SK hynix's decision to start full-scale HBM4 DRAM production at its new Cheongju M15X fab marks a pivotal moment in the race to power the next generation of AI.
This strategic move is fundamentally a response to a clear and powerful demand signal from the market. The primary driver is NVIDIA, which has confirmed that its next-generation AI platform, codenamed 'Rubin', will adopt HBM4 memory. When NVIDIA began sampling its Rubin GPUs in March 2026, it locked in the demand for HBM4, giving SK hynix a clear target and the confidence to accelerate its production timeline. This isn't just a hopeful bet; it's a calculated response to a customer's confirmed technology roadmap.
To meet this demand, SK hynix has been making a series of aggressive, forward-looking investments. First, the company placed a massive ₩11.95 trillion order for EUV lithography machines. These state-of-the-art tools are essential for manufacturing the complex, high-density HBM4 chips and represent a foundational investment in next-generation technology. Without this, mass production would be impossible.
Second, SK hynix is building a comprehensive 'HBM hub' in Cheongju by coordinating both front-end and back-end processes. The company opened the M15X cleanroom two months ahead of schedule to speed up wafer production (front-end). Simultaneously, it announced a ₩19 trillion investment in the adjacent P&T7 facility for advanced packaging (back-end). This integrated approach is designed to tackle production bottlenecks holistically, from creating the silicon wafers to stacking and packaging the final HBM chips.
These bold investments are further justified by the broader market outlook. SK Group's chairman publicly stated that the industry faces a wafer supply shortage of over 20% that could last until 2030. This long-term view validates the strategy of making large, preemptive investments to secure capacity and solidify market leadership. By acting now, SK hynix aims to resolve its production constraints, maintain its lead in the HBM market, and capture the immense growth fueled by the AI supercycle. However, the path forward includes challenges, such as potential yield issues with the new technology and intensifying competition from rivals like Samsung and Micron, who are also racing to bring their HBM4 products to market.
- HBM (High Bandwidth Memory): A type of high-performance stacked memory architecture used alongside GPUs to provide ultra-fast data access, crucial for AI and high-performance computing.
- EUV (Extreme Ultraviolet) Lithography: An advanced semiconductor manufacturing technology that uses extremely short wavelength light to etch circuits onto silicon wafers, enabling the creation of smaller and more powerful chips.
- Ramp-up: The period during which a factory increases its production volume to reach full capacity after starting initial operations.