Samsung could start mass production of HBM4 chips next month for Nvidia

Samsung’s HBM3 and HBM3E chips, which are used in AI accelerators, faced performance issues. The company had to redesign its HBM3E chips last year to secure Nvidia’s approval. Even then, those HBM3E chips were used only in select Nvidia AI accelerators sold in China. Samsung is now in a stronger position, as its HBM4 chips are reportedly close to receiving Nvidia’s approval.

According to a report from Bloomberg, Samsung’s HBM4 chips are nearing Nvidia’s approval. Their mass production is expected to begin in February 2026. The chips were sent to Nvidia in September 2025 and have reportedly entered the final stage of qualification. Once mass production starts, Samsung will be ready to supply HBM4 chips to Nvidia as well as other AI accelerator companies such as AMD and Google.

HBM stands for high-bandwidth memory. This high-performance memory technology uses multiple DRAM chips stacked vertically, delivering significantly higher bandwidth and speeds than conventional DDR or LPDDR memory used in phones, computers, servers, and other electronic devices. HBM chips are extremely complex and expensive to manufacture.

Despite being the world’s largest memory chip maker for decades, Samsung did not focus heavily on HBM early on. This allowed rivals such as SK Hynix and Micron to take the lead in the HBM market. By the time Samsung increased its focus during the fourth-generation HBM (HBM3) and fifth-generation HBM (HBM3E) eras, it had already fallen behind.

Samsung then intensified its efforts with the development of sixth-generation HBM (HBM4) chips. Some reports claim that the company's HBM4 chips outperform competing memory chips from SK Hynix and Micron. These HBM4 chips are expected to be used in Nvidia’s next-generation flagship AI processor, Rubin, which will launch this year.