Alibaba Group Holding (BABA) shares experienced a slight uptick following the unveiling of its new AI infrastructure, centered around the Zhenwu M890 AI accelerator. Announced at the Alibaba Cloud Summit in Hangzhou, the chip is a major step in China's domestic semiconductor push amid global technology restrictions. The M890 delivers three times the performance of its predecessor, the 810E, and supports both AI training and inference — a significant upgrade from the previous inference-only design.
Developed by Alibaba's semiconductor unit T-Head, the M890 packs 144GB of GPU memory, up from 96GB, enabling it to handle complex agentic AI workloads. Accompanying the chip is the Panjiu AL128 Supernode Server, which connects up to 128 accelerators in a single rack, underscoring Alibaba's system-level approach to high-performance AI computing. T-Head also revealed that approximately 560,000 Zhenwu chips have already been shipped to over 400 customers across 20 industries, including telecommunications giant China Unicom.
Looking ahead, Alibaba outlined a multi-year chip roadmap: the V900 is slated for Q3 2027, followed by the J900 in Q3 2028, each promising another roughly 3× performance leap. Alongside hardware, the company launched Qwen 3.7-Max, a flagship large language model optimized for long-running agent tasks (up to 35 hours) and advanced coding, tightly integrated with its cloud and chip ecosystem. Reports also suggest Alibaba is considering restructuring T-Head with potential employee ownership and a future IPO, a move that could unlock further capital.
The announcements reinforce Alibaba's commitment to building a domestic AI stack independent of Nvidia, especially as U.S. export controls limit access to advanced chips. CEO Eddie Wu previously stated that AI-related product revenue is expected to become the primary growth driver for the cloud segment. While the stock reaction was modest, the sustained chip roadmap and rapid enterprise adoption signal a deepening structural shift toward AI-driven revenue streams.
