Samsung began mass shipments of 12-layer HBM4 to Nvidia in February 2026, primarily for the Vera Rubin AI accelerator, with speeds of up to 11.7 Gbps. Samsung and SK hynix are preparing volume production of 8-layer stacks for Nvidia’s Rubin Ultra NVL systems in the second quarter of 2027. Shorter stacks are favored for their improved thermal management and higher manufacturing yield. SK hynix is projected to supply about two-thirds of Nvidia’s HBM4 volumes, while Samsung targets a 28% market share or more, with the two companies expected to control more than 80% of the market by mid-2027.
Source: Read the original article


