Home > Business > Industry

Korean experts urge gov't to prioritize AI infrastructure over LLMs

Published: 19 Feb. 2025, 18:54 Updated: 19 Feb. 2025, 18:56

LEE JAE-LIM
[email protected]

A seminar was convened on Feb. 19 to discuss ways to enhance Korea's AI competitiveness following the debut of China's DeepSeek. Government officials, professors and industry experts held a panel discussion at the seminar at the National Assembly in Yeouido, western Seoul. [LEE JAE-LIM]

To strengthen the nation’s competitiveness, Korea should prioritize investments in AI infrastructure over the development of cutting-edge large language models (LLMs) to outperform Chinese startup DeepSeek, according to industry experts.

Korean experts called for government-led support for data centers and related technologies, such as energy storage systems, uninterruptable power supply and liquid immersion cooling — areas with strong business potential — at a seminar convened at the National Assembly in western Seoul on Wednesday to discuss the country's AI advancement strategy.

They likened LLMs to just an “engine” within the broader AI industry, pointing out that the domestic industry is overfocusing on LLM development that is not yet profitable.

“Instead of blindly chasing OpenAI and DeepSeek to enter the cutthroat LLM development race, Korean firms should focus on AI infrastructure, where immediate revenue opportunities exist,” said Hee Yoon, Microsoft Korea’s director of government affairs.

“To give an example, Microsoft is investing 117 trillion won [$81.2 billion] in AI infrastructure in the first half of 2025 alone. […] But our investment in LLMs is actually very small; a majority of it is being poured into data centers and technologies to enhance power efficiency. Securing GPUs is important, but still, the budget for GPUs is miniscule compared to the overall investment required to operate AI infrastructure.”

It was confirmed on Wednesday that one of the world’s largest AI data centers will be built in South Jeolla, with an injection of as much as $35 billion from a global venture firm Stock Farm Road. The center's capacity is set to be three times of that of the ongoing Stargate project in the United States.

“However, it isn’t just to serve Korea — it’s meant to cover the entire Asia-Pacific region,” said Kim Dong-hwan, CEO of Korean AI startup 42Maru. “The key takeaway is that private sector firms are already making bold investments in AI infrastructure. The Korean government must match this momentum with more aggressive investments in AI infrastructure to stay competitive in the global AI race.”

The importance of Korean firms adopting a broader perspective and integrating AI across industries to drive profitable growth was emphasized.

U.S. software company Palantir Technologies was given as an example. Founded in 2003, the company specializes in building AI applications for enterprises and government agencies, and has seen its stock skyrocket 1,212 percent as of Tuesday from its initial listing in 2020.

The use of synthetic data, or artificially generated data, should be expanded to train AI models, as Korea lags behind global Big Tech firms in leveraging big data, according to Kim.

“Major tech companies abroad manage over 8 trillion data points each, whereas all of Korea combined still falls far behind,” he said. “In this situation, Korea still struggles to utilize synthetic data properly. We need to rethink how we structure and utilize our data to stay competitive, when the AI paradigm is shifting from simple information retrieval to reasoning-based AI, as seen in DeepSeek’s latest advancements.”

The government on Monday announced that it aims to secure 18,000 GPUs by the first half of 2026 to power a state-run AI computing center, but Kim urged more aggressive investments, pointing to Meta, which possesses 350,000 GPUs, and xAI, a U.S. AI startup run by Elon Musk, which said it trained its latest AI model, Grok3, with 200,000 GPUs.

BY LEE JAE-LIM [[email protected]]