The world’s top two memory-chip makers Samsung Electronics and SK hynix are set to go head-to-head this year for HBM4 dominance, with the next-generation chip poised to determine their positions in the artificial intelligence chip market.

In a bid to retain its leadership in the lucrative AI market, SK hynix showcased a sample of the six-generation HBM4 chip last week at GTC 2025, a developer conference hosted by the industry's largest customer for AI chips, Nvidia.

Having failed to gain an early edge in the lucrative market, Samsung Electronics renewed its commitment to make up for its past missteps to overtake the smaller rival with the upcoming HBM4 chips.

At the regular shareholders meeting last week, Samsung Vice Chairman Jun Young-hyun leading the company's semiconductor business pledged to boost its HBM business, and announced to kick off mass production of the HBM4 and custom HBM chips in the second half of this year -- moving up the launch by six months.

"We initially missed out on the early high-bandwidth memory market due to a delayed response, but we have since restructured the company and laid the foundation for all necessary technology development," Jun said Wednesday.

"We will do our utmost to meet shareholder expectations in our HBM business, with all employees working together to ensure we do not repeat past mistakes."

Jun also said the company would start mass production of the 12-layer HBM3E chips, the fifth generation of the AI chip, in the second quarter or the second half of this year.

Samsung, however, is expected to focus on the upcoming HBM4 chips rather than the 12-layer HBM3E chips, which SK hynix has already secured the position as the main supplier since 2024, according to industry sources. Samsung is still waiting to clear Nvidia's qualification test for the 12-layer HBM3E for supply.

"SK hynix, which has secured majority supply for Nvidia, said its HBM chips for this year have already 'sold out,' meaning the HBM3E market is saturated," an industry official said.

"Samsung would not be able to profit so much from the 12-layer HBM3E even if it does start supply to Nvidia, so Samsung would likely bet on the upcoming HBM4 for dominance."

The key for the chipmakers is to secure a supply contract for Nvidia’s next-generation Rubin architecture, which the GPU giant unveiled at the GTC 2025 event last week and is set for launch in the second half of next year.

SK hynix, the world's second-largest memory chip maker after Samsung, is keeping up momentum in the HBM race, and has introduced the prototype of the 12-layer HBM4 chip at GTC 2025 last week. The product, which is currently under development, is expected to be mass-produced in the second half of this year.

During the GTC 2025 event, Nvidia Senior Vice President Jeff Fisher visited the SK hynix booth and acknowledged the HBM4 samples, hinting at their potential collaboration.

SK hynix is currently delivering 12-layer HBM3E chips to global tech giants, mainly Nvidia, and is preparing for production of the more advanced 16-layer HBM3E chips in the first half of this year.

One highlight of the HBM4 rivalry between Samsung and SK hynix is that both companies will use the same packaging methods, unlike previous generation models.

Samsung, which previously used the Thermally Conductive Non-Conductive Film, is transitioning to the Mass Reflow-Molded Underfill method to stack the DRAM chips for its HBM4 product. SK hynix, which already uses MR-MUF will continue with the same approach for its next-generation product.

For the logic die featured for the first time with HBM4, Samsung will use the ones produced in-house by its foundry division, while SK hynix will source its supply from TSMC, the Taiwan-based contract chip manufacturer.

By: DocMemory
Copyright © 2023 CST, Inc. All Rights Reserved