Samsung ships latest HBM4 chips to catch-up in AI race

By Hyunjoo Jin

SEOUL, Feb 12 (Reuters) – Samsung Electronics said on Thursday it had started shipping its most advanced HBM4 chips to unnamed customers, as it tries to narrow the gap with rivals in supplying critical parts for Nvidia’s AI accelerators.

The global rush to build AI data centers has fuelled demand for such chips that feed massive amounts of data to AI accelerators for training and running AI models.

Samsung, the world’s top memory chipmaker, had been slow in responding to the advanced HBM chip market, lagging behind rivals, including SK Hynix, in supplying previous-generation HBM chips.

But Samsung aims to narrow the gap with SK Hynix, the dominant leader. Song Jai-hyuk, chief technology officer for Samsung Electronics’ chip division, touted the performance of its HBM4 chips on Wednesday, saying customer feedback had been “very satisfactory.”

Samsung said its HBM4 delivers a consistent processing speed of 11.7 gigabits-per-second (Gbps), a 22% increase from its predecessor, HBM3E. Samsung said its latest chips can achieve the maximum speed of 13Gbps, which helps mitigate growing data bottlenecks.

Samsung also said it planned to deliver samples for next-generation HBM4E chips in the second half of this year.

Samsung shares ended up 6.4% on Thursday, while SK Hynix closed up 3.3%.

RISING COMPETITION

SK Hynix said in January that it aims to maintain its “overwhelming” market share in the next-generation HBM4 chips, which are in volume production, as it faces rising competition from Samsung Electronics.

It also added that it aimed to achieve the production yields of HBM4 similar to those of current-generation HBM3E chips.

Micron’s CFO also said it was in high-volume production of HBM4 and has commenced customer shipments of HBM4, according to media reports.

(Reporting by Hyunjoo Jin and Heekyong Yang; Editing by Rashmi Aich, Louise Heavens and Anil D’Silva)