Physical samples of sixth-generation high-bandwidth memory, HBM4 and HBM3E, are on display at Samsung Electronics’ booth at the Semiconductor Exhibition (SEDEX) 2025, held at COEX in southern Seoul. (Yonhap)
SEOUL, Feb. 2 (Korea Bizwire) — Samsung Electronics and SK hynix are heading toward a high-stakes showdown over sixth-generation high bandwidth memory, or HBM4, with sharply contrasting strategies underscoring a broader battle for leadership in the artificial intelligence chip supply chain.
HBM, a stacked memory technology essential for training and running advanced AI models, is expected to enter a new growth phase this year as next-generation AI processors from companies like Nvidia begin adopting HBM4.
Major cloud companies, including Google, Microsoft and Amazon Web Services, are also expected to follow suit as they expand their use of custom AI chips.
Samsung is pursuing an aggressive “first-mover” strategy, betting that early mass production using cutting-edge manufacturing processes will allow it to regain momentum after losing ground to SK hynix in the previous HBM3E generation.
SK hynix, by contrast, is emphasizing stability, yield and cost competitiveness, seeking to extend its dominant position as the market leader.
On Jan. 8 (local time), the third day of CES, a video showcasing HBM4 is playing at the SK hynix booth at the Venetian Convention Center in Las Vegas, Nevada. (Yonhap)
During earnings calls in late January, Samsung said it had completed key quality tests and was already moving HBM4 into mass production, with shipments of top-tier products scheduled to begin in February at the request of major customers.
The company is using its latest sixth-generation 10-nanometer-class DRAM process alongside a 4-nanometer logic die produced by its foundry unit, aiming to deliver industry-leading performance.
The approach carries risks. Advanced processes can drive up costs and depress early production yields, but Samsung appears willing to sacrifice short-term profitability to secure early market share and reshape customer perceptions.
Samsung Electronics’ semiconductor production lines at its Pyeongtaek campus (image provided by Samsung Electronics).
With the largest production capacity among memory makers, the company hopes an early foothold will translate into rapid sales expansion.
SK hynix, which supplies the vast majority of current HBM3E volumes, is taking a more conservative path. It is building HBM4 using a previous-generation DRAM process and a logic die produced by Taiwan Semiconductor Manufacturing Company, while still achieving the same top speeds targeted by Samsung.
Executives said the company expects to maintain high yields using its proprietary packaging technology, reinforcing both cost advantages and manufacturing stability.
Industry sources say Nvidia is likely to allocate roughly two-thirds of its initial HBM4 orders to SK hynix, reflecting the close partnership forged during the HBM3E cycle.
An employee working at a semiconductor fab (factory) (Image courtesy of SK hynix)
Market researchers at Counterpoint estimate that SK hynix will command about 54 percent of the global HBM4 market this year, compared with 28 percent for Samsung and 18 percent for Micron.
As HBM4 becomes the backbone of the next wave of AI hardware, the divergent strategies highlight a fundamental choice facing chipmakers: whether to lead with speed and advanced technology, or with reliability and scale. The outcome could reshape the balance of power in one of the semiconductor industry’s most lucrative segments.
Kevin Lee (kevinlee@koreabizwire.com)
