South Korean semiconductor giant SK Hynix has notified its supply chain partners of plans to significantly expand production of its next-generation HBM4 (High Bandwidth Memory 4) memory chips as early as August, according to reports from the Korean newspaper Chosun Ilbo on June 9. Sources familiar with the matter revealed that SK Hynix's N2 facility in Pyeongtaek will double its monthly HBM4 wafer capacity from the current 30,000 units to over 60,000 per month, signaling the company's determination to capitalize on the explosive demand for AI infrastructure components.
HBM4: The Bottleneck Component for AI Chips
HBM (High Bandwidth Memory) has become an indispensable companion for AI training chips. Unlike conventional DRAM, HBM employs 3D stacking technology to vertically layer memory dies and connects them through silicon through-vias (TSV), achieving ultra-high bandwidth data transfer. In leading AI training processors such as NVIDIA's B200 and AMD's MI400, each GPU typically requires between 8 and 16 HBM3e or HBM4 memory chips, delivering aggregate bandwidth measured in multiple terabytes per second.
HBM4 represents the next standard after HBM3e, boosting bandwidth from the current 4.3GB/s to over 8.5GB/s while reducing power consumption by approximately 30 percent. Crucially, HBM4 introduces a more flexible chiplet architecture that can accommodate everything from cloud-scale training clusters to edge computing inference tasks. These attributes make HBM4 the most sought-after component among AI chip manufacturers, and SK Hynix's expansion is a direct response to the severe supply-demand imbalance currently gripping the market.
The Expansion Race: SK Hynix vs. Samsung
SK Hynix's move is part of a broader expansion race among South Korea's semiconductor titans. Rival Samsung Electronics is also planning to ramp up HBM capacity in the third quarter, with supply chain reports indicating that Samsung's Hwaseong facility has been urgently procuring advanced packaging equipment for HBM4 production. Japan's Micron is simultaneously accelerating its own HBM4 roadmap, aiming to increase annual output from 800,000 wafers to 1.5 million.
Tom O'Malley, lead semiconductor analyst at Goldman Sachs, noted: "The global HBM market is in a period of extreme supply-demand tension. SK Hynix is trying to exploit this window to widen its capacity lead over competitors. But Samsung will not let the gap grow unchecked. This is shaping up to be a defining moment for the memory industry."
Supply Chain Ripple Effects
The supply chain implications of the HBM expansion are being felt around the world. HBM4 manufacturing requires advanced packaging technologies such as TSMC's CoWoS (Chip-on-Wafer-on-Substrate), high-purity silicon interposers, and specialized cooling solutions. TSMC's shares rose 4.2 percent following news of SK Hynix's expansion, while Japanese equipment makers including Tokyo Electron and SCREEN Holdings reported a surge in orders for packaging-related machinery.
Chinese semiconductor firms are closely watching developments. Although domestic companies currently lag behind Korean and American peers in HBM technology, several major memory chip manufacturers have designated HBM4 as a priority R&D direction. A technical executive at Yangtze Memory Technologies Corp (YMTC) told media: "HBM4 is the most certain growth vector in the global memory market over the next three years. We cannot afford to fall behind."
The Golden Decade for HBM
Industry research firm TrendForce forecasts the global HBM market will reach $28 billion in 2026, grow to $42 billion in 2027, and potentially surpass $60 billion by 2028. The primary driver remains the insatiable demand for compute and memory bandwidth from AI large-language-model training. A senior semiconductor analyst speaking on condition of anonymity said: "From now through 2030, HBM will be a textbook seller's market. The companies that expand first will capture the dividend of this AI infrastructure boom."
SK Hynix's expansion plan is merely one indicator of how the global semiconductor industry is accelerating its restructuring under the AI wave. As memory chips become the "new oil" of the AI era, the race for capacity, technology leadership, and supply chain dominance has only just begun.