SK Hynix To Supply Next-Gen Memory For Big Tech's AI Push, Prioritizes Nvidia Amid Supply Crunch

SK Hynix has won custom high-bandwidth memory (HBM) orders from Nvidia Corp NVDA, Microsoft Corp MSFT, and Broadcom Inc AVGO. This marks a growing demand as big tech shifts away from general-purpose memory to optimize the performance of its AI services.

HBM chips are a type of computer memory designed for high-performance computing, particularly in graphics processing and AI. They achieve high bandwidth and efficiency by stacking multiple DRAM (Dynamic Random Access Memory) dies vertically, connected via through-silicon vias (TSVs), and using a wide interface. 

Also Read: Micron And Trump Announce $200-Billion US Chip Expansion Plan, Aiming To Create 90K Jobs

Contract chipmaker Taiwan Semiconductor Manufacturing Co TSM will make logic dies for SK Hynix’s custom HBM.

The South Korean chip supplier prioritized delivery of its leading client, Nvidia, because it was unable to acclimate to the demand frenzy from Big Tech giants, the Korea Economic Daily reported on Thursday.

SK Hynix’s first custom HBM will be its seventh generation or HBM4E, likely for launch in the second half of 2026.

An August 2024 report indicated that Chinese technology companies, including Huawei Technologies Co and Baidu Inc BIDU, were hoarding HBM semiconductors from Samsung Electronics in anticipation of potential U.S. export restrictions on the chips.

Read Next:

• Samsung’s Outlook Dims Amid HBM Supply Struggles for Nvidia and Sluggish Consumer Chip Sales

Image: Shutterstock