CXMT Reportedly Starts HBM3E Risk Production, With 2027 Supply Still Unproven
T-Head and Cambricon are reportedly evaluating samples, but no public CXMT specification, yield, capacity, or committed shipment date is available.
Original title: CXMT reportedly begins risk production of HBM3E memory in breakthrough for Chinese DRAM production — company could be in mass production in 2027
Tom’s Hardware, citing a report by The Information, says ChangXin Memory Technologies has begun risk production of HBM3E and that Chinese processor developers including Alibaba’s T-Head and Cambricon are evaluating samples. If qualification proceeds as described, the report gives 2027 as the earliest window for commercial use. These details remain attributed reporting. The public material listed by Tom’s Hardware does not include a CXMT confirmation of product specifications, customers, or a mass-production date.
Risk production is not the same as volume shipment. The report explicitly says the specifications of CXMT’s samples are unknown and that a commercial device could differ from the risk-production version. Customer evaluation must still test bandwidth, power, thermals, package reliability, and operation with the target accelerator. Only after qualification and repeatable supply would a trial run become measurable HBM availability.
Tom’s Hardware uses the general JEDEC HBM3E envelope to explain the generation: a 1,024-bit interface, transfer rates of up to 9.6 GT/s per pin, and eight-high or 12-high stacks. At the corresponding rate, one stack can provide as much as about 1.228 TB/s of bandwidth. With 24Gb DRAM dies, eight-high and 12-high configurations can provide 24GB and 36GB respectively. Those are standard-level capabilities, not evidence that CXMT’s reported sample reaches every value.
The manufacturing challenge also extends beyond making DRAM dies. The report identifies through-silicon vias, stacking interconnects, thermal management, advanced packaging, and testing as required parts of an HBM product. Even if the risk-production report is accurate, the public record cited here gives no yield, monthly capacity, packaging partner, number of samples, or customer qualification result. It therefore does not establish how far CXMT is from stable volume output, and 2027 should not be read as a committed delivery date.
The next decisive evidence would be a public CXMT or customer disclosure that identifies the sample specification and qualification status, followed by packaging details, yield, capacity, and commercial shipments. If only additional anonymous schedules appear, the story remains a report of risk production. A qualified product and documented deliveries would be the evidence needed to confirm that domestically supplied HBM3E has entered the Chinese accelerator market.