China’s CXMT begins small-batch HBM3E production for AI chips

Signals Inbox·September 2, 2026·AI Chips

CXMT has reportedly started small-batch HBM3E production, putting China's leading domestic memory maker only one generation behind the newest high-bandwidth memory entering mass production at global leaders. HBM is one of the hardest bottlenecks in AI accelerators, so this is strategically bigger than another DRAM launch.

Q1What is officially known about CXMT?

CXMT's official product portfolio already shows advanced DDR5 and LPDDR5X memory, including LPDDR5X at 10,667 Mbps. The new HBM3E production detail was reported by The Information from people familiar with the matter.

Q2Why is HBM so important?

AI accelerators need enormous memory bandwidth to keep expensive compute cores fed with data. HBM stacks memory close to the processor and has become a critical constraint for Nvidia-class systems, not a commodity component that can be swapped easily.

Q3How far behind is China now?

The report says CXMT is producing HBM3E in small quantities. Samsung, SK Hynix and Micron are moving into the next generation, so China is still behind on scale and leading-edge performance, but the gap is no longer several generations.

Q4Who could use it?

Chinese AI-chip designers such as Cambricon and Alibaba's T-Head are obvious candidates because export controls make access to the best foreign AI hardware and memory less predictable. Domestic HBM reduces one more dependency in that stack.

Q5What is the real test now?

Yield and volume. Making a small batch proves technical progress, but AI infrastructure needs huge quantities of consistently packaged HBM. If CXMT can scale production through 2027, the strategic effect becomes much larger than the first shipment.