SK Hynix has moved up the mass production schedule of HBM4 at its newly established Cheongju M15X plant in South Korea—a facility specifically designated for HBM4 manufacturing—by four months, bringing the start date to February of the coming year. Initially, the plant will have a production capacity of roughly 10,000 wafers, with projections to scale up to tens of thousands of wafers by the year's end. The installation of pertinent equipment is proceeding swiftly.

This strategic adjustment is viewed as a preemptive move to meet NVIDIA's requirements, given that NVIDIA's next-gen AI accelerator, dubbed "Rubin," is anticipated to hit the market in the latter half of next year and will necessitate HBM4 memory. Presently, SK Hynix has furnished samples and is working in tandem with NVIDIA and TSMC for evaluation purposes.

The global HBM market is characterized by intense rivalry. Samsung has already supplied samples and carried out testing, whereas Micron's HBM4 production capacity for the upcoming year is completely booked. NVIDIA is slated to unveil its final supplier selection and share allocation early next year, with power consumption efficiency emerging as a pivotal assessment criterion. Samsung, for its part, is making concerted efforts to enhance its process yield.