SK hynix samples next-gen 12-layer HBM4E chips, stock at new high

by Candice Kim Posted : June 18, 2026, 09:09Updated : June 18, 2026, 09:24
SK hynixs HBM4E Courtesy of SK hynix
SK hynix's HBM4E/ Courtesy of SK hynix

SEOUL, June 18 (AJP) - South Korea's SK hynix said on Thursday it has begun supplying samples of its next-generation 12-layer High Bandwidth Memory (HBM4E) chips to major customers, moving to solidify its position in the booming artificial intelligence hardware market.

The world’s leading supplier of HBM chips said the new HBM4E product delivers a data processing speed of up to 16 gigabits per second (Gbps) per pin while improving energy efficiency by more than 20 percent compared to the previous HBM4 generation. The improvements are aimed at reducing latency and expanding bandwidth for demanding AI training and inference workloads.

The announcement comes amid intense competition among global chipmakers to lock in supply deals with major AI processor designers like Nvidia Corp. SK hynix currently holds a significant lead in the HBM market, but domestic rival Samsung Electronics and U.S.-based Micron Technology are aggressively expanding their own advanced memory portfolios.

SK hynix said it utilized an advanced Mass Reflow Molded Underfill (MR-MUF) process to stack 12 DRAM chips, achieving a total capacity of 48 gigabytes (GB). The proprietary packaging method injects a liquid protective material between the stacked chips to improve structural integrity and heat dissipation.

According to the company, the HBM4E design reduces thermal resistance by approximately 17 percent compared to the standard HBM4 architecture, a key factor in maintaining operational stability under the intense heat generated by high-performance AI data centers.

"Based on our mass-production experience across previous HBM generations, we are working closely with key clients to ensure timely mass production," the company said.

Shares of SK hynix rallied on with record climb, adding 3 percent to 2,601,000 won a quarter after opening at 9:00 a.m. Thursday and nearly 24 percent in a week.  

Analysts note that as AI models grow in complexity, memory bottlenecks have become a critical constraint for data center operators, making thermal control and energy efficiency primary selling points for next-generation hardware.

This sampling milestone establishes a foundation for us to sustain our technological leadership," said Ahn Hyun, head of development at SK hynix. "We will continue to address system bottlenecks by working closely with our industry partners to deliver tailored memory solutions.