hankyoreh

Lee Jong-ho, South Korea’s minister of science and ICT. (Yonhap)

The Korean government is teaming up with domestic corporations to support the development of locally made artificial intelligence (AI) semiconductors so that services utilizing large AI language models can be operated without the need for graphics processing units (GPU) manufactured by Nvidia, a US company.

Instead of directly challenging Nvidia’s dominance in AI training chips, the plan aims to reenact the “myth of memory chips” in the market for low-power AI inference neural processing units (NPU), which domestic AI chipmakers possess the development capabilities for.

On Monday, the country’s Ministry of Science and ICT held its third strategic dialogue for top AI chip officials at the NHN Cloud building in Seongnam, Gyeonggi Province, where it announced the launch of the first stage of the “K-Cloud Project.” Officials from conglomerates like Samsung Electronics and SK Hynix as well as domestic AI chip startups and the cloud computing industry were in attendance.

The K-Cloud Project, which the ministry has been promoting since December of last year, seeks to put to use Korean-made NPUs in the early stages of commercialization at actual data centers so that references of market verification can be secured to test out and ultimately export cloud platforms. The plan is to advance the technological prowess of domestic AI chips by investing 826.2 billion won (US$634.5 million) by the year 2030.

More specifically, during the first stage of the project, which will see investments of 100 billion won by the end of 2025, both the private and public sectors will establish data centers with a total operational capacity of 39.9 petaFLOPS (PF), where 1 PF signifies the capability to perform 1 quadrillion floating-point operations in one second. Cloud companies like Naver Cloud, NHN Cloud, and KT Cloud, as well as AI chip companies like Rebellions, Sapeon Korea, and FuriosaAI and AI service developers like Upstage, Laon Road, and Nota will be participating as a consortium.

“Considering that NPUs cost less and consume less electricity, relatively speaking, they are suitable as an alternative to CUDA, Nvidia’s high-cost, high-power GPU acceleration platform,” an official from the ministry said.

Lee Jong-ho, Korea’s minister of science and ICT, remarked, “Global corporations are competing fiercely in order to make high-performance, low-power AI chips optimized for AI operation,” adding the MSIT will provide “proactive support” in order to “win the global chip war.”

Gartner, a global market research company, projected that the AI chip market, worth US$34.7 billion in 2021, will rapidly grow at a yearly rate of 16%, reaching a market size of US$86.1 billion by 2026, approximately half that of the memory chip market today.

By Chung In-seon, staff reporter

Please direct questions or comments to [english@hani.co.kr]