Samsung to develop custom chips for Microsoft, Meta

2024. 11. 12. 06:39
글자크기 설정 파란원을 좌우로 움직이시면 글자크기가 변경 됩니다.

이 글자크기로 변경됩니다.

(예시) 가장 빠른 뉴스가 있고 다양한 정보, 쌍방향 소통이 숨쉬는 다음뉴스를 만나보세요. 다음뉴스는 국내외 주요이슈와 실시간 속보, 문화생활 및 다양한 분야의 뉴스를 입체적으로 전달하고 있습니다.

(MK database)
Samsung Electronics Co. is developing high-bandwidth memory (HBM) chips customized for Microsoft Corp. and Meta Platforms Inc., according to industry sources on Monday.

“Microsoft has its own AI chip, Maia100, while Meta uses Artemis. Samsung is an ideal partner for these big tech firms as it has both memory and fabless divisions,” an industry insider said.

Big tech companies, many of which operate proprietary AI data centers, are increasingly looking to lower chip procurement costs. As a result, they develop custom AI accelerator chips alongside chips from fabless designers like Nvidia Corp. and Advanced Micro Devices Inc. (AMD).

Samsung Electronics aims to complete development of its HBM4 chips by the end of 2025, with mass production to follow. While specifications for the custom HBM4 chips for Microsoft and Meta have not been disclosed, Samsung shared HBM4 details at the 2024 International Solid-State Circuits Conference in February. The HBM4 offers a data transfer rate of 2 terabytes per second - 66 percent faster than the HBM3E- and stacks 16 layers of DRAM, boosting capacity to 48GB, a 33 percent increase over the current 36GB.

HBM4, labeled as a computing-in-memory chip, improves data transmission speeds and performance, supporting client-specific computational demands.

According to industry tracker TrendForce, the global HBM market is projected to grow to $46.7 billion in 2025 from $18.2 billion this year, a 2.5-fold increase. With Nvidia set to start mass production of its next-generation GPU, Rubin, in 2026, competition for HBM adoption is expected to intensify.

Copyright © 매일경제 & mk.co.kr. 무단 전재, 재배포 및 AI학습 이용 금지

이 기사에 대해 어떻게 생각하시나요?