Large GPU shipment expected to boost Korea’s AI

Nvidia’s pledge to supply 260,000 latest-generation graphics processing units is expected to ease a severe shortage that has hampered South Korea’s AI industries. The influx is expected to energize government-led projects, including the sovereign AI initiative, and boost private-sector AI development, enhancing South Korea’s competitiveness in the field.

Nvidia’s GPUs are in short supply worldwide amid a global AI boom. Each unit costs about 100 million won, and buyers often face months-long waits even after placing orders. The shortage reflects limited production and large purchases by tech giants such as OpenAI, Google and Amazon. U.S. export controls aimed at curbing China’s technological rise have further tightened supply.

South Korea has long underinvested in AI infrastructure, resulting in a severe GPU shortage. Government agencies, universities and companies lacked the computing resources needed for AI research and development, raising concerns that the country’s competitiveness in advanced industries could decline. By the end of 2023, public and private holdings in South Korea totaled only about 2,000 GPUs, far fewer than the tens of thousands held by major global tech companies. Combined, the government and domestic firms still fall short of the computing capacity of a single foreign tech giant.

Analysts say Korea missed the first wave of the race around foundation models, and the current phase, when AI is expanding into physical applications and reasoning tasks such as robotics, autonomous driving and advanced medicine, represents a narrow window to regain global leadership. AI chips, including Nvidia’s GPUs, are fueling a second boom as systems begin performing advanced reasoning and decision-making.

The large GPU delivery gives South Korea a chance to accelerate its AI efforts. Officials and industry observers said the sovereign AI program, aimed at positioning Korea alongside the United States and China, is expected to gain momentum. In August, the government selected five teams, including Naver, Upstage, SK Telecom, NC AI and LG AI Research Institute, to develop independent foundation models.

The GPU surge is also expected to benefit memory-chip makers such as Samsung Electronics and SK hynix. Nvidia’s Blackwell GPUs supplied to Korea use high-bandwidth memory, including HBM3E from the two companies. There is also potential for an expanded HBM4 supply for Nvidia’s next-generation Rubin GPUs.


박현익 beepark@donga.com