[ChangXin Breaks Through with HBM3E: Just One Generation Behind Global Leaders] Beating AI Newsflash, citing two insiders from *The Information*, reports that ChangXin Memory has begun small-scale production of HBM3E and plans to ramp up production by 2027. HBM is high-speed memory placed next to AI chips, responsible for rapidly delivering data to the chips. Products above HBM2 have consistently been a critical shortfall for domestic AI chips. Domestic chip teams such as Cambricon and Alibaba's T-Head are already testing ChangXin's HBM, and if progress goes smoothly, it could be integrated into products as early as next year. HBM3E is also the generation of memory used in NVIDIA's H200 and Blackwell. In terms of product generations, ChangXin is currently only one generation behind SK Hynix, Samsung, and Micron, who are mass-producing HBM4. However, small-scale production is still far from mature mass production. ChangXin's HBM3E currently has relatively low yields, and its speed and reliability are not yet on par with the three major manufacturers. SK Hynix began mass production of HBM3E as early as 2024, and the three major manufacturers have already entered the HBM4 stage and started sampling HBM4E. This time, ChangXin has crossed the hurdle of "can it be made," but the next challenge is to truly improve yield rates and production capacity. [Original Link]