Chips · Oct 21, 2026, 13:26
Samsung, SK hynix reportedly move China fabs to 286-layer NAND
Samsung is said to be converting its Xian plant to 286-layer flash by year-end, and SK hynix its Dalian plant to 240+ layers, to hold off YMTC and CXMT as AI data centers drive demand.
Samsung Electronics and SK hynix are reportedly upgrading their Chinese memory plants to make more advanced chips for AI data centers, according to a Reuters report relayed by Tom's Hardware on 21 October 2026.
Citing four people familiar with the plans, the report says Samsung aims to switch its Xian plant to 286-layer flash by the end of the year. SK hynix is targeting chips with 240 or more layers at its Dalian site, which it took over in 2021, though no timescale for that upgrade has been given. Xian currently makes chips of around 200 layers and accounts for roughly 40% of Samsung's global NAND output, while Dalian currently makes 192-layer parts.
The upgrades matter for AI infrastructure because flash storage sits alongside HBM and DRAM in the supply chain that feeds accelerator clusters. Higher layer counts pack more capacity into each die, and every new AI data center needs large pools of fast storage next to its GPUs. The report frames the move as a defence against Chinese rivals such as YMTC, which has told investors it aims to become the world's largest NAND maker by the end of 2027, and CXMT, which is reportedly eyeing 3D NAND after reporting $22.41 billion in revenue for the first half of 2026.
US export controls complicate the plan. Since 2022, South Korean firms have needed licences to install US-built chipmaking tools in Chinese fabs, and both companies say they are working in consultation with US authorities. That extra approval step is cited as one reason the upgrades have been slow.
Still unconfirmed: neither company has publicly announced the schedules, the licence status of the specific tools is not stated, and the report gives no output volumes. Voltgrid will track whether the Xian and Dalian conversions are confirmed, since NAND supply for AI servers also draws on the same fab power and equipment budgets as memory for accelerators.