinese e-commerce giant introduced its Zhenwu M890 AI accelerator, delivering triple the computational power of the earlier Zhenwu 810E model for agentic AI applications.
- T-Head, Alibaba’s chip design division, engineered the processor, which is now accessible to enterprise clients in China through the Alibaba Cloud Bailian platform.
- The company revealed an ambitious chip development timeline featuring the V900 (arriving Q3 2027) and J900 (scheduled Q3 2028), each targeting approximately 3x performance improvements.
- Simultaneously, Alibaba introduced Qwen 3.7-Max, an enhanced large language model optimized for complex coding workflows and extended agent operations lasting up to 35 hours.
- The company has distributed more than 560,000 Zhenwu processors to date, serving over 400 clients spanning 20 different sectors.
On Wednesday, Alibaba introduced a domestically developed AI processor alongside enhancements to its leading large language model, reinforcing its commitment to establishing a self-sufficient AI infrastructure that reduces reliance on Nvidia.
The Zhenwu M890 accelerator represents a significant leap forward, offering three times the computational capacity of the Zhenwu 810E. The announcement took place during the Alibaba Cloud Summit in Hangzhou, with immediate availability for Chinese business clients via the Bailian platform.
Alibaba stock (BABA) experienced a modest 0.60% decline when the announcements were made.
Alibaba Group Holding Limited, BABA
T-Head, Alibaba’s chip engineering subsidiary, designed the M890 with 144GB of GPU memory—a substantial upgrade from the 96GB capacity in the prior generation, enabling processing of significantly larger datasets.
The M890 represents a fundamental shift from its predecessor by supporting both model training and inference operations, rather than inference exclusively. This dual capability positions it as a comprehensive solution for organizations developing and deploying AI systems.
The processor is specifically engineered for agentic AI applications—intelligent systems capable of executing sophisticated, multi-stage operations with minimal human oversight. These demanding workloads require models to maintain extensive contextual information and synchronize activities in real time.
According to T-Head, cumulative shipments of Zhenwu processors have surpassed 560,000 units, serving more than 400 enterprise clients across 20 sectors including automotive manufacturing and financial services.
Multi-Year Development Strategy Unveiled
The Chinese tech giant went beyond a single product release, presenting an extensive chip development strategy that encompasses the V900, anticipated in Q3 2027, and the subsequent J900 model slated for Q3 2028.
The V900 is expected to achieve approximately another 3x performance enhancement compared to the M890. This consistent development rhythm demonstrates that Alibaba views semiconductor innovation as a foundational infrastructure investment rather than a temporary initiative.
Complementing the M890 announcement, Alibaba revealed the Panjiu AL128, a server architecture that integrates 128 of the new processors into a unified rack system.
Language Model Receives Significant Enhancement
The company simultaneously launched Qwen 3.7-Max, the newest iteration of its primary large language model. This model features specialized optimization for sophisticated programming tasks and extended agent operations, maintaining stable performance for continuous runs up to 35 hours.
Both the Qwen enhancement and the chip introduction were presented together at the Hangzhou event, framing them as complementary components of a unified strategy focused on agentic AI capabilities.
This development arrives amid ongoing U.S. export restrictions limiting Nvidia’s access to the Chinese market for cutting-edge processors. Alibaba maintained significant purchasing relationships with Nvidia before these controls accelerated the transition toward domestic chip alternatives.
Several Chinese semiconductor firms, including Biren Technology and MetaX, have recently completed public listings on mainland China and Hong Kong stock exchanges, signaling broader industry expansion.
Alibaba has previously committed over 380 billion yuan (approximately $53 billion) toward cloud computing and AI infrastructure investments across a three-year period—representing the company’s largest capital allocation to this sector.
During a recent quarterly earnings discussion, CEO Eddie Wu indicated that revenue from AI-related products is anticipated to emerge as the primary growth engine for the cloud division. The organization has positioned its internally developed processors as delivering “the highest value for money compute power,” with potential to enhance profit margins.
