China's CXMT starts small-scale HBM3E production to supply domestic AI chipmakers
China's CXMT has initiated small-scale production of HBM3E, an advanced high-bandwidth memory, to support domestic AI chipmakers. This move comes as Chinese technology firms face tighter restrictions on accessing foreign semiconductors. The company is supplying these chips in limited volumes for testing to developers like Alibaba Group's T-Head and Cambricon Technologies. If trials succeed, these domestic processor developers aim to integrate the memory into commercial products next year, ahead of CXMT's planned capacity expansion in 2027. This development reflects China's push for self-sufficiency in critical semiconductor components amid ongoing trade measures.
CXMT's small-scale HBM3E production marks a significant step for China's domestic chip supply chain, particularly for AI chip developers like Alibaba's T-Head and Cambricon. The ability to produce advanced memory chips, even in limited volumes, addresses a critical bottleneck created by US export restrictions on high-end memory and fabrication tools. This initiative reflects China's strategic response to secure its AI ambitions and reduce reliance on foreign technology. However, the immediate impact on the broader market will be limited. CXMT reportedly faces low manufacturing yields and lags behind leaders like SK hynix and Samsung Electronics by an estimated three to five years. South Korean manufacturers began commercial HBM3E production in 2024 and have already advanced to HBM4. While CXMT's recent 57.9 billion yuan (USD 8.6 billion) funding will support its HBM push, its capability remains unproven at scale, as noted by Counterpoint's Neil Shah.
分享这篇文章
相关文章
6 则
What’s next after China court freezes US$318m in Nexperia assets amid Wingtech battle?
Our past coverage explored the broader implications of China's tech sector navigating international restrictions.

中美需联手对抗另一强大敌人:人工智能
微软人工智能在模拟诊断测试中击败医生 - Fox News

Oracle 拓展其 AI 云服务,整合 Google TPU、定制芯片及 Gemini 数据链接

我让我的编码代理解释它自己的拉取请求

