Samsung races ahead of SK hynix with HBM4 rollout

Home > Business > Industry

Samsung races ahead of SK hynix with HBM4 rollout

Published: 12 Feb. 2026, 17:50 Updated: 12 Feb. 2026, 19:29

LEE JAE-LIM
[email protected]

Audio report: written by reporters, read by AI

Samsung Electronics' sixth-generation high bandwidth memory, or HBM4, chips began mass production and shipments to major customers, the company said on Feb. 12. [SAMSUNG ELECTRONICS]

Samsung Electronics said Thursday it has kicked off mass production of its sixth-generation high bandwidth memory (HBM4) chips, becoming the first in the industry to do so and started shipments to major customers including Nvidia ahead of rival SK hynix.

The Suwon, Gyeonggi-based chipmaker has initially planned to begin mass shipments shortly after Korea's Lunar New Year holiday, which runs through Wednesday. However, the timeline was moved up by about a week following discussions with customers.

HBM4 was developed using Samsung's latest 10-nanometer-class 1c dynamic random access memory (DRAM) process. The company said the new process achieved stable yields and industry-leading performance from the outset without requiring redesigns.

A truck with HBM4 chips that kicks off delivery to be supplied to customers worldwide on Feb. 12. [SAMSUNG ELECTRONICS]

“Instead of taking the conventional path of utilizing existing proven designs, Samsung took the leap and adopted the most advanced nodes like the 1c DRAM and 4-nanometer logic process for HBM4,” said Hwang Sang-joon, executive vice president and head of memory development at Samsung Electronics. “By leveraging our process competitiveness and design optimization, we are able to secure substantial performance headroom, enabling us to satisfy our customers’ escalating demands for higher performance, when they need them.”

Samsung said its HBM4 delivers the industry's highest processing speed of 11.7 gigabits per second (Gbps), 46 percent above the current industry standard of 8 Gbps and 22 percent faster than maximum 9.6 Gbps pin speed of its predecessor, HBM3E.

Performance can be further boosted to 13 Gbps to help ease data bottlenecks that intensify as AI models grow larger and more complex.

The 12-layer HBM4 offers capacities ranging from 24 gigabytes (GB) to 36 GB. The lineup will later be expanded to include 16-layer versions with up to 48 GB, in line with customer production schedules.

Compared with HBM3E, total memory bandwidth per stack has nearly tripled to a maximum of 3.3 terabytes per second. Power efficiency improved by 40 percent through the use of low-voltage through-silicon via (TSV) technology and optimized power distribution networks. Thermal resistance increased by 10 percent, while heat dissipation improved by 30 percent.

Samsung Electronics' sixth-generation high bandwidth memory (HBM) chips [SAMSUNG ELECTRONICS]

Samsung said it is consistently receiving supply requests from global hyperscalers that design their own customized AI chips and plans to deepen technological collaboration with those clients.

HBM revenue this year is expected to triple compared with last year, driven by strong AI-driven memory demand. The company is expanding production capacity and plans to begin manufacturing HBM chips at its P5 facility in Pyeongtaek, Gyeonggi, from 2028.

Samsung is also preparing to ship samples of next-generation HBM4E in the second half of this year, with customized HBM products scheduled for release in 2027. Custom HBM refers to memory tailored to a customer's AI accelerator or GPU architecture, with specifications — including capacity, speed, power characteristics and interface — designed to meet specific requirements.

“The quality and supply stability secured through the 1c process in the mass production of HBM4 are expected to serve as key advantages as the company transitions to higher-value products, including HBM4E and custom HBM variants,” a Samsung Electronics spokesperson said in a statement.

BY LEE JAE-LIM [[email protected]]