Samsung Electronics (005930.KS) and SK hynix (000660.KS) each strengthened partnerships to expand artificial intelligence (AI) memory supply through Computex 2026, held in Taipei, one of the global hubs for AI chips. SK hynix pledged to double its production capacity over the next five years, while Samsung Electronics unveiled successive technology leads in high-bandwidth memory (HBM).
A year ago, SK had focused on technology innovation while Samsung concentrated on capacity expansion. Now the two companies have swapped strategies to shore up their respective weaknesses, analysts say. Samsung and SK, the flagships of K-semiconductors, are reinforcing both technology and production while expanding cooperation with global firms to further solidify their positions.
SK Group Chairman Chey Tae-won visited the SK hynix booth at Computex 2026 on Tuesday and told reporters that "the memory bottleneck will persist through 2030," adding, "We will double our capacity over the next five years." With Nvidia CEO Jensen Huang having announced in his GTC Taipei keynote the previous day that Nvidia would double the production capacity of its 'Vera Rubin' platform through Taiwan's supply chain, SK hynix likewise faces the need to expand DRAM supply.
"Building a new fab now takes at least three years, and starting from a greenfield site takes more than five years," Chey said. "Resolving the bottleneck involves a major hurdle on a roughly 10-year scale."
Behind SK hynix's emphasis on capacity lies a structural memory supply shortage driven by the AI boom. According to industry sources, the volume that the three memory makers, including SK hynix, can currently supply stands at only around 50% of the medium- to long-term demand required by global Big Tech firms. SK hynix is accordingly investing 31 trillion won in its first fab (Y1) at the Yongin semiconductor cluster in Gyeonggi Province to aggressively expand capacity. Y1 is expected to begin full operation after completion in the first quarter of next year and to reach DRAM production capacity of 300,000 wafers per month by year-end.
To accelerate the capacity race, SK hynix plans to mobilize the entire group's resources to build out 'AI factories.' "In the future, many AI factories will be needed to produce more intelligence, so SK also wants to build more AI factories," Chey said.
Samsung Electronics, for its part, unveiled an HBM innovation roadmap aimed at staying a step ahead of rapidly advancing AI accelerators. Song Jai-hyuk, Chief Technology Officer of Samsung Electronics' Device Solutions (DS) division, held a technology briefing at Samsung Display's booth at Computex 2026 the same day and presented an HBM5 prototype. Samsung's HBM5 applies a base (logic) die built on its own 2-nanometer (one-billionth of a meter) foundry process, demonstrating a technology gap that vaults well beyond the current 4nm.
Samsung had initially been seen as somewhat lagging in the HBM race. The company was passive in developing HBM1, when end-demand was unclear, and struggled to pass qualification tests for HBM3E used in Nvidia's Blackwell GPU during the most competitive phase. However, Samsung maximized development speed through its vertically integrated structure spanning all business divisions, including memory, and has been leading the industry since HBM4 earlier this year.
"In responding to a rapidly changing industry, our total solution encompassing memory, foundry, logic and packaging has created differentiation," Song said. "The number of stacked layers has not been finalized, but we are considering structures of up to 20 layers."
