Samsung May Begin HBM4 Mass Production in H2 2025

Samsung failed to penetrate the HBM market as it missed out on the early AI boom. Its arch-rival SK Hynix secured deals with all the key clients, capturing a larger share of the market. The Korean firm now aims to bounce back in this space with its next-generation HBM4 solution. It plans to supply HBM4 samples in August and begin mass production in the second half of the year.
Samsung ramps up HBM4 strategy to attract clients
Samsung is actively looking for big orders from major AI chip makers in the industry. It recently signed a deal with AMD to supply its 12-layer HBM3E chip for next-generation accelerators, the MI350 series. The company is now hoping to gain approval from NVIDIA to join its supply chain as well. At the same time, it aims to introduce its HBM4 chips to the market as early as possible to attract clients.
During the recent biannual meetings, Samsung looked for several ways to regain its position in the growing HBM market. It discussed business-specific challenges, sales strategies, and performance goals. One of the key topics was HBM chip sales and production plans. More importantly, Samsung reviewed plans to start mass production of its HBM4 chips in the second half of the year, hoping this will help it make a comeback in this space.
Of course, the company will first need to secure major deals with AI chip makers. As demand for high-performance computing (HPC) and artificial intelligence (AI) applications increases, the company remains hopeful that it will regain ground in the HBM4 era. Samsung is now eyeing to supply its HBM4 samples to clients in August. It has already started negotiating with Nvidia, Broadcom, and Google for the supply of a customized HBM4 solution.
Speaking of its development progress, last month, it achieved nearly a 40% yield in 1c DRAM, the core foundation of HBM4 chips. It aims to begin full-scale mass production in the second half of the year. This will pave the way for the Korean firm to start mass production of 6th-generation high-bandwidth memory solutions as scheduled.










