SK Hynix to boost memory production 3x ... you can wait another 8 years, right? - The Register

SK Hynix, the world's largest supplier of HBM memory used in high-end GPUs, plans to triple its wafer capacity to meet the surging demand for AI infrastructure. Despite this ambitious expansion, new c... Read original →

SK Hynix’s bet on tripling wafer capacity by 2034 reveals a structural mismatch between AI memory demand and manufacturing reality. Technically, HBM’s reliance on TSV and CoWoS packaging is forcing upstream OSATs like TSMC and ASE to co-develop advanced integration flows, while DRAM scaling near the 1α-node leaves little room for yield gains per wafer. Geopolitically, shifting some fab construction to Japan mitigates export control risks but adds 15–20% to CAPEX due to logistics, talent localization, and U.S.-Japan-Korea supply chain realignment. Samsung will likely counter with early HBM4 volume production, while Micron leverages CHIPS Act subsidies to accelerate its Idaho expansion—intensifying the battle for premium memory pricing power. Over the next 18 months, even as AI training demand moderates, inference-driven HBM consumption will keep prices rigid, delaying any meaningful DRAM/NAND price correction until SK’s first new fab comes online in 2027.