GMAsia
    🇨🇳China·AI News·22 Sept 2026·via Tech Node

    T-Head unveils Zhenwu V900 AI chip in Alibaba’s push to expand its AI infrastructure stack

    Alibaba’s chip subsidiary T-Head unveiled its Zhenwu V900 AI chip at the 2026 Apsara Conference. The V900 is designed for both training and inference, delivering three times the performance of its predecessor. It features 216GB of memory and 1,200GB/s of inter-chip bandwidth. T-Head also detailed its roadmap for Yitian server CPUs, with new models like the Yitian 720 and 730 expected in Q3 2027.

    Nexa's Summary

    T-Head’s Zhenwu V900 is not just another accelerator. It is a full-stack play, integrating computing, storage, and networking. The V900 supports low-precision computing formats including FP8 and FP4. Its architecture allows more than 1,000 chips to function as a single system. This focus on system-level design addresses the scaling challenges of trillion-parameter models and agentic AI.

    Alibaba’s expansion challenges Nvidia’s dominance in China’s AI infrastructure. Alibaba Cloud’s new AI computing center network architecture can scale to 500,000 accelerators. This creates a powerful in-house alternative for Chinese enterprises. The Zhenwu series already serves over 650 enterprise customers, including those in autonomous driving and large language models.

    The test for T-Head is whether its Yitian 750 CPU can truly integrate with Zhenwu AI chips. This direct interconnectivity, planned for a future Yitian release, would be a critical differentiator. It would further reduce reliance on external chip designs. The Yitian 730, with 1.4 times the single-core performance of the Yitian 710, sets a high bar for future integration.

    #heavy hitters#news#semiconductor#alibaba#highlight#ai#chips
    Go deeper
    Original reporting by Tech NodeWe don't republish, read the full story →

    Related reading

    6 stories