SK Hynix Plans to Double Wafer Capacity as AI Memory Chip Shortage Intensifies

The Quick Take

• SK Hynix has announced plans to double its wafer production capacity in response to a shortage of AI-driven memory chips
• The capacity expansion targets the surging demand for High Bandwidth Memory (HBM) chips used in AI accelerators
• The doubling plan represents one of the largest capex commitments in the memory chip sector's recent history

SK Hynix, the South Korean memory chip maker and world's leading producer of High Bandwidth Memory used in NVIDIA's AI accelerators, has announced plans to double its wafer production capacity in direct response to accelerating AI-driven memory chip demand. The memory chip shortage — characterized by more demand for HBM3 and HBM3E than current production capacity can meet — has put SK Hynix in the rare position of being supply-constrained rather than demand-constrained, a reversal of the oversupply cycle that defined the memory market in 2022-2023. The capacity doubling is a multi-year capital commitment that fundamentally reshapes the memory supply landscape.

The expansion has broad sector implications. Equipment suppliers including ASML, Applied Materials, Lam Research, and Tokyo Electron will see demand for lithography and process equipment increase substantially as SK Hynix builds out cleanroom capacity. Samsung, which is working to catch up to SK Hynix's HBM technology lead, and Micron Technology — the US-based memory giant competing aggressively for AI chip supply deals — face a strategic challenge: if SK Hynix doubles capacity, their own market share in the AI memory segment becomes harder to reclaim without matching investment. The timing of SK Hynix's capacity addition will determine whether the HBM shortage extends into 2027 or resolves sooner.

Key forward signals include SK Hynix's formal capex guidance update and the timeline for when the doubled capacity comes online — typically 18-24 months from groundbreaking to production. Investors should watch NVIDIA's Q2 earnings for any comments on HBM supply adequacy, as NVIDIA's ability to ship more H100/H200/B-series AI accelerators depends directly on SK Hynix memory availability. The macro variable that determines whether the investment pays off is the sustained growth rate of AI infrastructure buildout: if enterprise AI capex growth moderates materially, memory chip oversupply could return before the new capacity is absorbed.

Synthesized from 4 sources.