Key Points:
- Samsung Electronics will supply Mach-1 AI chip to Naver Corp. in a deal worth up to 1 trillion won ($752 million) by the end of the year.
- The collaboration aimed to reduce Naver’s reliance on Nvidia for AI chips and expand Samsung’s client base.
- Mach-1 chip promises enhanced AI acceleration with minimized data bottlenecks and power consumption while offering a competitive price.
- Samsung aims to compete with SK Hynix in the advanced HBM segment with recent developments in HBM technology.
Samsung Electronics Co., the leading memory chipmaker globally, has announced plans to supply its upcoming artificial intelligence chip, Mach-1, to Naver Corp. in a deal valued at up to 1 trillion won ($752 million) by the end of the year. This move is poised to significantly lessen Naver’s dependence on Nvidia Corp. for its AI chip requirements.
Under the agreement, Samsung’s System LSI business division has reached a supply deal with Naver, with final negotiations underway to determine the exact volume and pricing. Sources familiar with the matter suggest that Samsung aims to price the Mach-1 AI chip at around 5 million won ($3,756) each, with Naver seeking to procure between 150,000 and 200,000 units of the AI accelerator.
Naver, a prominent Korean online platform, is expected to integrate Mach-1 chips into its servers for AI inference, replacing Nvidia chips currently in use. Leveraging this collaboration as a springboard, Samsung plans to expand its client base to include major tech firms, with ongoing supply discussions with Microsoft Corp. and Meta Platforms Inc. showing promising signs.
The Mach-1 AI chip, currently under development and set for mass production of a prototype by year-end, promises to revolutionize AI acceleration. It functions as a system-on-chip (SoC) designed to mitigate bottlenecks between the graphics processing unit (GPU) and high bandwidth memory (HBM) chips. Samsung’s Kyung Kye-hyun, head of the semiconductor business, highlights Mach-1’s efficiency in reducing bottlenecks and enhancing power efficiency, making it suitable for large language model inference.
Unlike Nvidia’s AI accelerator, which relies on GPUs and HBM chips, Mach-1 integrates Samsung’s proprietary processors and low-power (LP) DRAM chips. This design minimizes data bottlenecks and power consumption while offering a competitive price point at one-tenth the cost of Nvidia products.
Naver’s decision to adopt Samsung’s Mach-1 chips aligns with its strategy to diversify its AI chip suppliers, which had previously transitioned from Nvidia to Intel’s CPU-based servers. The global IT landscape has witnessed discontent with Nvidia’s GPU pricing hikes and supply shortages, prompting companies like Naver to seek alternative solutions.
For Samsung, the collaboration with Naver presents an opportunity to compete with industry rivals, particularly in the advanced HBM segment dominated by SK Hynix Inc. Samsung’s recent developments in HBM technology, including the industry’s first 12-stack HBM3E DRAM, signify its commitment to challenging competitors and expanding its market share.
