[Reporter’s Notebook] GPU Crisis, the Minister’s Lament

"With the National Assembly budget (supplementary budget) frozen, we have already lost several months this year. We are suffering tremendous damage due to the Graphics Processing Unit (GPU) shortage. This year is the most worrisome."

Yoo Sang-im, Minister of Science and ICT, sighed deeply while making these remarks at a recent briefing held at the Government Seoul Office. She expressed her frustration as the supplementary budget, which includes GPU purchases, remains stalled in the National Assembly. Minister Yoo also muttered to herself, "I wish I had enough money to buy these (GPUs) myself..." and lamented, "If we waste the remaining nine months of this year, it would be equivalent to losing three years in the AI technology market."

Minister Yoo’s comments are understandable given the global competition for dominance in the AI market. Nowadays, GPUs are so scarce that their prices are sky-high. Nvidia GPUs, which occupy more than 80% of the AI semiconductor market, are uncertain in terms of supply even if purchased now. Despite this situation, South Korea’s GPU procurement has been hampered by political strife and budget issues. In a 100-meter sprint, while other competitors are running at full speed, the Korean athlete is just pacing nervously at the starting line.

The government is in a hurry. Failure to secure GPUs in a timely manner could set the country back not by three years, but possibly by 30 years. The National Assembly must reach a bipartisan agreement to secure the supplementary budget so that by next year, a national AI computing center equipped with 20,000 GPUs can be built and world-class large language models (LLMs) can be developed. Although large corporations and platform companies are developing AI, it is difficult without partnering with overseas big tech firms. To build our own AI data center and train models, purchasing a large number of GPUs is the first step. The future where domestically produced fully autonomous vehicles run on roads and K-robots work in factories and homes instead of humans can only be realized with GPUs.

Nvidia is racing ahead, announcing GPUs with improved performance every year. Although it is number one, it does not rest on its laurels. This year, it is focusing on selling the Blackwell GPU, and from next year, it will mass-produce the next-generation GPU called 'Rubin.' These products are sold in large quantities to overseas big tech companies and are used to deliver ever-evolving AI technologies.

Even Jensen Huang, CEO who has led Nvidia for over 30 years to the top position globally, runs the company with a sense of crisis, saying, "We could go bankrupt within the next 30 days." Yet, ironically, the National Assembly, which should be most aware of this crisis, is stuck in a deadlock over the supplementary budget discussions. When it comes to the 'AI supplementary budget,' bipartisan political disputes must be set aside, and a magnanimous agreement should be reached with concern for the national economy and future.