BEIJING — China's CXMT plans to start production of HBM3E, an advanced type of high-bandwidth memory chip used with AI processors, despite as little as 20% of the chips it makes being usable, two people familiar with the matter told Reuters.
Source: Reuters
ChangXin Memory Technologies — China's largest maker of dynamic random-access memory (DRAM) — has sent HBM3E samples to several domestic AI chip designers, the people said. It expects to begin limited production over the next two months, one person said. Seoul Economic Daily, reporting the same Reuters disclosure on Wednesday, said this is the first time a specific yield figure for CXMT's HBM3E has been made public.
Source: Seoul Economic Daily (Reuters-Yonhap)
The decision highlights China's intense shortage of the AI memory chips, which are essential to a government-backed push to develop domestic alternatives to Nvidia's AI processors at scale amid US curbs on access to foreign-made technology, Reuters said. High-bandwidth memory is made by vertically stacking memory chips so they can exchange data rapidly with AI processors, and is crucial to the specialised chips used to train and run large language models.
The people declined to be identified because the plans are not public, Reuters said. CXMT did not respond to a request for comment, and SK Hynix did not immediately respond to a request for comment on the yield estimate. The Information first reported CXMT's HBM3E plans; Reuters said it is reporting details about the product yields and delivery timeline for the first time.
Source: Reuters
What is solid
- CXMT has shipped HBM3E samples to several Chinese AI chip designers and plans limited production within two months — reported by Reuters and carried by Seoul Economic Daily on October 28.
- The yield comparison — 20-30% at CXMT against about 70% at SK Hynix — is attributed in both reports to sources familiar with the matter, not to either company's published figures.
- US restrictions on certain advanced HBM exports to China were tightened in 2024, a fact stated in both reports.
What is only claimed
- The production plan, yields and timeline rest on two anonymous sources cited by Reuters; CXMT has not confirmed them publicly and did not respond to a request for comment, so the figures should be read as source-based reporting, not company-confirmed data.
What happens next: if limited production begins on the reported timeline, CXMT's output will still be constrained by low yields, which raise the cost of each finished stack — meaning China's AI chipmakers are unlikely to see relief from the memory shortage quickly, even as domestic production starts.
