China's CXMT to Launch HBM3E Production Despite Low Yields, Report Says
China's leading memory chipmaker is preparing to begin commercial production of advanced high-bandwidth memory chips used in artificial intelligence applications, even as its manufacturing yields remain a fraction of those achieved by global industry leaders, according to sources familiar with the matter.
According to Reuters' exclusive report published on October 28, 2026, Changxin Memory Technologies (CXMT) is moving forward with plans to produce HBM3E chips — a type of high-bandwidth memory (HBM) critical for powering AI processors — despite current yield rates of only around 20% to 30%.
Two sources cited in the report said CXMT has already distributed HBM3E samples to several domestic AI chip design companies and is expected to commence small-scale production within the next two months. The decision underscores the acute shortage of AI memory chips in China, where the government has been aggressively pushing for semiconductor self-sufficiency as U.S. export restrictions continue to limit Chinese firms' access to advanced foreign technology, including Nvidia's high-end AI chips.
The yield gap between CXMT and its global rivals is significant. South Korea's SK Hynix, the dominant player in the HBM market, operates at yield rates of approximately 70% — more than double CXMT's current performance. Yield rate is a key metric in chip manufacturing, directly affecting production costs and commercial viability.
Neither CXMT nor SK Hynix responded to requests for comment on the yield figures, according to Reuters. The sources declined to be identified as the plans have not been publicly disclosed.
The move reflects a broader strategic imperative in Beijing. With Washington tightening controls over exports of advanced semiconductors and chipmaking equipment, China has been channeling substantial resources into developing domestic alternatives across the chip supply chain. HBM chips are essential components for AI accelerators, and a reliable domestic supply would be pivotal to China's ambitions of fielding competitive homegrown AI processors at scale.
While CXMT's decision to proceed with low-yield production suggests urgency over efficiency, analysts may view the step as a necessary phase in building manufacturing experience and ramping up capacity over time. If yield rates improve materially, CXMT could emerge as a significant supplier to China's fast-growing AI hardware ecosystem — reducing, at least partially, the country's dependence on foreign memory suppliers.