Jensen Huang, the CEO of Nvidia, announced at the company’s GTC 2025 developer conference in San Jose that a new AI chip called Rubin would be released in the latter half of 2026. (Image courtesy of Yonhap)
SEOUL, March 24 (Korea Bizwire) — In a high-stakes battle for dominance in artificial intelligence technology, major semiconductor manufacturers are racing to develop HBM4, the next generation of high-bandwidth memory chips crucial for advanced AI processors.
The competition intensified on March 18 when Jensen Huang, the CEO of Nvidia, announced at the company’s GTC 2025 developer conference in San Jose that a new AI chip called Rubin would be released in the latter half of 2026. Rubin will be Nvidia’s first processor to incorporate HBM4 technology, marking a significant milestone in the evolution of AI hardware.
SK Hynix, the current market leader in high-bandwidth memory, made a surprise announcement the same day, revealing it had become the first manufacturer to provide 36GB HBM4 12-layer samples to major customers.
The announcement’s significance was underscored when Jeff Fisher, a senior vice president at Nvidia, visited SK Hynix’s conference booth and left a congratulatory message about the HBM4 sample.
The timing of SK Hynix’s announcement coincided with Samsung Electronics’ annual shareholders’ meeting, where Kyung Kye-hyun, head of Samsung’s semiconductor division, acknowledged the company’s delayed response to the HBM market.
“We were late in reading market trends for HBM, which is a key component in the AI era,” he said. “We are methodically preparing for next-generation products like HBM4 to avoid similar mistakes.”
Jeff Fisher, a senior vice president at Nvidia, visited SK Hynix’s conference booth and left a congratulatory message about the HBM4 sample. (Image courtesy of Yonhap)
Samsung Electronics and SK Hynix are targeting mass production of HBM4 in the second half of 2025, while U.S.-based Micron Technology aims to begin production in 2026.
Industry observers expect Samsung to focus more heavily on HBM4 than its predecessor, HBM3E, where SK Hynix has already established market dominance. “The HBM3E market is already saturated, with SK Hynix announcing it has sold out this year’s production capacity,” said a semiconductor industry insider, speaking on condition of anonymity.
Technical innovation will be crucial in the HBM4 competition, particularly in chip stacking methods. Both Samsung and SK Hynix are expected to adopt a new hybrid bonding technique for certain HBM4 products, departing from their current different approaches.
Samsung uses TC-NCF (thermal compression non-conductive film), while SK Hynix employs MR-MUF (mass reflow-molded underfill) technology.
The manufacturing process for HBM4′s logic die will also be critical. Samsung’s memory division plans to strengthen collaboration with Samsung Foundry, which has achieved high yields with its 4-nanometer process technology.
Meanwhile, SK Hynix and Micron, lacking their own foundry capabilities, plan to partner with Taiwan’s TSMC for HBM4 production.
Kevin Lee (kevinlee@koreabizwire.com)
