[THE VIEW] Korea Joins the AI Supremacy War, But...

Naver, LG, and Lotte Ignite the Model Competition
Government Presence Fades in GPU Support and Education Strategies
Companies Are Running, but the Government Is Only Walking

As NVIDIA's stock price continues to soar and countries around the world fiercely compete for supremacy in artificial intelligence (AI), major domestic IT companies in Korea have finally thrown their hats into the ring after a long period of anticipation.

Naver has unveiled its general-purpose foundation model, HyperCLOVA, which significantly enhances Korean language comprehension, positioning "the AI that understands Korean best" as its differentiating strategy. LG AI Research has drawn attention by fully disclosing the model weights of its latest model, Exaone 4.0. Lotte Innovate has adopted an open-source strategy from the outset, launching iMember 3.0 and emphasizing its strength in providing full-stack AI services that are not limited to specific domains.

Although there are criticisms that these efforts amount to little more than tweaking existing foundation models, the very fact that each company is attempting to implement actual models is in itself a step forward, and these initiatives by large domestic corporations certainly hold positive significance.

Overseas companies are also making various attempts. The latest version of Google's Gemini is designed to analyze extremely long inputs?up to several million tokens at once?by implementing internal reasoning processes. Alibaba's Qwen 3 selectively engages in reasoning based on task difficulty, deeply reasoning through complex problems while minimizing thought or calculation for simpler ones, thereby increasing problem-solving speed.

While each model has slightly different distinguishing features, they all share the same goal: to overcome the limitations of the transformer architecture and pursue high-performance, general-purpose AI. To use a construction analogy, it is akin to repeatedly revising blueprints and sourcing different materials for each space to build the best possible house.

By constantly experimenting with how much weight to assign to each parameter and how to set values, these companies are inching closer to the dream of "AI that surpasses human intelligence." Just as revising blueprints and varying materials room by room leads to a better house, the process of actually implementing these models accumulates valuable know-how.

The issue is whether the Korean government is playing its part in the global race for supremacy. Europe has lost its competitiveness due to excessive regulation and a strategy of government-driven AI resource allocation. As of now, France is the only European country in the top 20 of the AI Comprehensive Intelligence Index.

A data center built using NVIDIA's GPUs. Photo by NVIDIA

In the United States, a performance-based reward culture that enables high-value incentives, combined with pro-business policies such as tax benefits and regulatory easing, allows for corporate-led growth. For example, in June 2025, OpenAI CEO Sam Altman revealed that "Meta offered $100 million per core researcher." This demonstrates a commitment to providing truly substantial rewards to attract talent.

China is betting on its massive talent pool. Starting in 2025, the city of Beijing has made AI education mandatory in elementary and middle schools, integrating AI classes into the curriculum. As of 2024, 535 universities in China have established AI-related departments. This tightly integrated system?linking government, industry, and education?reflects the government's direction to prioritize securing the scale and number of AI professionals.

Korea has not managed to secure the strengths of either the United States or China. First, with strong corporate regulations in place and a cultural tendency for significant rewards based on performance to be met with criticism rather than recognition, Korea lags behind the U.S. in terms of performance-based compensation systems. Organizational cultures that emphasize internal freedom and creativity have also yet to take root.

This has led to a side effect where a large number of STEM talents seek employment in the United States, drawn by its performance-oriented and assured reward systems. In other words, Korea finds it difficult to implement an American-style reward system due to strong corporate regulations and social resistance to performance-based compensation, and, unlike China, it does not have a vast reservoir of STEM talent.

So how should the Korean government set its policy direction? Current policies?such as government-led AI resource allocation and large enterprise-focused computing support?need to be broken down into more detailed, decentralized components, with increased flexibility in support. Individual developers and startups should be able to experiment freely, and resources such as GPUs should be provided more flexibly, moving away from rigid processes like business bidding and evaluation.

Rather than simply establishing new AI departments, it may be faster to provide experimental environments for students at the middle and high school levels to nurture talent. By granting appropriate credits and supporting GPU computing power for young people interested in AI to try various things, the country will naturally foster individuals who are deeply committed to AI.

While large corporations should be supported as needed, it is essential to continuously cultivate countless talents who can "stand on the shoulders of giants." Just as Korea once seized semiconductor supremacy, there is hope that, through more flexible and detailed policies and by fostering an environment of free competition among companies, the nation can once again claim leadership in AI.

Kyung Nakyung, Professor of Computer Science, National University of Singapore