At the AI Infrastructure Summit in Santa Clara, California, the future trajectory of AI memory took center stage. Samsung revealed its latest product roadmap, showcasing its next-generation AI memory, zHBM, which is poised to dramatically enhance the response speed of AI accelerators by up to ten times. Samsung aims to elevate the number of tokens processed per second by conversational AI systems from 100 to 1,000. The performance of zHBM is projected to reach up to eight times that of HBM5, while improving energy efficiency by up to threefold.

Additionally, Samsung announced its intention to start providing samples of its three-dimensional storage solution, zNAND-O, beginning in 2028. During the summit, CXL technology encountered skepticism from certain attendees. Although some industry experts believe that CXL technology could mitigate the challenges posed by high HBM prices and constrained supply, representatives from OpenAI and Intel highlighted that CXL technology is primarily suited for storing cold data and cannot serve as a substitute for HBM. Samsung and SK Hynix, currently at the forefront of the HBM market, are anticipated to sustain their robust growth momentum.