SK hynix to Delay HBM4 Mass Production and Expansion

According to ZDNet Korea, memory giant SK hynix has revised its HBM4 production roadmap, pushing mass production from the original February 2026 target to around March–April 2026, while large-scale output is now expected in the third quarter rather than the second. As a result, the ramp-up of materials and component preparation for HBM4 has also slowed.

According to people familiar with the matter, SK hynix had planned to steadily increase HBM4 output in the first half of 2026 and raise the overall share of HBM products by the end of Q2. That plan has changed. For at least the first half of 2026, HBM3E will remain the company's primary production focus. The shift is widely seen as closely tied to changes in NVIDIA's strategy and evolving market demand.

According to industry sources, discussions between SK hynix and NVIDIA about 2026 HBM supply revealed a sharp increase in NVIDIA's orders for HBM3E. This points to two key developments: first, NVIDIA's next-generation "Rubin" AI chip—designed to pair with HBM4—may be launching later than expected; second, demand for Blackwell GPUs using HBM3E remains strong, encouraging SK hynix to prioritize supply of the current generation.

According to the same sources, potential delays to Rubin are not just about timing, but also technology. HBM4 significantly raises the bar, doubling I/O counts to 2,048 for higher bandwidth and adopting a non-DRAM process for the control die, which is expected to be produced on TSMC's foundry lines. At the same time, TSMC's 2.5D advanced packaging platform, CoWoS, is facing capacity constraints, adding another layer of complexity to the schedule.

According to SK hynix, while it did not confirm specific operational details, the company said it will continue to adjust plans flexibly based on market demand.