SK Hynix Inc. plans to use more advanced technology than initially planned for its next-generation memory chips, responding to pressure from key artificial intelligence customers including Nvidia Corp.
The South Korean chipmaker will manufacture its HBM4 memory using 3-nanometer technology instead of the originally planned 5nm process, with shipments starting in the second half of 2025, according to industry sources. The upgrade is expected to boost performance by up to 30% compared to the previous design.
The shift comes as SK Hynix, which controls about half the global high-bandwidth memory market, races to meet accelerated timelines requested by Nvidia CEO Jensen Huang. The GPU maker asked SK Group Chairman Chey Tae-won to expedite delivery of 12-layer HBM4 chips by six months.
For broader market applications, SK Hynix will partner with Taiwan Semiconductor Manufacturing Co. to produce general-use HBM4 chips using 12nm technology. The company faces competition from Samsung Electronics Co., which plans to use 4nm process for its HBM4 products.
The strategy aligns with SK Hynix’s pivot toward U.S. tech giants like Nvidia, Google LLC and Microsoft Corp., reducing its exposure to China as Washington tightens export controls on advanced chips.
Tesla Inc. is also evaluating HBM4 prototypes from both SK Hynix and Samsung for potential supply deals.
