SK Hynix Accelerates Capacity Expansion, Samsung's DRAM Capacity Lead Narrows to 14%
Nashnova编辑部
SK Hynix is growing DRAM wafer capacity at nearly 10% annually while Samsung's growth drops to 1.3%, narrowing the gap from 27% to roughly 14% — a material shift in the DRAM duopoly's balance of power within two years.
How much has the capacity gap actually shrunk?
In 2025 Samsung leads SK Hynix by 1.635 million wafers (~27%); by 2027 that falls to 990,000 wafers (~13.6%).
This means → the gap contracts by roughly 39% in two years — SK Hynix moves from "clearly behind" to "close pursuit."
The driver is a widening growth-rate scissors: SK Hynix sustains 9.5–9.9% annual expansion in 2026-2027, while Samsung drops from 6.2% to just 1.3%.
Why is Samsung deliberately slowing expansion?
Capital spending is pivoting from "build more wafer lines" to "upgrade existing lines to more advanced process nodes."
Pyeongtaek's share of Samsung's total DRAM capacity rises from 49.5% (2025) to 55.4% (2027), driven by conversion to the 6th-gen 10nm-class (1c) node — a more advanced manufacturing process that packs more data bits per wafer.
In plain terms = Samsung is choosing "make each wafer more valuable" over "make more wafers" — wafer count slows, but bit output per wafer rises.
What is Samsung betting on?
Samsung's HBM4 is in mass production using 1c DRAM paired with a 4nm logic base die, achieving a stable 11.7 Gbps processing speed.
HBM revenue is projected to more than triple in 2026; HBM4E samples have shipped to key customers.
This means → Samsung's logic: during an AI demand surge, selling high-value products (HBM, DDR5, SOCAMM2) is more profitable than stacking wafer volume — the slowdown is a strategic choice, not a stumble.
What gives SK Hynix the confidence to expand aggressively?
The M15X fab is accelerating mass-production timelines; Yongin Phase 1 cleanrooms are set to open in early 2027 with a rapid capacity ramp.
SK Hynix has signed long-term supply agreements with roughly 10 customers — the reason is straightforward: existing supply already cannot meet demand.
This reflects SK Hynix's conviction that AI-infrastructure DRAM demand is not a short-term pulse but a multi-year structural uptrend, justifying continued investment within a capex-discipline framework.
What does the 2027 DRAM market look like?
Samsung expects server DRAM, enterprise SSD, and HBM demand to accelerate in H2 2026, with the overall market remaining undersupplied.
In plain terms = the two rivals differ in tactics but agree on direction — both believe AI-driven memory demand is far from peaking.
Whether SK Hynix can leverage its faster capacity cadence to compress the gap further is the key variable shaping the 2027 DRAM market landscape.
Content is for reference only, not financial advice.