650 customers onboard! Alibaba's Wu Yongming: T-Head AI chips will continue to ramp up in the second half of the year.

On August 20, Alibaba Group released its financial results for the first quarter of fiscal year 2027.

During the analyst conference call, Alibaba Group CEO Eddie Wu stated that T-Head has established a full-stack self-developed chip portfolio covering GPUs, CPUs, and network chips. As of early August, the Zhenwu chip has served more than 650 customers.

In addition, the super-node instance based on the next-generation T-Head chip, Zhenwu M890, recently launched on Alibaba Cloud for large-scale sales, with continued volume expansion in the second half of the year to meet strong customer demand.

It is reported that the Zhenwu M890 made its debut at the Alibaba Cloud Summit. The chip adopts T-Head's self-developed parallel computing architecture, features 144GB of HBM high-bandwidth memory, and achieves 800GB/s inter-chip interconnect bandwidth, delivering overall performance three times that of the previous-generation Zhenwu 810E.

The chip natively supports multiple data precisions from FP32 to FP4, covering full scenarios including high-precision training, low-precision inference, and ultra-low-precision inference.

Paired with the self-developed ICN Switch 1.0 interconnect chip, it enables full-bandwidth interconnect across 64 cards, significantly improving the computing efficiency and stability of large-scale intelligent computing clusters.

The latest financial report shows that the Zhenwu chip, through Alibaba Cloud, has achieved broad commercial application across more than 650 external customers in over 20 industries, including autonomous driving, internet, and finance.

Furthermore, Alibaba Cloud has compressed the delivery cycle for large-scale AI data centers to 100 days, and this year the production capacity efficiency of its self-developed modular data centers will more than double to accommodate stronger AI computing demand.